View in EDS HTML Full Text PDF Full Text

Detecting Autism Spectrum Disorder and Attention Deficit Hyperactivity Disorder Using Multimodal Time-Frequency Analysis with Machine Learning Using the Electroretinogram from Two Flash Strengths

Saved in:
Bibliographic Details
Title: Detecting Autism Spectrum Disorder and Attention Deficit Hyperactivity Disorder Using Multimodal Time-Frequency Analysis with Machine Learning Using the Electroretinogram from Two Flash Strengths
Language: English
Authors: Sultan Mohammad Manjur, Luis Roberto Mercado Diaz, Irene O. Lee, David H. Skuse, Dorothy A. Thompson, Fernando Marmolejos-Ramos, Paul A. Constable, Hugo F. Posada-Quintero (ORCID 0000-0003-4514-4772)
Source: Journal of Autism and Developmental Disorders. 2025 55(4):1365-1378.
Availability: Springer. Available from: Springer Nature. One New York Plaza, Suite 4600, New York, NY 10004. Tel: 800-777-4643; Tel: 212-460-1500; Fax: 212-460-1700; e-mail: customerservice@springernature.com; Web site: https://link.springer.com/
Peer Reviewed: Y
Page Count: 14
Publication Date: 2025
Document Type: Journal Articles
Reports - Research
Descriptors: Autism Spectrum Disorders, Attention Deficit Hyperactivity Disorder, Symptoms (Individual Disorders), Clinical Diagnosis, Measurement Equipment, Physiology, Screening Tests
DOI: 10.1007/s10803-024-06290-w
ISSN: 0162-3257
1573-3432
Abstract: Purpose: Autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD) are conditions that similarly alter cognitive functioning ability and challenge the social interaction, attention, and communication skills of affected individuals. Yet these are distinct neurological conditions that can exhibit diverse characteristics which require different management strategies. It is desirable to develop tools to assist with early distinction so that appropriate early interventions and support may be tailored to an individual's specific requirements. The current diagnostic procedures for ASD and ADHD require a multidisciplinary approach and can be lengthy. This study investigated the potential of electroretinogram (ERG), an eye test measuring retinal responses to light, for rapid screening of ASD and ADHD. Methods: Previous studies identified differences in ERG amplitude between ASD and ADHD, but this study explored time-frequency analysis (TFS) to capture dynamic changes in the signal. ERG data from 286 subjects (146 control, 94 ASD, 46 ADHD) was analyzed using two TFS techniques. Results: Key features were selected, and machine learning models were trained to classify individuals based on their ERG response. The best model achieved 70% overall accuracy in distinguishing control, ASD, and ADHD groups. Conclusion: The ERG to the stronger flash strength provided better separation and the high frequency dynamics (80-300 Hz) were more informative features than lower frequency components. To further improve classification a greater number of different flash strengths may be required along with a discrimination comparison to participants who meet both ASD and ADHD classifications and carry both diagnoses.
Abstractor: As Provided
Entry Date: 2025
Accession Number: EJ1464139
Database: ERIC
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
    Url: https://content.ebscohost.com/cds/retrieve?content=AQICAHj0k_4E0hTGH8RJwT4gCJyBsGNe_WN95AvKlDbXJGqwxwFK1U1ChUqBKOUPhx9jTWDpAAAA4zCB4AYJKoZIhvcNAQcGoIHSMIHPAgEAMIHJBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDE4fhvzCAF-d9cM3OQIBEICBm_6xRcl-M90-W64p8Cjd__-ByQ6TolwckQo1YVJu4jsJFZ4nWKZRjnWiKl77r6fGXAf5aQlk85fO0yz2SY0goaoUdCGW4EGZ4e8qL1WiG0zsKpySiL8cVUq1nCSMJj9qRPjBAk3Ti--HzGt21qxImC7yavjkbMdbP6fcdE8jYSZmjWTJh88PQuqLWpeP_MfogQLjdCNL6OgqZK1K
Text:
  Availability: 1
  Value: <anid>AN0183973035;aut01apr.25;2025Mar26.05:27;v2.2.500</anid> <title id="AN0183973035-1">Detecting Autism Spectrum Disorder and Attention Deficit Hyperactivity Disorder Using Multimodal Time-Frequency Analysis with Machine Learning Using the Electroretinogram from Two Flash Strengths </title> <p>Purpose: Autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD) are conditions that similarly alter cognitive functioning ability and challenge the social interaction, attention, and communication skills of affected individuals. Yet these are distinct neurological conditions that can exhibit diverse characteristics which require different management strategies. It is desirable to develop tools to assist with early distinction so that appropriate early interventions and support may be tailored to an individual's specific requirements. The current diagnostic procedures for ASD and ADHD require a multidisciplinary approach and can be lengthy. This study investigated the potential of electroretinogram (ERG), an eye test measuring retinal responses to light, for rapid screening of ASD and ADHD. Methods: Previous studies identified differences in ERG amplitude between ASD and ADHD, but this study explored time-frequency analysis (TFS) to capture dynamic changes in the signal. ERG data from 286 subjects (146 control, 94 ASD, 46 ADHD) was analyzed using two TFS techniques. Results: Key features were selected, and machine learning models were trained to classify individuals based on their ERG response. The best model achieved 70% overall accuracy in distinguishing control, ASD, and ADHD groups. Conclusion: The ERG to the stronger flash strength provided better separation and the high frequency dynamics (80–300 Hz) were more informative features than lower frequency components. To further improve classification a greater number of different flash strengths may be required along with a discrimination comparison to participants who meet both ASD and ADHD classifications and carry both diagnoses.</p> <p>Keywords: Electroretinogram; Autism spectrum disorder; Attention deficit hyperactivity disorder; Time-frequency analysis; Machine learning; Psychology and Cognitive Sciences Psychology</p> <p>Supplementary Information The online version contains supplementary material available at https://doi.org/10.1007/s10803-024-06290-w.</p> <hd id="AN0183973035-2">Introduction</hd> <p>Neurodevelopmental conditions, specifically autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD), have an increasing reported prevalence with estimates currently at 1% (Zeidan et al., [<reflink idref="bib97" id="ref1">97</reflink>]) and 5% (Song et al., [<reflink idref="bib85" id="ref2">85</reflink>]) respectively and sometimes co-occur. ASD is a lifelong condition characterized by limitations in social and reciprocal communication skills alongside repetitive and restrictive actions or interests (Hodges et al., [<reflink idref="bib45" id="ref3">45</reflink>]). On the other hand, ADHD is characterized by heightened impulsivity, hyperactivity, and inattentiveness (Wilens & Spencer, [<reflink idref="bib93" id="ref4">93</reflink>]). The unique challenges posed by these disorders often necessitate specialized assistance (Young et al., [<reflink idref="bib95" id="ref5">95</reflink>]) and detecting these conditions at an early stage allows appropriate support (Hadders-Algra, [<reflink idref="bib41" id="ref6">41</reflink>]). Behavioral therapies, educational strategies, and support services tailored to the specific needs of individuals with ASD and ADHD contribute to improved adaptive skills, better social integration, and improved quality of life (Elder et al., [<reflink idref="bib32" id="ref7">32</reflink>]). The current diagnosis of ASD and ADHD involves a comprehensive and multidimensional assessment with input from clinicians, psychologists, parents, and educators (Hatch et al., [<reflink idref="bib43" id="ref8">43</reflink>]).</p> <p>The potentially lengthy diagnostic procedure and the substantial heterogeneity evident in ASD and ADHD coupled with overlap and variations in symptom severity complicates and can delay diagnosis (Duvall et al., [<reflink idref="bib30" id="ref9">30</reflink>]; Hatch et al., [<reflink idref="bib43" id="ref10">43</reflink>]; Malwane et al., [<reflink idref="bib59" id="ref11">59</reflink>]). The absence of specific biological or genetic tests for the conclusive diagnosis of ASD and ADHD underscores the need for innovative diagnostic approaches that may have screening benefits or otherwise support the clinical diagnosis. In this study, we have investigated the feasibility of using the electroretinogram (ERG) (Quintana et al., [<reflink idref="bib75" id="ref12">75</reflink>]; Youssef et al., [<reflink idref="bib96" id="ref13">96</reflink>]), derived from the electrical activity of the neuroretina, as a potential biomarker to facilitate the diagnosis and management of ASD and ADHD. Whilst previous studies have identified group differences in ERGs, to date the application of ML models to classification has not been explored fully with respect to ASD and ADHD using a minimal flash strength number (two in this case) to evaluate the potential of a rapid and noninvasive test for ASD and/or ADHD.</p> <p>Many studies have explored potential biomarkers for the diagnosis of ASD and ADHD with promising results, in particular with biosignals such as electroencephalography (EEG) (Baygin et al., [<reflink idref="bib9" id="ref14">9</reflink>]; Bosl et al., [<reflink idref="bib11" id="ref15">11</reflink>]; Djemal et al., [<reflink idref="bib26" id="ref16">26</reflink>]), heart rate variability (HRV) (Bellato et al., [<reflink idref="bib10" id="ref17">10</reflink>]), eye movements (Schmitt et al., [<reflink idref="bib79" id="ref18">79</reflink>]) and facial scanning paths (Aktas et al., [<reflink idref="bib2" id="ref19">2</reflink>]; Muszkat et al., [<reflink idref="bib69" id="ref20">69</reflink>]). Despite encouraging outcomes with these noninvasive signals, challenges arise in their recording, necessitating specialized skills and knowledge (Puce & Hämäläinen, [<reflink idref="bib74" id="ref21">74</reflink>]). Additionally, these signals are prone to being influenced by noise, external stimuli and conditions (Tiwari et al., [<reflink idref="bib89" id="ref22">89</reflink>]), adding a layer of complexity to their accurate capture and interpretation which hinders their application towards ASD and ADHD diagnosis. Hence, the quest for a biomarker that operates independently of emotion and stress, and is easily recordable, will assist the development of a practical test for the early diagnosis of neurodevelopmental conditions such as ASD and ADHD.</p> <p>The retina, housing a highly organized and layered array of neurons and photoreceptors, serves as a unique "window to the central nervous system," offering insights into the underlying intricate neural processes associated with neurological disorders (London et al., [<reflink idref="bib55" id="ref23">55</reflink>]). The electroretinogram (ERG) is a rapid and non-invasive diagnostic test, providing valuable insights into the electrical activity of retinal cells and their synaptic responses when subjected to light stimulation (Asi & Perlman, [<reflink idref="bib8" id="ref24">8</reflink>]; Kolb et al., [<reflink idref="bib48" id="ref25">48</reflink>]). By recording the superimposed electrical activity, the ERG captures the main responses of the photoreceptor, bipolar, and the ganglion cells in the retina (Asi & Perlman, [<reflink idref="bib8" id="ref26">8</reflink>]; Constable et al., 2022; Perlman, [<reflink idref="bib72" id="ref27">72</reflink>]; Stockton & Slaughter, [<reflink idref="bib87" id="ref28">87</reflink>]). Glutamate, as the primary excitatory neurotransmitter, GABA, the principal inhibitory neurotransmitter, and dopamine, with its multifaceted functions, all contribute to the dynamics of the recorded ERG waveform but also hold relevance in understanding the pathophysiology of ASD and ADHD (I. O. Lee et al., [<reflink idref="bib54" id="ref29">54</reflink>]).</p> <p>Previous studies have highlighted the intricate relationship between retinal neurotransmitter activity, as reflected in the ERG, and the broader implications for neurological and psychiatric conditions such as ASD, ADHD, Bipolar disorder, Parkinson's disease, and Schizophrenia (Constable et al., [<reflink idref="bib18" id="ref30">18</reflink>], [<reflink idref="bib21" id="ref31">21</reflink>], [<reflink idref="bib19" id="ref32">19</reflink>]; Gross et al., [<reflink idref="bib40" id="ref33">40</reflink>]; Hébert et al., [<reflink idref="bib44" id="ref34">44</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref35">54</reflink>]; Mello et al., [<reflink idref="bib66" id="ref36">66</reflink>]). Conventional analyses of the ERG waveform traditionally concentrate on time-domain parameters, specifically the amplitudes and time to peaks of the principal waveform components, the a- and b-waves (Constable et al., [<reflink idref="bib19" id="ref37">19</reflink>]; Demmin et al., [<reflink idref="bib24" id="ref38">24</reflink>]; Friedel et al., [<reflink idref="bib33" id="ref39">33</reflink>]; Hamilton et al., [<reflink idref="bib42" id="ref40">42</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref41">54</reflink>]). Recently, Lee et al. conducted an analysis of various time-domain indices of the ERG that revealed lower b-wave amplitude in individuals with ASD compared to neurotypicals (I. O. Lee et al., [<reflink idref="bib54" id="ref42">54</reflink>]). Conversely, the authors reported an increased b-wave amplitude in individuals with ADHD (I. O. Lee et al., [<reflink idref="bib54" id="ref43">54</reflink>]). Leveraging these distinct ERG patterns, the authors successfully differentiated between subjects with ASD and ADHD, achieving an Area Under the Curve (AUC) score of 0.88 with 84% sensitivity and 57% specificity with a flash strength of 1.2 log photopic cd.s.m<sups>− 2</sups> (I. O. Lee et al., [<reflink idref="bib54" id="ref44">54</reflink>]). Just focused on time-domain indices, this study offered compelling evidence that expanding the analytical scope to include spectral-domain investigations may unveil nuanced insights, allowing for greater understanding of the differences between ASD and ADHD.</p> <p>The distinct components of the ERG waveform represent different stages of neural processing in the retina (Kolb et al., [<reflink idref="bib48" id="ref45">48</reflink>]; Perlman, [<reflink idref="bib72" id="ref46">72</reflink>]), and we can better understand the frequency-specific characteristics of retinal activity by subjecting these components to spectral analysis (Constable et al., [<reflink idref="bib20" id="ref47">20</reflink>]; Zhdanov et al., 2023). Aberrations in retinal responses, that manifest as deviations from typical spectral profiles, may serve as potential biomarkers for neurodevelopmental disorders (Manjur et al., [<reflink idref="bib61" id="ref48">61</reflink>]). The application of Time-Frequency (TF) analysis to ERG, specifically employing Discrete Wavelet Transform (DWT) (Shensa, [<reflink idref="bib82" id="ref49">82</reflink>]), has proved effective in characterizing distinct stages of the retinal response (Gauvin et al., [<reflink idref="bib34" id="ref50">34</reflink>], [<reflink idref="bib35" id="ref51">35</reflink>], [<reflink idref="bib36" id="ref52">36</reflink>], [<reflink idref="bib37" id="ref53">37</reflink>]). This analytical approach allows a detailed examination of the ERG signal's temporal and frequency components, and valuable insights into the dynamic changes associated with various phases of the retinal response. Gauvin et al. proposed 'Haar' wavelet analysis of the ERG waveform to identify pivotal central frequencies linked to the ON- and OFF-retinal pathways in the 'a' and 'b' waves, specifically at 20 and 40 Hz (Gauvin et al., [<reflink idref="bib34" id="ref54">34</reflink>], [<reflink idref="bib35" id="ref55">35</reflink>], [<reflink idref="bib36" id="ref56">36</reflink>], [<reflink idref="bib37" id="ref57">37</reflink>]). Additionally, Gauvin et al. pinpointed central frequencies associated with early and late oscillatory potentials (OPs) at 80 and 160 Hz, respectively (Gauvin et al., [<reflink idref="bib36" id="ref58">36</reflink>]). Recently, we have proposed another TF analysis technique named variable frequency complex demodulation (VFCDM) (Wang et al., [<reflink idref="bib92" id="ref59">92</reflink>]) to analyze ERG signal with higher spectral resolution (Manjur et al., [<reflink idref="bib60" id="ref60">60</reflink>]; Posada-Quintero et al., [<reflink idref="bib73" id="ref61">73</reflink>]). VFDCM analysis showed that the spectral characteristics in the higher frequency region (80–300 Hz) were most affected in neurodevelopmental conditions (Manjur et al., [<reflink idref="bib60" id="ref62">60</reflink>]; Posada-Quintero et al., [<reflink idref="bib73" id="ref63">73</reflink>]). Our previous analyses also showed the potential of VFCDM analysis of the ERG to differentiate between these conditions in a binary classification framework of ASD vs control (Manjur et al., [<reflink idref="bib60" id="ref64">60</reflink>]; Posada-Quintero et al., [<reflink idref="bib73" id="ref65">73</reflink>]).</p> <p>Expanding upon the findings from prior studies, we undertook an exploration using multimodal (DWT and VFCDM) spectral analysis of the ERG waveform to extract features that demonstrate a statistically significant effect for the differentiation of the groups examined. Using these features, we then developed and trained a three-class machine learning (ML) model using the extracted features with the objective of proficiently differentiating between individuals with ASD, ADHD, and the control group. We experimented with well-known classical ML classifiers and evaluated the models using a Leave k-groups out validation strategy to assess the overall model performance.</p> <hd id="AN0183973035-3">Related Works</hd> <p>The typical clinical assessment procedures for ASD and ADHD involve a multifaceted screening process encompassing a comprehensive array of diagnostic tools. The evaluation can involve clinical interviews, behavioral observations, and the administration of standardized rating scales. For instance, the Autism Diagnostic Observation Schedule (ADOS) (Lord et al., [<reflink idref="bib58" id="ref66">58</reflink>]) and the Autism Diagnostic Interview (ADI) (Lord et al., [<reflink idref="bib57" id="ref67">57</reflink>]) are the acknowledged gold standards for the assessment of ASD. Complementing these benchmarks, additional observational ASD assessment tools include the Child Autism Rating Scale (CARS) (Schopler et al., [<reflink idref="bib80" id="ref68">80</reflink>]), the Social Communication Questionnaire (SCQ) (Rutter, [<reflink idref="bib77" id="ref69">77</reflink>]), the Social Responsiveness Scale (SRS) (Constantino, [<reflink idref="bib22" id="ref70">22</reflink>]), and the Developmental, Dimensional, and Diagnostic Interview (3Di) (Skuse et al., [<reflink idref="bib84" id="ref71">84</reflink>]). Similarly, clinical assessments for ADHD utilize various tools such as the Conners Rating Scales (Conners et al., [<reflink idref="bib17" id="ref72">17</reflink>]), ADHD Rating Scale-IV (DuPaul et al., [<reflink idref="bib29" id="ref73">29</reflink>]), and Vanderbilt ADHD Diagnostic Rating Scales (Anderson et al., [<reflink idref="bib5" id="ref74">5</reflink>]). Behavioral observations, such as the Direct Observation Form (DOF) (McConaughy & Achenbach, [<reflink idref="bib65" id="ref75">65</reflink>]) for ADHD and ADHD-specific Brown Attention-Deficit Disorder (Brown, [<reflink idref="bib13" id="ref76">13</reflink>]) scales contribute to the comprehensive evaluation process to form a clinical diagnosis.</p> <p>These diagnostic procedures are time consuming and require the examiner's interpretation and expertise. Therefore, there is a need to develop a rapid and objective early detection process to complement or support these diagnostic instruments and procedures.</p> <p>Over the past decades, various modalities have been investigated with the aim of identifying a reliable and objective indicator for ASD and ADHD that complement the established clinical diagnostic processes. For instance, Kuttala et al. proposed a dense attentive GAN-based model to detect ASD and ADHD by analyzing T1-weighted longitudinal structural magnetic resonance images (MRI) (Kuttala et al., [<reflink idref="bib51" id="ref77">51</reflink>]). Their developed model was trained in a binary framework to separate ASD from control and ADHD from control and achieved 0.91 and 0.85 AUC score respectively (Kuttala et al., [<reflink idref="bib51" id="ref78">51</reflink>]). Ghiiassian et al. proposed a model using Histogram of Gradient (HOG) based features extracted from structural and functional MRI to detect ASD and ADHD in a holdout fashion achieving 65% and 69.6% accuracy respectively (Ghiassian et al., [<reflink idref="bib38" id="ref79">38</reflink>]). In recent investigations, electroencephalography (EEG) has been a focal point to delve into the different patterns of electrical activity within the brain, to discern distinctive neurophysiological markers associated with ASD and ADHD (Adamou et al., [<reflink idref="bib1" id="ref80">1</reflink>]; Ari et al., [<reflink idref="bib6" id="ref81">6</reflink>]; Arns et al., [<reflink idref="bib7" id="ref82">7</reflink>]; Bosl et al., [<reflink idref="bib11" id="ref83">11</reflink>], [<reflink idref="bib12" id="ref84">12</reflink>]; Clarke et al., [<reflink idref="bib16" id="ref85">16</reflink>]; Djemal et al., [<reflink idref="bib26" id="ref86">26</reflink>]). Matlis et al. reported reduced posterior/anterior power ratio in the alpha (8–14 Hz) frequency band along with reduced global density and reduced mean connectivity strength in ASD subjects (Matlis et al., [<reflink idref="bib64" id="ref87">64</reflink>]). Arns et al. showed that the theta/beta ratio (TBR) deviated significantly in a subgroup of ADHD subjects compared to neurotypicals and suggested TBR could offer prognostic information (Arns et al., [<reflink idref="bib7" id="ref88">7</reflink>]). Individuals with neurodevelopmental disorders often display atypical visual scan paths and previous researchers have undertaken analyses of eye movement to detect ASD and ADHD (Elbattah et al., [<reflink idref="bib31" id="ref89">31</reflink>]; Kleberg et al., [<reflink idref="bib47" id="ref90">47</reflink>]; Lee et al., [<reflink idref="bib53" id="ref91">53</reflink>]; Schmitt et al., [<reflink idref="bib79" id="ref92">79</reflink>]; Tsang & Chu, [<reflink idref="bib90" id="ref93">90</reflink>]). Kang et al. performed a multimodal analysis combining features from EEG and eye movement data to predict ASD with approximately 0.93 AUC score using SVM classifier (Kang et al., [<reflink idref="bib46" id="ref94">46</reflink>]). Merzon et al. proposed a method to detect ADHD subjects with 0.92 AUC score using features from eye tracking data collected whilst the subjects played a lifelike prospective memory game (Merzon et al., [<reflink idref="bib68" id="ref95">68</reflink>]).</p> <p>There is now a burgeoning interest in investigating the ERG (retinal response to light flashes) as a means of studying psychiatric disorders (Al Abdlseaed et al., [<reflink idref="bib3" id="ref96">3</reflink>]; Chiquita et al., [<reflink idref="bib15" id="ref97">15</reflink>]; Gauvin et al., [<reflink idref="bib35" id="ref98">35</reflink>]; Gross et al., [<reflink idref="bib40" id="ref99">40</reflink>]; Hébert et al., [<reflink idref="bib44" id="ref100">44</reflink>]; Shoeibi et al., [<reflink idref="bib83" id="ref101">83</reflink>]; Stockton & Slaughter, [<reflink idref="bib87" id="ref102">87</reflink>]). While numerous studies have highlighted encouraging findings in using the ERG to explore various disorders, showcasing its potential as a biomarker, they confined their analysis mostly within time domain (Constable et al., [<reflink idref="bib18" id="ref103">18</reflink>], [<reflink idref="bib21" id="ref104">21</reflink>]; Hébert et al., [<reflink idref="bib44" id="ref105">44</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref106">54</reflink>]). In this study we have employed high resolution time-frequency analysis techniques along with ML models to develop a complete model for prediction of ASD and ADHD using a minimal number of flash strengths under light-adapted conditions.</p> <hd id="AN0183973035-4">Methods</hd> <p>Ethical approval for the studies that provided the data analyzed in this paper was granted by the Flinders University Human Research Ethics Committee and the Southeast Scotland Research Ethics Committee in the United Kingdom. For this study we utilized a subset of pre-existing ERG recordings obtained from neurotypicals and individuals with ASD and ADHD (Constable et al., [<reflink idref="bib21" id="ref107">21</reflink>], [<reflink idref="bib20" id="ref108">20</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref109">54</reflink>]). The recordings were sourced from both eyes and encompassed two flash strengths.</p> <hd id="AN0183973035-5">Participants</hd> <p>A cumulative cohort of 286 individuals, comprising 146 controls, 94 ASD and 46 ADHD participated in earlier studies conducted at two different sites in London (UK) and Adelaide (Australia). The sex distribution among participants indicated that within the control group, 53 individuals (37%) were males, while the ADHD group had 24 males (53%), and the ASD group 56 males (59%). The mean ages (± standard deviation) of participants in the control, ASD, and ADHD groups were 14.5 years (± 5.4), 11 years (± 4.5), and 13.2 years (± 3.4), respectively. Participants under the age of 16 were included upon obtaining written informed consent from their legal guardian. The age disparity among groups is deemed inconsequential in influencing the ERG amplitudes within this young cohort with clear optical media (Sannita et al., [<reflink idref="bib78" id="ref110">78</reflink>]). Diagnostic assessments were performed by pediatric psychiatrist or clinical psychologist at the social communication disorder clinics at Great Ormond Street Hospital for Children in the UK or local Child and Adolescent Mental Health clinics in South Australia. All participants diagnosed with ASD met DSM-IV (American Psychiatric Association & Association, [<reflink idref="bib4" id="ref111">4</reflink>]) or DSM-V (<emph>Diagnostic and Statistical Manual of Mental Disorders</emph>, [<reflink idref="bib25" id="ref112">25</reflink>]) criteria, as determined through assessments employing the ADOS (Lord et al., [<reflink idref="bib56" id="ref113">56</reflink>]), ADOS-2 (Gotham et al., [<reflink idref="bib39" id="ref114">39</reflink>]), or 3Di (Skuse et al., [<reflink idref="bib84" id="ref115">84</reflink>]) and ADHD diagnosis were made based on the Conners Rating Scales or ADHD Rating Scale-IV with clinical reports. Control participants had no familial background of ASD or psychiatric conditions. Exclusions from ASD, ADHD and control groups comprised individuals with a history of strabismus surgery, inherited retinal disease, chromosomal disorders (e.g., Down syndrome), comorbid ASD + ADHD, or a history of traumatic brain injury.</p> <hd id="AN0183973035-6">Electroretinogram Recording Protocol</hd> <p>The study employed a customized Light-Adapted (LA) full-field ERG series, conducted in accordance with the guidelines outlined by the International Society for Clinical Electrophysiology of Vision (ISCEV) (Robson et al., [<reflink idref="bib76" id="ref116">76</reflink>]). The RETeval device (LKC Technologies Inc Gaithersburg, MD, USA) was utilized for data acquisition. The selection of LA-ERG for capturing retinal responses was deliberate, driven by its expediency compared to the dark-adapted ERG counterpart, which requires 20 min adaptation to a dark environment (Al Abdlseaed et al., [<reflink idref="bib3" id="ref117">3</reflink>]). Notably, all recordings were collected under normal room lighting conditions, maintaining an illuminance of 250–350 lx. Pupils remained undilated throughout the procedure. Building upon findings from our previous investigations LAERGs to two flash strengths were strategically chosen for examination: 113 Troland-Seconds (Td.s) and 446 Td.s, presented on a 1130 Td white background (I. O. Lee et al., [<reflink idref="bib54" id="ref118">54</reflink>]; Manjur et al., [<reflink idref="bib61" id="ref119">61</reflink>], [<reflink idref="bib60" id="ref120">60</reflink>]; Posada-Quintero et al., [<reflink idref="bib73" id="ref121">73</reflink>]). Assuming a 6 mm pupil diameter, the equivalent flash strengths translated to 0.60 and 1.20 log photopic cd.s.m<sups>-2</sups>, respectively, on a 40 cd.m<sups>-2</sups> white background (Constable et al., [<reflink idref="bib21" id="ref122">21</reflink>]). The ERG recordings were sequentially obtained from the right eye followed by the left eye and flashes were administered at a stimulus frequency of 2 flashes/ second (2 Hz).</p> <p>ERGs were averaged (n 30–60) to generate the final reported ERG waveform and the sampling frequency for the collected ERG signals was set at 2000 Hz, ensuring high temporal resolution. Individual traces were excluded from the averaged reported waveform they deviated beyond the upper or lower 25th percentile of the averaged waveform to ensure the inclusion of representative and consistent ERGs free from artefacts such as blinks during the recording run. Replicates of the recordings were conducted for each eye, with Table 1 listing the total number of ERG signals recorded from the participants for the different combinations of flash strength and eye. Participants were seated throughout the data collection process and looked towards a central red LED within the Ganzfeld. Fixation was monitored with an in-built IR camera. Electrode positions were checked from the reported image captured by the RETeval using a scaled graticule, and recordings excluded if the electrode was placed 4 mm or more below the lower eyelid. A comprehensive discussion on the ERG recording protocol, providing detailed procedural insights, has been delineated in the original studies from which part of the dataset for this study was derived.</p> <p>Table 1 Number of ERG (averaged) samples included for each flash strength</p> <p> <ephtml> <table frame="hsides" rules="groups"><thead><tr><th align="left"><p>Flash Strength</p><p>(Td.s)</p></th><th align="left"><p>Eye</p></th><th align="left"><p>Control</p></th><th align="left"><p>ASD</p></th><th align="left"><p>ADHD</p></th></tr></thead><tbody><tr><td align="left" rowspan="2"><p>446</p></td><td align="left"><p>Right</p></td><td char="." align="char"><p>310</p></td><td char="." align="char"><p>183</p></td><td char="." align="char"><p>112</p></td></tr><tr><td align="left"><p>Left</p></td><td char="." align="char"><p>325</p></td><td char="." align="char"><p>172</p></td><td char="." align="char"><p>107</p></td></tr><tr><td align="left" rowspan="2"><p>113</p></td><td align="left"><p>Right</p></td><td char="." align="char"><p>313</p></td><td char="." align="char"><p>182</p></td><td char="." align="char"><p>112</p></td></tr><tr><td align="left"><p>Left</p></td><td char="." align="char"><p>318</p></td><td char="." align="char"><p>159</p></td><td char="." align="char"><p>114</p></td></tr></tbody></table> </ephtml> </p> <hd id="AN0183973035-7">Data Processing</hd> <p>We examined the ERG signal encompassing both time and time-frequency analysis to extract features that could potentially discriminate between the control, ASD, and ADHD groups. Employing two distinct time-frequency analysis techniques, we aimed to provide a detailed characterization of the divergent electroretinographic profiles observed within each cohort based on two flash strengths.</p> <hd id="AN0183973035-8">Time-Domain Analysis</hd> <p>The ERG serves as a recorded summed electrical response of the retina to a brief flash of light, offering an objective and quantitative assessment of retinal function (Perlman, [<reflink idref="bib72" id="ref123">72</reflink>]). The entry of light into the eye induces a graded hyperpolarization and depolarization of retinal cell membranes at distinct temporal intervals (Perlman, [<reflink idref="bib72" id="ref124">72</reflink>]). The collective ERG signal waveform reflects these dynamic changes and is typically recorded from electrodes positioned near the corneal anterior pole of the eye and manifests as voltage changes over time and is illustrated in Figure SI-1 (Top).</p> <p>The ERG waveform has several distinctive features; an initial negative trough termed the a-wave, predominantly reflecting the hyperpolarization of cone outer segments followed by the positive.</p> <p>b-wave that is dominated by the depolarization of second-order bipolar cells in the retinal ON- and OFF- pathways with additional influence from potassium currents in glial cells (Constable et al., [<reflink idref="bib18" id="ref125">18</reflink>], [<reflink idref="bib21" id="ref126">21</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref127">54</reflink>]; Thompson et al., [<reflink idref="bib88" id="ref128">88</reflink>]) shaping the b-wave. The ascending limb of b-wave has some high frequency low amplitude oscillations reflecting responses initiated by amacrine cells which are termed the oscillatory potentials (OPs) (Constable et al., [<reflink idref="bib18" id="ref129">18</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref130">54</reflink>]). Ganglion cells contribute to the Photopic Negative Response (PhNR) that may be reported as either the PhNR<subs>min</subs> (Lowest trough after b-wave) or PhNR<subs>72</subs> (PhNR response at 72 ms after flash onset) (Constable et al., [<reflink idref="bib18" id="ref131">18</reflink>], [<reflink idref="bib21" id="ref132">21</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref133">54</reflink>]; Posada-Quintero et al., [<reflink idref="bib73" id="ref134">73</reflink>]). Previous studies that explored the amplitude and time-to-peak characteristics of the a- and b-waves, as well as the PhNR found the b-wave amplitude was the paramount discriminator of control, ASD, and ADHD groups (Asi & Perlman, [<reflink idref="bib8" id="ref135">8</reflink>]; Constable et al., [<reflink idref="bib18" id="ref136">18</reflink>], [<reflink idref="bib21" id="ref137">21</reflink>]; Gross et al., [<reflink idref="bib40" id="ref138">40</reflink>]; Hébert et al., [<reflink idref="bib44" id="ref139">44</reflink>]; I. O. Lee et al., [<reflink idref="bib54" id="ref140">54</reflink>]). As shown in Figure SI-1 (bottom) individuals with ASD exhibit a reduced b-wave amplitude, while those with ADHD manifest an elevated b-wave amplitude compared to control counterparts. In this study we will focus on the timing and amplitudes of a- and b-waves for time domain analysis.</p> <hd id="AN0183973035-9">Time-Frequency Analysis</hd> <p>Time domain characteristics exhibit promising outcomes but to further elucidate the intricate dynamics of the ERG, we used two distinct time-frequency analysis techniques- DWT (Shensa, [<reflink idref="bib82" id="ref141">82</reflink>]) and VFCDM (Wang et al., [<reflink idref="bib92" id="ref142">92</reflink>]).</p> <hd id="AN0183973035-10">Discrete Wavelet Transform (DWT)</hd> <p>The DWT can be mathematically expressed as a convolution operation between the signal of interest and the appropriately scaled and translated mother wavelet as shown in Eq. (<reflink idref="bib1" id="ref143">1</reflink>).</p> <p>1</p> <p>Graph</p> <p>Here represents the DWT coefficients, a and b are scaling and translational factors. The DWT, with its convolution and scaling-translational operations, facilitates the extraction of information at different scales, enabling a multiresolution analysis of the reported ERG signal waveform. DWT has been utilized previously to characterize morphological features of the ERG (Gauvin et al., [<reflink idref="bib34" id="ref144">34</reflink>], [<reflink idref="bib35" id="ref145">35</reflink>], [<reflink idref="bib36" id="ref146">36</reflink>], [<reflink idref="bib37" id="ref147">37</reflink>]) through various DWT descriptors within the Time-Frequency Spectrogram (TFS), as depicted in Figure SI-2.</p> <p>Within the ERG signal, the a-wave and the b-wave are centered in the low frequency range (0–40 Hz) and are characterized by descriptors 20a, 40a, 20b, and 40b (Gauvin et al., [<reflink idref="bib37" id="ref148">37</reflink>]). The OPs are situated in higher frequency regions at 80 and 160 Hz respectively and are identified by OP80 and OP160 descriptors (Gauvin et al., [<reflink idref="bib36" id="ref149">36</reflink>]). In this study, we utilized a 7 level DWT decomposition of the ERG signal using 'Symlet 2' wavelet to compute these coefficients and analyzed those coefficients located in the region of interest from 20, 40, 80 and 160 Hz frequency bands.</p> <hd id="AN0183973035-11">Variable Frequency Complex Demodulation (VFCDM)</hd> <p>The DWT offers valuable time-localized spectral information through its coefficients, yet it comes with inherent limitations, primarily stemming from its down sampling factor at each decomposition level, constraining the achievable resolution particularly in the higher frequency regions. Acknowledging this constraint, we have opted for VFCDM an alternative time-frequency analysis technique to explore the dynamics within the ERG signals, especially in higher frequency regions critical for capturing subtle variations in retinal responses relating to the OPs in ASD and ADHD that have been identified as atypical (Constable et al., 2016, 2022). VFCDM does not apply any down sampling which enables it to retain accurate amplitude distribution while providing greater frequency resolution. The mathematical formulation behind VFCDM has been described in our previous studies (Posada-Quintero et al., 2023; Wang et al., [<reflink idref="bib92" id="ref150">92</reflink>]). We have decomposed the ERG signal into (= 24) non-overlapping equal width frequency sub bands or components.</p> <p>2</p> <p>Graph</p> <p>In this equation is the ERG signal and represents the frequency components. Since our sampling frequency was 2000 Hz, each component contains approximately 41.67 Hz according to Nyquist's theorem. Figure SI-3 shows VFCDM decomposition of the ERG signal and their power spectral density. From the spectrogram in Figure SI-3 (top row) we can see that most of the information of the ERG signals within 300 Hz as expected based on the band pass filtering of the raw signal and that's why we included only the first 8 VFCDM components in our analysis. Figure SI-3 (bottom row) shows the decomposed VFCDM components and their respective spectral content. We computed several statistical features from these VFCDM components to examine subtle changes occurring especially in the high frequency range.</p> <hd id="AN0183973035-12">Feature Set and Analysis</hd> <p>In this study, based on two flash strengths, a comprehensive analysis was conducted, encompassing time-domain indices, DWT coefficients, and statistical features derived from VFCDM components. Table 2 summarizes the set of features we extracted and analyzed to select the optimal set of features for training ML model.</p> <p>Table 2 Features extracted from ERG time domain and TFS analysis</p> <p> <ephtml> <table frame="hsides" rules="groups"><thead><tr><th align="left"><p>Modality</p></th><th align="left"><p>Feature name</p></th></tr></thead><tbody><tr><td align="left"><p>Time Domain indices</p></td><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq8.gif" />, <inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq9.gif" />, <inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq10.gif" />, <inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq11.gif" /></p></td></tr><tr><td align="left"><p>DWT coefficients</p></td><td align="left"><p>DWT coefficients in a20, a40, b20, b40, OP80 and OP160 descriptors</p></td></tr><tr><td align="left"><p>Statistical Features from VFCDM components</p></td><td align="left"><p>mean, maximum, minimum, standard deviation, kurtosis, inter-quartile range, root mean square, median</p></td></tr></tbody></table> </ephtml> </p> <p>We conducted statistical comparisons across control, ADHD, and ASD groups for each of the features detailed in Table 2. Notably, both time-domain indices and TFS features did not conform to a normal distribution based on the Shapiro-Wilk normality test (Marmolejo-Ramos & González-Burgos, [<reflink idref="bib63" id="ref151">63</reflink>]; Shapiro & Wilk, [<reflink idref="bib81" id="ref152">81</reflink>]). To compare these groups, we employed the Kruskal-Walli's test (Kruskal & Wallis, [<reflink idref="bib50" id="ref153">50</reflink>]) a non-parametric version of one-way ANOVA, followed by post hoc analysis using the Dunn's test (Dunn, [<reflink idref="bib28" id="ref154">28</reflink>]) with Holm-Bonferroni adjustment for pairwise comparisons. Features exhibiting a statistically significant difference underwent further scrutiny through receiver operating characteristic analysis and Youden's J index (Youden, [<reflink idref="bib94" id="ref155">94</reflink>]) to perform binary classification between groups i.e., control vs. ASD, control vs. ADHD and ASD vs. ADHD. Only those features demonstrating robust binary classification performance were selected for integration into the final ML model, this ensured the inclusion of only highly discriminatory features.</p> <hd id="AN0183973035-13">Machine Learning Model</hd> <p>Following the selection of an optimal feature set from time-domain, DWT, and VFCDM modalities, we applied a diverse array of classical ML models i.e., Random Forest (RF), Adaptive Boosting (AdaBoost), Gradient Boosting (GradBoost), Extreme gradient Boosting (XGBoost), Support Vector Machine (SVM), K-nearest neighbor (KNN) and multi-layer perceptron (MLP) to discern and classify ASD and ADHD from neurotypicals. Given instances where multiple samples, representing repeated measurements, were obtained from the same subject for a specific flash strength and eye, a rigorous 10-fold subject-wise cross-validation strategy was implemented. This ensured that replicates from the same subject were distinctly allocated to either the training or testing sets, thereby eliminating any potential overlap with subject or repeated measures from a subject. The entire dataset was systematically divided into 10 subject subsets, with 9 subsets utilized for training and the remaining subset for testing in each iteration as shown in Fig. 1.</p> <p>Graph: Fig. 1 Feature extraction, analysis, and ML model pipeline</p> <p>For each combination of flash strength and eye, as indicated in Table 1, the dataset exhibited an imbalance that could potentially introduce bias towards the majority class (control). To rectify this, we applied synthetic minority over-sampling technique (SMOTE) (Chawla et al., [<reflink idref="bib14" id="ref156">14</reflink>]) to establish balance among the three classes—control, ASD, and ADHD. Subsequently, the widely utilized 'GridsearchCV' (LaValle et al., [<reflink idref="bib52" id="ref157">52</reflink>]) strategy was employed for hyperparameter optimization, aiming to identify optimal parameters for each classifier. To ensure impartiality and mitigate subject bias in parameter optimization, a subject-independent group 3-fold cross-validation strategy was applied to the training data. An exhaustive exploration of hyperparameters was conducted, and the best set of hyperparameters was selected based on the highest average cross-validation F1-score. Lastly, in the initial feature extraction and selection phase, where features were initially chosen based on individual predictiveness without considering inter-feature correlations, potentially leading to the presence of redundant and repetitive information. Consequently, we employed Random Forest-based feature selection, utilizing a mean decrease in impurity-based approach to identify and retain the most informative features. To report the performance of the ML models we opted for four different performance criteria i.e., accuracy, precision, recall and AUC-score. The formula to calculate these metrices are defined as:</p> <p>3</p> <p>Graph</p> <p>4</p> <p>Graph</p> <p>5</p> <p>Graph</p> <p>6</p> <p>Graph</p> <p>where TP is the total number of true positives, FP is total false positives, TN represents total number of true negatives and FN is the total false negatives. Accuracy can be misleading due to the existing imbalance in our dataset and to overcome that we report precision and recall for each class along with overall accuracy.</p> <hd id="AN0183973035-14">Results</hd> <p>The results of the time and time-frequency analysis yielded discriminative features crucial for distinguishing between control, ASD, and ADHD. In the classification phase, these characterized features played a pivotal role, manifesting in the accurate differentiation of these groups and are discussed in this section.</p> <hd id="AN0183973035-15">Time-Domain Analysis</hd> <p>Previous investigations focused on four key time domain parameters—, , , and —of the ERG. Among these parameters, emerged as a significant discriminative factor, particularly noteworthy in distinguishing between control vs. ASD and ASD vs. ADHD groups. Figure SI-4 visually presents the distribution of these parameters, and it is important to note that they deviated from normal distribution based on the Shapiro-Wilk normality test (Marmolejo-Ramos & González-Burgos, [<reflink idref="bib63" id="ref158">63</reflink>]; Shapiro & Wilk, [<reflink idref="bib81" id="ref159">81</reflink>]). Detailed results of the statistical analysis are presented in Table 3. Box plots in Figure SI-4 and the insights derived from Table 3 collectively emphasize that time domain indices, with a specific emphasis on , exhibit the high discriminatory power between control vs. ASD and ASD vs. ADHD groups. However, from Dunn's post hoc analysis it is crucial to acknowledge that time domain indices fall short in effectively separating the control and ADHD groups in this analysis of two flash strengths.</p> <p>Table 3 Statistical analysis of time domain indices</p> <p> <ephtml> <table frame="hsides" rules="groups"><thead><tr><th align="left" rowspan="2"><p>Intensity</p></th><th align="left" rowspan="2"><p>Eye</p></th><th align="left" rowspan="2"><p>Parameter</p></th><th align="left" rowspan="2"><p>Kruskal-Walli's test</p><p>(<italic>P</italic>-value)</p></th><th align="left" colspan="3"><p>Dunn's post-hoc analysis pairwise comparison (<italic>P</italic>-value)</p></th></tr><tr><th align="left"><p>Control vs. ASD</p></th><th align="left"><p>Control vs. ADHD</p></th><th align="left"><p>ASD vs. ADHD</p></th></tr></thead><tbody><tr><td align="left" rowspan="4"><p>446</p></td><td align="left" rowspan="4"><p>Right</p></td><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq18.gif" /></p></td><td align="left"><p>0.54</p></td><td align="left"><p>0.93</p></td><td align="left"><p><bold>0.92</bold></p></td><td align="left"><p>0.83</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq19.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.05</bold></p></td><td align="left"><p>< 0.05</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq20.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.16</bold></p></td><td align="left"><p>< 0.05</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq21.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.21</bold></p></td><td align="left"><p>< 0.001</p></td></tr><tr><td align="left" rowspan="4"><p>446</p></td><td align="left" rowspan="4"><p>Left</p></td><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq22.gif" /></p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>0.62</p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>< 0.05</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq23.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>0.27</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq24.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.39</bold></p></td><td align="left"><p>< 0.05</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq25.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.08</bold></p></td><td align="left"><p>< 0.001</p></td></tr><tr><td align="left" rowspan="4"><p>113</p></td><td align="left" rowspan="4"><p>Right</p></td><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq26.gif" /></p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>< 0.05</p></td><td align="left"><p><bold>0.05</bold></p></td><td align="left"><p>0.7</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq27.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.51</bold></p></td><td align="left"><p>< 0.001</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq28.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.15</bold></p></td><td align="left"><p>< 0.05</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq29.gif" /></p></td><td align="left"><p>< 0.001</p></td><td align="left"><p>< 0.001</p></td><td align="left"><p><bold>0.12</bold></p></td><td align="left"><p>< 0.001</p></td></tr><tr><td align="left" rowspan="4"><p>113</p></td><td align="left" rowspan="4"><p>Left</p></td><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq30.gif" /></p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>< 0.05</p></td><td align="left"><p><bold>0.29</bold></p></td><td align="left"><p>0.48</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq31.gif" /></p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>< 0.05</p></td><td align="left"><p><bold>0.99</bold></p></td><td align="left"><p>< 0.07</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq32.gif" /></p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>< 0.05</p></td><td align="left"><p><bold>0.37</bold></p></td><td align="left"><p>0.23</p></td></tr><tr><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq33.gif" /></p></td><td align="left"><p>< 0.05</p></td><td align="left"><p>< 0.05</p></td><td align="left"><p><bold>0.27</bold></p></td><td align="left"><p>< 0.05</p></td></tr></tbody></table> </ephtml> </p> <p>Statistical analysis with 0.05 and 0.001 level of significance. Time domain indices do not differ significantly between control and ADHD subjects shown in bolded <emph>P</emph>-value</p> <hd id="AN0183973035-16">DWT Descriptors</hd> <p>Given the limitations of time domain parameters in effectively distinguishing between control and ADHD groups, we carried out a further exploration of the DWT descriptors. Specifically, the individual coefficients of the 20a, 20b, 40a, 40b, OP80 and OP160 descriptors, to discern their discriminative potential. We performed a thresholding-based classification and Figure SI-5 presents the classification results, revealing that achieves AUC scores of 66%, 54%, and 69% in separating controls vs. ASD, control vs. ADHD, and ASD vs. ADHD, respectively. Intriguingly, DWT coefficients within OP160 descriptors consistently outperformed , particularly evident in differentiating control and ADHD groups (Figure SI-5, Top right). Additionally, some of these coefficients exhibited superior performance compared to in discerning between the ASD and ADHD groups (Figure SI-5, bottom). While the presented figure displays only a subset of coefficients due to their abundance in OP80 and OP160 descriptors, the highlighted ones were specifically chosen based on their superior AUC scores compared to .</p> <hd id="AN0183973035-17">VFCDM Features</hd> <p>DWT coefficients, particularly within OP160, exhibited notable efficacy in distinguishing control and ADHD subjects, as evidenced by high AUC scores. However, their performance falls short when tasked with separating control vs. ASD and ASD vs. ADHD. Furthermore, it is essential to acknowledge the limitation of DWT in providing high resolution in the higher frequency range. In response to these limitations, we pivot to the discussion of VFCDM components and their discriminative characteristics among control, ASD, and ADHD groups. This section aims to explore the potential of VFCDM in overcoming the identified limitations. Figure SI-6 shows the boxplot of interquartile range (IQR) of the VFCDM components and results of statistical analysis. We can see that as we go higher in the frequency range the difference between control vs. ASD and ASD vs. ADHD groups becomes more significant. We found similar patterns with other statistical features computed from VFCDM components. Figure SI-7 shows the thresholding-based classification performance of these features to separate control, ASD, and ADHD groups. We can see that the statistical features from the VFCDM frequency components achieves significantly higher AUC score compared to . Moreover, higher frequency components achieve higher AUC scores compared to lower frequency components. We note that the control and ADHD groups have a greater variance compared to the ASD group – especially within the 5th VFCDM component. This may reflect a less robust clinical diagnosis of the ADHD subjects with a wide heterogeneity compared to the ASD group where standardized tests are available [<reflink idref="bib50" id="ref160">50</reflink>], [<reflink idref="bib51" id="ref161">51</reflink>].</p> <p>Our exploration of the ERG signal encompassed the extraction of a comprehensive array of features through the utilization of DWT and VFCDM as listed in Table 2. In the critical phase of selecting an initial feature set for training the ML models, we systematically evaluated the discriminative capacity of each feature across different conditions. Notably, DWT coefficients within OP160 descriptors exhibited superior discriminative power for control vs. ADHD separation, surpassing the effectiveness of and achieving an AUC score exceeding 0.6, thus becoming the exclusive candidates for inclusion in our ML modeling. Moreover, our exploration of VFCDM, the components within the higher frequency range, specifically the last two components, exhibited the highest AUC score for distinguishing control vs. ASD and ASD vs. ADHD. Consequently, we selectively incorporated statistical features from these last two VFCDM components for ML modeling. Table 4 provides a detailed enumeration of the initial features meticulously chosen for training our ML models.</p> <p>Table 4 feature set to train ML models</p> <p> <ephtml> <table frame="hsides" rules="groups"><thead><tr><th align="left"><p>Modality</p></th><th align="left"><p>Feature name</p></th></tr></thead><tbody><tr><td align="left"><p>Time Domain indices</p></td><td align="left"><p><inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq40.gif" />, <inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq41.gif" />, <inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq42.gif" /></p></td></tr><tr><td align="left"><p>DWT coefficients</p></td><td align="left"><p>DWT coefficients in OP160 descriptors surpassing <inline-graphic mime-subtype="GIF" href="10803_2024_6290_Article_IEq43.gif" />and AUC score of minimum 0.6</p></td></tr><tr><td align="left"><p>Statistical Features from VFCDM components</p></td><td align="left"><p>maximum, minimum, standard deviation, inter-quartile range of 7th and 8th VFCDM component</p></td></tr></tbody></table> </ephtml> </p> <hd id="AN0183973035-18">Machine Learning Model Performance</hd> <p>In this subsection, we report the classification performance of the selected ML models across various combinations of the two flash strengths and each eye. The graphical representation in Figure SI-8 illustrates the accuracy attained by the trained ML models, providing a comprehensive overview of their performance. From Figure SI-8 the Random Forest classifier achieved the highest accuracy of 70% for the combination of 446 Td.s applied to right eye. Additionally, the detailed classification results for each individual class are presented in Table 5 shedding light on their specific performance concerning different classes. Figure SI-8 shows that Random Forest and XGBoost classifier consistently achieved better performance compared to other classifiers for almost all flash strength and eye combinations. For that reason, we only report these two classifiers in Table 5.</p> <p>Table 5 Classification performance of ML models</p> <p> <ephtml> <table frame="hsides" rules="groups"><thead><tr><th align="left"><p>Classifier</p></th><th align="left"><p>Flash strength (Td.s) / Eye</p></th><th align="left" colspan="3"><p>Precision</p></th><th align="left" colspan="3"><p>Recall</p></th><th align="left" rowspan="2"><p>Overall Accuracy</p></th></tr><tr><th align="left" colspan="2"><p>'</p></th><th align="left"><p><bold>Control</bold></p></th><th align="left"><p><bold>ASD</bold></p></th><th align="left"><p><bold>ADHD</bold></p></th><th align="left"><p><bold>Control</bold></p></th><th align="left"><p><bold>ASD</bold></p></th><th align="left"><p><bold>ADHD</bold></p></th></tr></thead><tbody><tr><td align="left" rowspan="4"><p>Random Forest</p></td><td align="left"><p><bold>446 / Right</bold></p></td><td align="left"><p><bold>0.76</bold></p></td><td align="left"><p><bold>0.65</bold></p></td><td align="left"><p><bold>0.64</bold></p></td><td align="left"><p><bold>0.71</bold></p></td><td align="left"><p><bold>0.75</bold></p></td><td char="." align="char"><p><bold>0.62</bold></p></td><td char="." align="char"><p><bold>0.70</bold></p></td></tr><tr><td align="left"><p>446 / Left</p></td><td align="left"><p>0.73</p></td><td align="left"><p>0.48</p></td><td align="left"><p>0.51</p></td><td align="left"><p>0.61</p></td><td align="left"><p>0.69</p></td><td char="." align="char"><p>0.58</p></td><td char="." align="char"><p>0.61</p></td></tr><tr><td align="left"><p>113 / Right</p></td><td align="left"><p>0.63</p></td><td align="left"><p>0.50</p></td><td align="left"><p>0.31</p></td><td align="left"><p>0.54</p></td><td align="left"><p>0.59</p></td><td char="." align="char"><p>0.63</p></td><td char="." align="char"><p>0.54</p></td></tr><tr><td align="left"><p>113 / Left</p></td><td align="left"><p>0.63</p></td><td align="left"><p>0.50</p></td><td align="left"><p>0.34</p></td><td align="left"><p>0.54</p></td><td align="left"><p>0.63</p></td><td char="." align="char"><p>0.55</p></td><td char="." align="char"><p>0.54</p></td></tr><tr><td align="left" rowspan="4"><p>XGBoost</p></td><td align="left"><p>446 / Right</p></td><td align="left"><p>0.72</p></td><td align="left"><p>0.60</p></td><td align="left"><p>0.59</p></td><td align="left"><p>0.66</p></td><td align="left"><p>0.67</p></td><td char="." align="char"><p>0.68</p></td><td char="." align="char"><p>0.66</p></td></tr><tr><td align="left"><p>446 / Left</p></td><td align="left"><p>0.69</p></td><td align="left"><p>0.49</p></td><td align="left"><p>0.46</p></td><td align="left"><p>0.60</p></td><td align="left"><p>0.66</p></td><td char="." align="char"><p>0.59</p></td><td char="." align="char"><p>0.59</p></td></tr><tr><td align="left"><p>113 / Right</p></td><td align="left"><p>0.61</p></td><td align="left"><p>0.50</p></td><td align="left"><p>0.35</p></td><td align="left"><p>0.54</p></td><td align="left"><p>0.64</p></td><td char="." align="char"><p>0.50</p></td><td char="." align="char"><p>0.54</p></td></tr><tr><td align="left"><p>113 / Left</p></td><td align="left"><p>0.63</p></td><td align="left"><p>0.49</p></td><td align="left"><p>0.36</p></td><td align="left"><p>0.53</p></td><td align="left"><p>0.59</p></td><td char="." align="char"><p>0.56</p></td><td char="." align="char"><p>0.53</p></td></tr></tbody></table> </ephtml> </p> <p>The model with the best performance is in bold</p> <hd id="AN0183973035-19">Discussion</hd> <p>This study marks the first initial comprehensive analysis of the ERG signal aimed at detecting individuals with ADHD and ASD from neurotypical subjects using a unified model. Our findings illuminate the significance of the among various time domain indices, showcasing its paramount role in achieving the highest separation, with an AUC score of 0.69, specifically between ASD and ADHD subjects. This notable discrimination power can be attributed to the distinct characteristics observed in ASD subjects, characterized by smaller b-waves, and ADHD subjects, marked by higher b-waves. also emerges as a robust discriminator between control and ASD subjects, underscoring its versatility. However, its limitations become apparent in distinguishing ADHD subjects from controls, yielding an AUC score of 0.54, akin to chance-level performance.</p> <p>Our in-depth examination of the ERG signal through the rigorous application of DWT and VFCDM has yielded a repertoire of highly sensitive features crucial for distinguishing between control, ASD, and ADHD conditions. Notably, our investigation has underscored the pivotal role of the higher frequency range (80–300 Hz) in harboring the most pertinent information for accurate differentiation. Exemplifying this, DWT coefficients within the OP160 descriptor exhibited an AUC score of 0.80, markedly improving upon the discriminatory capabilities of traditional time domain indices. Furthermore, VFCDM components encompassing the higher frequency range emerged as potent discriminators, surpassing the performance of the by a substantial margin in distinguishing between control and ASD, as well as ASD and ADHD. Our meticulous analysis involved the careful scrutiny of DWT descriptors and VFCDM components from varying frequency ranges, culminating in the selection of an initial set of highly sensitive features. These features were subsequently employed to train ML models, demonstrating the immense potential of TFS analysis of ERG signals in effectively identifying the distinct phenotypes associated with ASD and ADHD in comparison to neurotypicals.</p> <p>The retinal neurotransmitters and their regulation may contribute to the electrophysiological findings observed between ASD and ADHD. Whilst the pathophysiology of ASD and ADHD remains elusive – several genetic links to GABA and glutamate have been proposed along with genes encoding synaptic development and maturation (Del Campo et al., [<reflink idref="bib23" id="ref162">23</reflink>]; Mandic-Maravic et al., [<reflink idref="bib62" id="ref163">62</reflink>]; Mereu et al., [<reflink idref="bib67" id="ref164">67</reflink>]; Nguyen et al., [<reflink idref="bib70" id="ref165">70</reflink>]; Pavăl & Micluția, [<reflink idref="bib71" id="ref166">71</reflink>]). The involvement of the OPs and higher frequency components suggests a dopaminergic difference in the two clinical groups. Dopamine receptor agonists are potential therapeutic targets in ASD and the findings of this study may provide a potential marker with which to identify individuals most likely to benefit from dopamine agonists based on lower OPs [<reflink idref="bib92" id="ref167">92</reflink>] and evidence that dopamine antagonists selectively reduce the OPs (Wachtmeister, [<reflink idref="bib91" id="ref168">91</reflink>]). For ADHD the dopamine has been implicated in the pathophysiology using mouse [<reflink idref="bib93" id="ref169">93</reflink>] and human studies (Spencer et al., [<reflink idref="bib86" id="ref170">86</reflink>]) and consequently the elevated OPs may help in screening and management of ADHD (Del Campo et al., [<reflink idref="bib23" id="ref171">23</reflink>]; Mandic-Maravic et al., [<reflink idref="bib62" id="ref172">62</reflink>]; Mereu et al., [<reflink idref="bib67" id="ref173">67</reflink>]).</p> <p>A recent report by Dubois et al. (Dubois et al., [<reflink idref="bib27" id="ref174">27</reflink>]) in older subjects with ADHD reported a longer a-wave time to peak in female subjects but no significant findings in the b-wave amplitude under LA conditions. The authors suggest that a larger cohort is required to fully evaluate the potential of the ERG as a biomarker for ADHD. Our findings provide some support for the ERG as a potential biomarker for ADHD and ASD, although heterogeneity within these clinical populations will likely require a larger dataset with further evaluation of the effects of medications on the OPs, although Lee et al. [<reflink idref="bib28" id="ref175">28</reflink>] reported no significant medication interactions with the b-wave in 5 ADHD subjects.</p> <p>Our research outcomes, derived from a comprehensive analysis of critical features, resonate with previous studies, such as the work by Gauvin et al. (Gauvin et al., [<reflink idref="bib36" id="ref176">36</reflink>], [<reflink idref="bib37" id="ref177">37</reflink>]). They observed that OPs situated in the high-frequency range are typically affected in vasculopathies like diabetic retinopathy, while the a-wave and b-wave, found in the low-frequency range, remain relatively unaffected (Gauvin et al., [<reflink idref="bib36" id="ref178">36</reflink>], [<reflink idref="bib37" id="ref179">37</reflink>]). Our findings align with these observations, emphasizing that features extracted from high-frequency regions harbor the most informative content for detecting ASD as previously observed with the 'Haar' mother wavelet (Constable et al., 2022) and also ADHD in this study. This underscores the effectiveness of TFs analysis in extracting OPs and high-frequency information, showcasing their sensitivity in detection. The collective results suggest that ASD and ADHD predominantly impact the OPs of the ERG signal originating from amacrine cells, indicating a potential deficit in the dopaminergic signaling pathway (Wachtmeister, [<reflink idref="bib91" id="ref180">91</reflink>]). In our analysis, we evaluated two flash strengths based on prior studies identifying significant group differences at these flash strengths. Specifically, the stronger 446 Td.s flash strength, associated with the photopic hill plateau and dominated by the ON-pathway, exhibited superior classification performance compared to the 113 Td.s flash strength associated with the peak of the photopic hill and dominated by the OFF-pathways (Hamilton et al., [<reflink idref="bib42" id="ref181">42</reflink>]). Moreover, our findings highlighted enhanced classification performance for signals collected from the right eye. This may be attributed to the right eye data collection sequence; participants may be more alert, co-operative and have improved fixation during this first phase. Another contributing factor could be the practical ease of handling the RETeval by right-handed examiners for whom aligning with the right eye is easier than stretching across the body for left eye testing, potentially influencing response levels for the left eye in our experimental setup.</p> <p>Acknowledging study limitations, including the restriction to singular ASD or ADHD diagnoses, warrants future exploration of mixed diagnostic (Krakowski et al., [<reflink idref="bib49" id="ref182">49</reflink>]) classifications. Factors like sex proportion, medication history, and age effects (childhood vs. adult) require further scrutiny to ensure model validity. Despite these considerations, our study highlights the potential utility of the ERG, specifically with spectral-domain analysis, as a supplementary tool for identifying ASD phenotypes, offering valuable insights for future biomarker investigations in this domain.</p> <hd id="AN0183973035-20">Conclusion</hd> <p>This study represents the inaugural attempt to leverage TFS analysis for extracting highly sensitive features from a minimal number of ERG signals to classify common neurodevelopmental disorders, culminating in the development of a ML model capable of detecting ASD and ADHD from neurotypical Controls with an overall accuracy of 70%. Notwithstanding these advancements, it is crucial to acknowledge the potential co-occurrence of ASD and ADHD within the same individual, a facet not addressed in the current study. Future investigations will delve into the prospect of employing TFS analysis to differentiate between concomitant ASD and ADHD in comparison to isolated cases of ASD and ADHD. Notably, optimal ASD detection was attained when ERG signals were obtained using a flash strength of 446 Td.s applied to the right eye. While this discovery holds promise, further studies will investigate a wider range of flash strengths, the influence of sex, co-occurring conditions and clinical severity to further explore the retina as potential biomarker for neurodevelopmental conditions to support diagnosis and management.</p> <hd id="AN0183973035-21">Declarations</hd> <p></p> <hd id="AN0183973035-22">Conflict of interest</hd> <p>The authors declare no conflicts of interest.</p> <hd id="AN0183973035-23">Electronic supplementary material</hd> <p>Below is the link to the electronic supplementary material.</p> <p>Graph: Supplementary Material 1</p> <hd id="AN0183973035-24">Publisher's Note</hd> <p>Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.</p> <ref id="AN0183973035-25"> <title> References </title> <blist> <bibl id="bib1" idref="ref80" type="bt">1</bibl> <bibtext> Adamou, M, Fullen, T, & Jones, S. L. (2020). EEG for Diagnosis of Adult ADHD: A Systematic Review With Narrative Analysis. Frontiers in Psychiatry, 11. https://www.frontiersin.org/journals/psychiatry/articles/https://doi.org/10.3389/fpsyt.2020.00871.</bibtext> </blist> <blist> <bibl id="bib2" idref="ref19" type="bt">2</bibl> <bibtext> Aktas H, Esin IS, Dursun OB. Is it possible to recognize children diagnosed with ADHD from their facial anthropometric measures? A case-control study. Medical Hypotheses. 2020; 140: 109649. 10.1016/j.mehy.2020.109649. 32135446</bibtext> </blist> <blist> <bibl id="bib3" idref="ref96" type="bt">3</bibl> <bibtext> Al Abdlseaed A, McTaggart Y, Ramage T, Hamilton R, McCulloch DL. Light- and dark-adapted electroretinograms (ERGs) and ocular pigmentation: Comparison of brown- and blue-eyed cohorts. Documenta Ophthalmologica Advances in Ophthalmology. 2010; 121; 2: 135-146. 10.1007/s10633-010-9240-3. 20665068</bibtext> </blist> <blist> <bibl id="bib4" idref="ref111" type="bt">4</bibl> <bibtext> American Psychiatric Association, A. P, &, & Association, A. P. (1994). Diagnostic and statistical manual of mental disorders: DSM-IV (Vol. 4). American psychiatric association Washington, DC. https://<ulink href="http://www.gammaconstruction.mu/sites/default/files/webform/cvs/pdf-diagnostic-and-statistical-manual-of-mental-disorders-dsm-iv-american-psychiatric-association-pdf-download-free-book-9223cc7.pdf">www.gammaconstruction.mu/sites/default/files/webform/cvs/pdf-diagnostic-and-statistical-manual-of-mental-disorders-dsm-iv-american-psychiatric-association-pdf-download-free-book-9223cc7.pdf</ulink>.</bibtext> </blist> <blist> <bibl id="bib5" idref="ref74" type="bt">5</bibl> <bibtext> Anderson NP, Feldman JA, Kolko DJ, Pilkonis PA, Lindhiem O. National norms for the Vanderbilt ADHD diagnostic parent rating scale in children. Journal of Pediatric Psychology. 2022; 47; 6: 652-661. 10.1093/jpepsy/jsab132. 34986222. 9172842</bibtext> </blist> <blist> <bibl id="bib6" idref="ref81" type="bt">6</bibl> <bibtext> Ari B, Sobahi N, Alçin ÖF, Sengur A, Acharya UR. Accurate detection of autism using Douglas-Peucker algorithm, sparse coding based feature mapping and convolutional neural network techniques with EEG signals. Computers in Biology and Medicine. 2022; 143: 105311. 10.1016/j.compbiomed.2022.105311. 35158117</bibtext> </blist> <blist> <bibl id="bib7" idref="ref82" type="bt">7</bibl> <bibtext> Arns M, Conners CK, Kraemer HC. A decade of EEG Theta/Beta Ratio Research in ADHD: A Meta-analysis. Journal of Attention Disorders. 2013; 17; 5: 374-383. 10.1177/1087054712460087. 23086616</bibtext> </blist> <blist> <bibl id="bib8" idref="ref24" type="bt">8</bibl> <bibtext> Asi H, Perlman I. Relationships between the electroretinogram a-wave, b-wave and oscillatory potentials and their application to clinical diagnosis. Documenta Ophthalmologica Advances in Ophthalmology. 1992; 79; 2: 125-139. 10.1007/BF00156572. 1591967</bibtext> </blist> <blist> <bibl id="bib9" idref="ref14" type="bt">9</bibl> <bibtext> Baygin M, Dogan S, Tuncer T, Datta Barua P, Faust O, Arunkumar N, Abdulhay EW, Palmer E, Acharya R. Automated ASD detection using hybrid deep lightweight features extracted from EEG signals. Computers in Biology and Medicine. 2021; 134: 104548. 10.1016/j.compbiomed.2021.104548. 34119923</bibtext> </blist> <blist> <bibtext> Bellato A, Arora I, Kochhar P, Ropar D, Hollis C, Groom MJ. Heart rate variability in children and adolescents with autism, ADHD and co-occurring autism and ADHD, during Passive and active experimental conditions. Journal of Autism and Developmental Disorders. 2022; 52; 11: 4679-4691. 10.1007/s10803-021-05244-w. 34716841</bibtext> </blist> <blist> <bibtext> Bosl W, Tierney A, Tager-Flusberg H, Nelson C. EEG complexity as a biomarker for autism spectrum disorder risk. BMC Medicine. 2011; 9; 1: 18. 10.1186/1741-7015-9-18. 21342500. 3050760</bibtext> </blist> <blist> <bibtext> Bosl, W. J, Tager-Flusberg, H, & Nelson, C. A. (2018). EEG Analytics for early detection of Autism Spectrum disorder: A data-driven approach. Scientific Reports, 8(1). https://doi.org/10.1038/s41598-018-24318-x.</bibtext> </blist> <blist> <bibtext> Brown, T. E. (1996). Attention-deficit disorder scales. Psychological Corporation.</bibtext> </blist> <blist> <bibtext> Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE: Synthetic minority over-sampling technique. Journal of Artificial Intelligence Research. 2002; 16: 321-357. 10.1613/jair.953</bibtext> </blist> <blist> <bibtext> Chiquita S, Rodrigues-Neves AC, Baptista FI, Carecho R, Moreira PI, Castelo-Branco M, Ambrósio AF. The retina as a window or Mirror of the brain changes detected in Alzheimer's Disease: Critical aspects to Unravel. Molecular Neurobiology. 2019; 56; 8: 5416-5435. 10.1007/s12035-018-1461-6. 30612332</bibtext> </blist> <blist> <bibtext> Clarke AR, Barry RJ, McCarthy R, Selikowitz M. EEG analysis of children with attention-deficit/hyperactivity disorder and comorbid reading disabilities. Journal of Learning Disabilities. 2002; 35; 3: 276-285. 10.1177/002221940203500309. 15493324</bibtext> </blist> <blist> <bibtext> Conners, C. K, Pitkanen, J, & Rzepa, S. R. (2011). Conners 3rd edition (conners 3; conners 2008). Encyclopedia of Clinical Neuropsychology, 675–678.</bibtext> </blist> <blist> <bibtext> Constable, P. A, Gaigg, S. B, Bowler, D. M, Jägle, H, & Thompson, D. A. (2016). Full-field electroretinogram in autism spectrum disorder. Documenta Ophthalmologica. Advances in Ophthalmology, 132(2), 83–99. https://doi.org/10.1007/s10633-016-9529-y</bibtext> </blist> <blist> <bibtext> Constable, P. A, Lee, I. O, Marmolejo-Ramos, F, Skuse, D. H, & Thompson, D. A. (2021). The photopic negative response in autism spectrum disorder. Clinical and Experimental Optometry, 104(8), 841–847. https://doi.org/10.1080/08164622.2021.1903808</bibtext> </blist> <blist> <bibtext> Constable, P. A, Marmolejo-Ramos, F, Gauthier, M, Lee, I. O, Skuse, D. H, & Thompson, D. A. (2022). Discrete wavelet transform analysis of the electroretinogram in autism spectrum disorder and attention deficit hyperactivity disorder. Frontiers in Neuroscience, 16.</bibtext> </blist> <blist> <bibtext> Constable, P. A, Ritvo, E. R, Ritvo, A. R, Lee, I. O, McNair, M. L, Stahl, D, Sowden, J, Quinn, S, Skuse, D. H, & Thompson, D. A. (2020). Light-adapted electroretinogram differences in autism spectrum disorder. Journal of Autism and Developmental Disorders, 50, 2874–2885.</bibtext> </blist> <blist> <bibtext> Constantino, J. N. (2021). Social Responsiveness Scale. In F. R. Volkmar (Ed.), Encyclopedia of Autism Spectrum Disorders (pp. 4457–4467). Springer International Publishing. https://doi.org/10.1007/978-3-319-91280-6_296.</bibtext> </blist> <blist> <bibtext> Del Campo N, Chamberlain SR, Sahakian BJ, Robbins TW. The roles of dopamine and noradrenaline in the pathophysiology and treatment of attention-deficit/hyperactivity disorder. Biological Psychiatry. 2011; 69; 12: e145-157. 10.1016/j.biopsych.2011.02.036. 21550021</bibtext> </blist> <blist> <bibtext> Demmin DL, Davis Q, Roché M, Silverstein SM. Electroretinographic anomalies in schizophrenia. Journal of Abnormal Psychology. 2018; 127; 4: 417-428. 10.1037/abn0000347. 29745706</bibtext> </blist> <blist> <bibtext> Diagnostic and statistical manual of mental disorders: DSM-5™, 5th ed (pp. xliv, 947). (2013). American Psychiatric Publishing, Inc. https://doi.org/10.1176/appi.books.9780890425596.</bibtext> </blist> <blist> <bibtext> Djemal R, AlSharabi K, Ibrahim S, Alsuwailem A. EEG-Based computer aided diagnosis of Autism Spectrum Disorder using Wavelet, Entropy, and ANN. BioMed Research International. 2017; 2017: 9816591. 10.1155/2017/9816591. 28484720. 5412163</bibtext> </blist> <blist> <bibtext> Dubois MA, Pelletier CA, Mérette C, Jomphe V, Turgeon R, Bélanger RE, Grondin S, Hébert M. Evaluation of electroretinography (ERG) parameters as a biomarker for ADHD. Progress in Neuro-Psychopharmacology and Biological Psychiatry. 2023; 127: 110807. 10.1016/j.pnpbp.2023.110807. 37290571</bibtext> </blist> <blist> <bibtext> Dunn OJ. Multiple comparisons using Rank sums. Technometrics. 1964; 6; 3: 241-252. 10.1080/00401706.1964.10490181</bibtext> </blist> <blist> <bibtext> DuPaul, G. J, Power, T. J, Anastopoulos, A. D, & Reid, R. (1998). ADHD Rating Scale—IV: Checklists, norms, and clinical interpretation (pp. viii, 79). The Guilford Press.</bibtext> </blist> <blist> <bibtext> Duvall S, Armstrong K, Shahabuddin A, Grantz C, Fein D, Lord C. A road map for identifying autism spectrum disorder: Recognizing and evaluating characteristics that should raise red or pink flags to guide accurate differential diagnosis. The Clinical Neuropsychologist. 2022; 36; 5: 1172-1207. 10.1080/13854046.2021.1921276. 34121610</bibtext> </blist> <blist> <bibtext> Elbattah, M, Guérin, J. L, Carette, R, Cilia, F, & Dequen, G. (2020). NLP-Based Approach to Detect Autism Spectrum Disorder in Saccadic Eye Movement. 2020 IEEE Symposium Series on Computational Intelligence (SSCI), 1581–1587. https://doi.org/10.1109/SSCI47803.2020.9308238.</bibtext> </blist> <blist> <bibtext> Elder JH, Kreider CM, Brasher SN, Ansell M. Clinical impact of early diagnosis of autism on the prognosis and parent–child relationships. Psychology Research and Behavior Management. 2017; 10: 283-292. 10.2147/PRBM.S117499. 28883746. 5576710</bibtext> </blist> <blist> <bibtext> Friedel EBN, Schäfer M, Endres D, Maier S, Runge K, Bach M, Heinrich SP, Ebert D, Domschke K, van Tebartz L, Nickel K. Electroretinography in adults with high-functioning autism spectrum disorder. Autism Research: Official Journal of the International Society for Autism Research. 2022; 15; 11: 2026-2037. 10.1002/aur.2823. 36217563</bibtext> </blist> <blist> <bibtext> Gauvin, M, Lina, J. M, & Lachapelle, P. (2014). Advance in ERG analysis: From peak time and amplitude to frequency, power, and energy. BioMed Research International, 2014.</bibtext> </blist> <blist> <bibtext> Gauvin M, Little JM, Lina JM, Lachapelle P. Functional decomposition of the human ERG based on the discrete wavelet transform. Journal of Vision. 2015; 15; 16: 14. 10.1167/15.16.14. 26746684</bibtext> </blist> <blist> <bibtext> Gauvin, M, Dorfman, A. L, Trang, N, Gauthier, M, Little, J. M, Lina, J. M, & Lachapelle, P. (2016). Assessing the contribution of the Oscillatory potentials to the Genesis of the photopic ERG with the Discrete Wavelet transform. BioMed Research International, 2016(e2790194). https://doi.org/10.1155/2016/2790194.</bibtext> </blist> <blist> <bibtext> Gauvin M, Sustar M, Little JM, Brecelj J, Lina JM, Lachapelle P. Quantifying the ON and OFF contributions to the Flash ERG with the Discrete Wavelet transform. Translational Vision Science & Technology. 2017; 6; 1: 3. 10.1167/tvst.6.1.3</bibtext> </blist> <blist> <bibtext> Ghiassian, S, Greiner, R, Jin, P, & Brown, M. R. (2016). Using functional or structural magnetic resonance images and personal characteristic data to identify ADHD and autism. PloS One, 11(12), e0166934.</bibtext> </blist> <blist> <bibtext> Gotham K, Risi S, Pickles A, Lord C. The Autism Diagnostic Observation schedule: Revised algorithms for improved diagnostic validity. Journal of Autism and Developmental Disorders. 2007; 37; 4: 613-627. 10.1007/s10803-006-0280-1. 17180459</bibtext> </blist> <blist> <bibtext> Gross G, Tursini K, Albuisson E, Angioi-Duprez K, Conart JB, Dorr L, Schwan V, Schwitzer T. Bipolar disorders and retinal electrophysiological markers (BiMAR): Study protocol for a comparison of electroretinogram measurements between subjects with bipolar disorder and a healthy control group. Frontiers in Psychiatry. 2022; 13: 960512. 10.3389/fpsyt.2022.960512. 36159928. 9492949</bibtext> </blist> <blist> <bibtext> Hadders-Algra, M. (2021). Early Diagnostics and early intervention in Neurodevelopmental disorders—Age-Dependent challenges and opportunities. Journal of Clinical Medicine, 10(4). https://doi.org/10.3390/jcm10040861. Article 4.</bibtext> </blist> <blist> <bibtext> Hamilton R, Bees MA, Chaplin CA, McCulloch DL. The luminance-response function of the human photopic electroretinogram: A mathematical model. Vision Research. 2007; 47; 23: 2968-2972. 10.1016/j.visres.2007.04.020. 17889925</bibtext> </blist> <blist> <bibtext> Hatch B, Kadlaskar G, Miller M. Diagnosis and treatment of children and adolescents with autism and ADHD. Psychology in the Schools. 2023; 60; 2: 295-311. 10.1002/pits.22808. 37065905</bibtext> </blist> <blist> <bibtext> Hébert M, Mérette C, Gagné AM, Paccalet T, Moreau I, Lavoie J, Maziade M. The Electroretinogram May differentiate Schizophrenia from bipolar disorder. Biological Psychiatry. 2020; 87; 3: 263-270. 10.1016/j.biopsych.2019.06.014. 31443935</bibtext> </blist> <blist> <bibtext> Hodges H, Fealko C, Soares N. Autism spectrum disorder: Definition, epidemiology, causes, and clinical evaluation. Translational Pediatrics. 2020; 9; Suppl 1: S55-S65. 10.21037/tp.2019.09.09. 32206584. 7082249</bibtext> </blist> <blist> <bibtext> Kang J, Han X, Song J, Niu Z, Li X. The identification of children with autism spectrum disorder by SVM approach on EEG and eye-tracking data. Computers in Biology and Medicine. 2020; 120: 103722. 10.1016/j.compbiomed.2020.103722. 32250854</bibtext> </blist> <blist> <bibtext> Kleberg, J. L, Frick, M. A, & Brocki, K. C. (2023). Eye-movement indices of arousal predict ADHD and comorbid externalizing symptoms over a 2-year period. Scientific Reports, 13(1). https://doi.org/10.1038/s41598-023-31697-3.</bibtext> </blist> <blist> <bibtext> Kolb, H, Fernandez, E, & Nelson, R. (Eds.). (1995). Webvision: The Organization of the retina and visual system. University of Utah Health Sciences Center. <ulink href="http://www.ncbi.nlm.nih.gov/books/NBK11530/">http://www.ncbi.nlm.nih.gov/books/NBK11530/</ulink>.</bibtext> </blist> <blist> <bibtext> Krakowski AD, Cost KT, Szatmari P, Anagnostou E, Crosbie J, Schachar R, Duku E, Georgiades S, Ayub M, Kelley E, Nicolson R, Pullenayegum E, Barnett-Tapia C. Characterizing the ASD-ADHD phenotype: Measurement structure and invariance in a clinical sample. Journal of Child Psychology and Psychiatry and Allied Disciplines. 2022; 63; 12: 1534-1543. 10.1111/jcpp.13609. 35342939</bibtext> </blist> <blist> <bibtext> Kruskal WH, Wallis WA. Use of ranks in One-Criterion Variance Analysis. Journal of the American Statistical Association. 1952; 47; 260: 583-621. 10.2307/2280779</bibtext> </blist> <blist> <bibtext> Kuttala D, Mahapatra D, Subramanian R, Oruganti VRM. Dense attentive GAN-based one-class model for detection of autism and ADHD. Journal of King Saud University-Computer and Information Sciences. 2022; 34; 10: 10444-10458</bibtext> </blist> <blist> <bibtext> LaValle SM, Branicky MS, Lindemann SR. On the relationship between classical Grid search and probabilistic roadmaps. The International Journal of Robotics Research. 2004; 23; 7–8: 673-692. 10.1177/0278364904045481</bibtext> </blist> <blist> <bibtext> Lee, D. Y, Shin, Y, Park, R. W, Cho, S. M, Han, S, Yoon, C, Choo, J, Shim, J. M, Kim, K, Jeon, S. W, & Kim, S. J. (2023). Use of eye tracking to improve the identification of attention-deficit/hyperactivity disorder in children. Scientific Reports, 13(1). https://doi.org/10.1038/s41598-023-41654-9.</bibtext> </blist> <blist> <bibtext> Lee, I. O, Skuse, D. H, Constable, P. A, Marmolejo-Ramos, F, Olsen, L. R, & Thompson, D. A. (2022). The electroretinogram b-wave amplitude: A differential physiological measure for attention deficit hyperactivity disorder and autism spectrum disorder. Journal of Neurodevelopmental Disorders, 14(1), 30. https://doi.org/10.1186/s11689-022-09440-2</bibtext> </blist> <blist> <bibtext> London, A, Benhar, I, & Schwartz, M. (2013). The retina as a window to the brain—from eye research to CNS disorders. Nature Reviews Neurology, 9(1). https://doi.org/10.1038/nrneurol.2012.227.</bibtext> </blist> <blist> <bibtext> Lord C, Rutter M, Goode S, Heemsbergen J, Jordan H, Mawhood L, Schopler E. Autism diagnostic observation schedule: A standardized observation of communicative and social behavior. Journal of Autism and Developmental Disorders. 1989; 19; 2: 185-212. 10.1007/BF02211841. 2745388</bibtext> </blist> <blist> <bibtext> Lord C, Rutter M, Le Couteur A. Autism Diagnostic Interview-Revised: A revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. Journal of Autism and Developmental Disorders. 1994; 24; 5: 659-685. 10.1007/BF02172145. 7814313</bibtext> </blist> <blist> <bibtext> Lord C, Risi S, Lambrecht L, Cook EH, Leventhal BL, DiLavore PC, Pickles A, Rutter M. The Autism Diagnostic Observation Schedule—Generic: A standard measure of social and communication deficits associated with the spectrum of autism. Journal of Autism and Developmental Disorders. 2000; 30: 205-223. 11055457</bibtext> </blist> <blist> <bibtext> Malwane MI, Nguyen EB, Trejo S, Kim EY, Cucalón-Calderón JR. A delayed diagnosis of Autism Spectrum Disorder in the setting of Complex attention deficit hyperactivity disorder. Cureus. 2022; 14; 6: e25825. 10.7759/cureus.25825. 35836458. 9273190</bibtext> </blist> <blist> <bibtext> Manjur, S. M, Hossain, M.-B, Constable, P. A, Thompson, D. A, Marmolejo-Ramos, F, Lee, I. O, & Posada Quintero, H. F. (2023). Spectral analysis of electroretinography to differentiate autism spectrum disorder and attention deficit hyperactivity disorder. 2023 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI) (pp. 1–4). https://doi.org/10.1109/BHI58575.2023.10313406</bibtext> </blist> <blist> <bibtext> Manjur, S. M, Hossain, M.-B, Constable, P. A, Thompson, D. A, Marmolejo-Ramos, F, Lee, I. O, Skuse, D. H, & Posada-Quintero, H. F. (2022). Detecting autism spectrum disorder using spectral analysis of electroretinogram and machine learning: Preliminary results. 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) (pp. 3435–3438). https://doi.org/10.1109/EMBC48229.2022.9871173</bibtext> </blist> <blist> <bibtext> Mandic-Maravic V, Grujicic R, Milutinovic L, Munjiza-Jovanovic A, Pejovic-Milovancevic M. Dopamine in Autism Spectrum disorders—Focus on D2/D3 partial agonists and their possible use in treatment. Frontiers in Psychiatry. 2022; 12: 787097. 10.3389/fpsyt.2021.787097. 35185637. 8850940</bibtext> </blist> <blist> <bibtext> Marmolejo-Ramos F, González-Burgos J. A power comparison of various tests of univariate normality on ex-gaussian distributions. Methodology: European Journal of Research Methods for the Behavioral and Social Sciences. 2013; 9; 4: 137-149. 10.1027/1614-2241/a000059</bibtext> </blist> <blist> <bibtext> Matlis S, Boric K, Chu CJ, Kramer MA. Robust disruptions in electroencephalogram cortical oscillations and large-scale functional networks in autism. BMC Neurology. 2015; 15; 1: 97. 10.1186/s12883-015-0355-8. 26111798. 4482270</bibtext> </blist> <blist> <bibtext> McConaughy, S. H, & Achenbach, T. M. (2009). Manual for the ASEBA direct observation form. Burlington, VT: University of Vermont Research Center for Children. Youth, and Families.</bibtext> </blist> <blist> <bibtext> Mello, L. G. M, Paraguay, I. B. B, Andrade, T, de Rocha, S, Barbosa, A. A, Oyamada, E. R, M. K, & Monteiro, M. L. R. (2022). Electroretinography reveals retinal dysfunction in Parkinson's disease despite normal high-resolution optical coherence tomography findings. Parkinsonism & Related Disorders, 101, 90–95. https://doi.org/10.1016/j.parkreldis.2022.06.018.</bibtext> </blist> <blist> <bibtext> Mereu M, Contarini G, Buonaguro EF, Latte G, Managò F, Iasevoli F, de Bartolomeis A, Papaleo F. Dopamine transporter (DAT) genetic hypofunction in mice produces alterations consistent with ADHD but not schizophrenia or bipolar disorder. Neuropharmacology. 2017; 121: 179-194. 10.1016/j.neuropharm.2017.04.037. 28454982</bibtext> </blist> <blist> <bibtext> Merzon, L, Pettersson, K, Aronen, E. T, Huhdanpää, H, Seesjärvi, E, Henriksson, L, MacInnes, W. J, Mannerkoski, M, Macaluso, E, & Salmi, J. (2022). Eye movement behavior in a real-world virtual reality task reveals ADHD in children. Scientific Reports, 12(1). https://doi.org/10.1038/s41598-022-24552-4.</bibtext> </blist> <blist> <bibtext> Muszkat M, de Mello CB, Muñoz P, de Lucci OL, David TK, de Siqueira VF, Otta E. Face scanning in Autism Spectrum disorder and attention Deficit/Hyperactivity disorder: Human Versus Dog face scanning. Frontiers in Psychiatry. 2015; 6: 150. 10.3389/fpsyt.2015.00150. 26557097. 4615933</bibtext> </blist> <blist> <bibtext> Nguyen M, Roth A, Kyzar EJ, Poudel MK, Wong K, Stewart AM, Kalueff AV. Decoding the contribution of dopaminergic genes and pathways to autism spectrum disorder (ASD). Neurochemistry International. 2014; 66: 15-26. 10.1016/j.neuint.2014.01.002. 24412511</bibtext> </blist> <blist> <bibtext> Pavăl D, Micluția IV. The dopamine hypothesis of Autism Spectrum Disorder Revisited: Current status and future prospects. Developmental Neuroscience. 2021; 43; 2: 73-83. 10.1159/000515751. 34010842</bibtext> </blist> <blist> <bibtext> Perlman, I. (1995). The Electroretinogram: ERG. In H. Kolb, E. Fernandez, & R. Nelson (Eds.), Webvision: The Organization of the retina and visual system. University of Utah Health Sciences Center. <ulink href="http://www.ncbi.nlm.nih.gov/books/NBK11554/">http://www.ncbi.nlm.nih.gov/books/NBK11554/</ulink>.</bibtext> </blist> <blist> <bibtext> Posada-Quintero, H. F, Manjur, S. M, Hossain, Md. B, Marmolejo-Ramos, F, Lee, I. O, Skuse, D. H, Thompson, D. A, & Constable, P. A. (2023). Autism spectrum disorder detection using variable frequency complex demodulation of the electroretinogram. Research in Autism Spectrum Disorders, 109, 102258. https://doi.org/10.1109/EMBC48229.2022.9871173</bibtext> </blist> <blist> <bibtext> Puce A, Hämäläinen MS. A review of issues related to Data Acquisition and Analysis in EEG/MEG studies. Brain Sciences. 2017; 7; 6: 58. 10.3390/brainsci7060058. 28561761. 5483631</bibtext> </blist> <blist> <bibtext> Quintana, Q, Benedetto, M. L, Maldonado, M. M, E, A. C. V, & Contin, M. A. (2016). Electroretinography: A biopotential to assess the function/dysfunction of the retina. Journal of Physics: Conference Series, 705(1), 012053. https://doi.org/10.1088/1742-6596/705/1/012053.</bibtext> </blist> <blist> <bibtext> Robson AG, Frishman LJ, Grigg J, Hamilton R, Jeffrey BG, Kondo M, Li S, McCulloch DL. ISCEV Standard for full-field clinical electroretinography (2022 update). Documenta Ophthalmologica Advances in Ophthalmology. 2022; 144; 3: 165-177. 10.1007/s10633-022-09872-0. 35511377. 9192408</bibtext> </blist> <blist> <bibtext> Rutter, M. (2003). Social communication questionnaire. (No Title). https://cir.nii.ac.jp/crid/1370285712560901636.</bibtext> </blist> <blist> <bibtext> Sannita WG, Maggi L, Germini PL, Fioretto M. Correlation with age and sex of flash-evoked electroretinogram and retinal oscillatory potentials recorded with skin electrodes. Documenta Ophthalmologica Advances in Ophthalmology. 1989; 71; 4: 413-419. 10.1007/BF00152769. 2791848</bibtext> </blist> <blist> <bibtext> Schmitt LM, Cook EH, Sweeney JA, Mosconi MW. Saccadic eye movement abnormalities in autism spectrum disorder indicate dysfunctions in cerebellum and brainstem. Molecular Autism. 2014; 5; 1: 47. 10.1186/2040-2392-5-47. 25400899. 4233053</bibtext> </blist> <blist> <bibtext> Schopler E, Reichler RJ, DeVellis RF, Daly K. Toward objective classification of childhood autism: Childhood Autism Rating Scale (CARS). Journal of Autism and Developmental Disorders. 1980; 10; 1: 91-103. 10.1007/BF02408436. 6927682</bibtext> </blist> <blist> <bibtext> Shapiro SS, Wilk MB. An analysis of Variance Test for Normality (complete samples). Biometrika. 1965; 52; 3/4: 591-611. 10.2307/2333709</bibtext> </blist> <blist> <bibtext> Shensa MJ. The discrete wavelet transform: Wedding the a trous and Mallat algorithms. IEEE Transactions on Signal Processing. 1992; 40; 10: 2464-2482. 10.1109/78.157290</bibtext> </blist> <blist> <bibtext> Shoeibi, A, Sadeghi, D, Moridian, P, Ghassemi, N, Heras, J, Alizadehsani, R, Khadem, A, Kong, Y, Nahavandi, S, Zhang, Y. D, & Gorriz, J. M. (2021). Automatic Diagnosis of Schizophrenia in EEG Signals Using CNN-LSTM Models. Frontiers in Neuroinformatics, 15. https://www.frontiersin.org/articles/https://doi.org/10.3389/fninf.2021.777977.</bibtext> </blist> <blist> <bibtext> Skuse D, Warrington R, Bishop D, Chowdhury U, Lau J, Mandy W, Place M. The Developmental, Dimensional and Diagnostic interview (3di): A Novel Computerized Assessment for Autism Spectrum disorders. Journal of the American Academy of Child & Adolescent Psychiatry. 2004; 43; 5: 548-558. 10.1097/00004583-200405000-00008</bibtext> </blist> <blist> <bibtext> Song P, Zha M, Yang Q, Zhang Y, Li X, Rudan I. The prevalence of adult attention-deficit hyperactivity disorder: A global systematic review and meta-analysis. Journal of Global Health. 2021; 11: 04009. 10.7189/jogh.11.04009. 33692893. 7916320</bibtext> </blist> <blist> <bibtext> Spencer TJ, Biederman J, Madras BK, Faraone SV, Dougherty DD, Bonab AA, Fischman AJ. In vivo neuroreceptor imaging in attention-deficit/hyperactivity disorder: A focus on the dopamine transporter. Biological Psychiatry. 2005; 57; 11: 1293-1300. 10.1016/j.biopsych.2005.03.036. 15950001</bibtext> </blist> <blist> <bibtext> Stockton RA, Slaughter MM. B-wave of the electroretinogram. A reflection of ON bipolar cell activity. The Journal of General Physiology. 1989; 93; 1: 101-122. 10.1085/jgp.93.1.101. 2915211</bibtext> </blist> <blist> <bibtext> Thompson, D. A, Feather, S, Stanescu, H. C, Freudenthal, B, Zdebik, A. A, Warth, R, Ognjanovic, M, Hulton, S. A, Wassmer, E, van't Hoff, W, Russell-Eggitt, I, Dobbie, A, Sheridan, E, Kleta, R, & Bockenhauer, D. (2011). Altered electroretinograms in patients with KCNJ10 mutations and EAST syndrome. The Journal of Physiology, 589(Pt 7), 1681–1689. https://doi.org/10.1113/jphysiol.2010.198531</bibtext> </blist> <blist> <bibtext> Tiwari R, Kumar R, Malik S, Raj T, Kumar P. Analysis of Heart Rate Variability and Implication of different factors on Heart Rate Variability. Current Cardiology Reviews. 2021; 17; 5: e160721189770. 10.2174/1573403X16999201231203854. 33390146. 8950456</bibtext> </blist> <blist> <bibtext> Tsang, V, & Chu, P. C. K. (2018). Comparing Eye-tracking data of children with high-functioning ASD, Comorbid ADHD, and of a control watching social videos. Journal of Visualized Experiments: JoVE, 142. https://doi.org/10.3791/58694.</bibtext> </blist> <blist> <bibtext> Wachtmeister L. Further studies of the chemical sensitivity of the oscillatory potentials of the electroretinogram (ERG) I. GABA- and glycine antagonists. Acta Ophthalmologica. 1980; 58; 5: 712-725. 10.1111/j.1755-3768.1980.tb06684.x. 7211260</bibtext> </blist> <blist> <bibtext> Wang H, Siu K, Ju K, Chon KH. A high Resolution Approach to estimating time-frequency Spectra and their amplitudes. Annals of Biomedical Engineering. 2006; 34; 2: 326-338. 10.1007/s10439-005-9035-y. 16463086</bibtext> </blist> <blist> <bibtext> Wilens TE, Spencer TJ. Understanding attention-deficit/hyperactivity disorder from childhood to adulthood. Postgraduate Medicine. 2010; 122; 5: 97-109. 10.3810/pgm.2010.09.2206. 20861593. 3724232</bibtext> </blist> <blist> <bibtext> Youden, W. J. (1950). Index for rating diagnostic tests. Cancer, 3(1), 32–35. https://pubmed.ncbi.nlm.nih.gov/15405679/.</bibtext> </blist> <blist> <bibtext> Young S, Hollingdale J, Absoud M, Bolton P, Branney P, Colley W, Craze E, Dave M, Deeley Q, Farrag E, Gudjonsson G, Hill P, Liang H, Murphy C, Mackintosh P, Murin M, O'Regan F, Ougrin D, Rios P, Woodhouse E. Guidance for identification and treatment of individuals with attention deficit/hyperactivity disorder and autism spectrum disorder based upon expert consensus. BMC Medicine. 2020; 18; 1: 146. 10.1186/s12916-020-01585-y. 32448170. 7247165</bibtext> </blist> <blist> <bibtext> Youssef P, Nath S, Chaimowitz GA, Prat SS. Electroretinography in psychiatry: A systematic literature review. European Psychiatry: The Journal of the Association of European Psychiatrists. 2019; 62: 97-106. 10.1016/j.eurpsy.2019.09.006. 31553929</bibtext> </blist> <blist> <bibtext> Zeidan J, Fombonne E, Scorah J, Ibrahim A, Durkin MS, Saxena S, Yusuf A, Shih A, Elsabbagh M. Global prevalence of autism: A systematic review update. Autism Research: Official Journal of the International Society for Autism Research. 2022; 15; 5: 778-790. 10.1002/aur.2696. 35238171</bibtext> </blist> </ref> <aug> <p>By Sultan Mohammad Manjur; Luis Roberto Mercado Diaz; Irene O Lee; David H Skuse; Dorothy A. Thompson; Fernando Marmolejos-Ramos; Paul A. Constable and Hugo F. Posada-Quintero</p> <p>Reported by Author; Author; Author; Author; Author; Author; Author; Author</p> </aug> <nolink nlid="nl1" bibid="bib97" firstref="ref1"></nolink> <nolink nlid="nl2" bibid="bib85" firstref="ref2"></nolink> <nolink nlid="nl3" bibid="bib45" firstref="ref3"></nolink> <nolink nlid="nl4" bibid="bib93" firstref="ref4"></nolink> <nolink nlid="nl5" bibid="bib95" firstref="ref5"></nolink> <nolink nlid="nl6" bibid="bib41" firstref="ref6"></nolink> <nolink nlid="nl7" bibid="bib32" firstref="ref7"></nolink> <nolink nlid="nl8" bibid="bib43" firstref="ref8"></nolink> <nolink nlid="nl9" bibid="bib30" firstref="ref9"></nolink> <nolink nlid="nl10" bibid="bib59" firstref="ref11"></nolink> <nolink nlid="nl11" bibid="bib75" firstref="ref12"></nolink> <nolink nlid="nl12" bibid="bib96" firstref="ref13"></nolink> <nolink nlid="nl13" bibid="bib11" firstref="ref15"></nolink> <nolink nlid="nl14" bibid="bib26" firstref="ref16"></nolink> <nolink nlid="nl15" bibid="bib10" firstref="ref17"></nolink> <nolink nlid="nl16" bibid="bib79" firstref="ref18"></nolink> <nolink nlid="nl17" bibid="bib69" firstref="ref20"></nolink> <nolink nlid="nl18" bibid="bib74" firstref="ref21"></nolink> <nolink nlid="nl19" bibid="bib89" firstref="ref22"></nolink> <nolink nlid="nl20" bibid="bib55" firstref="ref23"></nolink> <nolink nlid="nl21" bibid="bib48" firstref="ref25"></nolink> <nolink nlid="nl22" bibid="bib72" firstref="ref27"></nolink> <nolink nlid="nl23" bibid="bib87" firstref="ref28"></nolink> <nolink nlid="nl24" bibid="bib54" firstref="ref29"></nolink> <nolink nlid="nl25" bibid="bib18" firstref="ref30"></nolink> <nolink nlid="nl26" bibid="bib21" firstref="ref31"></nolink> <nolink nlid="nl27" bibid="bib19" firstref="ref32"></nolink> <nolink nlid="nl28" bibid="bib40" firstref="ref33"></nolink> <nolink nlid="nl29" bibid="bib44" firstref="ref34"></nolink> <nolink nlid="nl30" bibid="bib66" firstref="ref36"></nolink> <nolink nlid="nl31" bibid="bib24" firstref="ref38"></nolink> <nolink nlid="nl32" bibid="bib33" firstref="ref39"></nolink> <nolink nlid="nl33" bibid="bib42" firstref="ref40"></nolink> <nolink nlid="nl34" bibid="bib20" firstref="ref47"></nolink> <nolink nlid="nl35" bibid="bib61" firstref="ref48"></nolink> <nolink nlid="nl36" bibid="bib82" firstref="ref49"></nolink> <nolink nlid="nl37" bibid="bib34" firstref="ref50"></nolink> <nolink nlid="nl38" bibid="bib35" firstref="ref51"></nolink> <nolink nlid="nl39" bibid="bib36" firstref="ref52"></nolink> <nolink nlid="nl40" bibid="bib37" firstref="ref53"></nolink> <nolink nlid="nl41" bibid="bib92" firstref="ref59"></nolink> <nolink nlid="nl42" bibid="bib60" firstref="ref60"></nolink> <nolink nlid="nl43" bibid="bib73" firstref="ref61"></nolink> <nolink nlid="nl44" bibid="bib58" firstref="ref66"></nolink> <nolink nlid="nl45" bibid="bib57" firstref="ref67"></nolink> <nolink nlid="nl46" bibid="bib80" firstref="ref68"></nolink> <nolink nlid="nl47" bibid="bib77" firstref="ref69"></nolink> <nolink nlid="nl48" bibid="bib22" firstref="ref70"></nolink> <nolink nlid="nl49" bibid="bib84" firstref="ref71"></nolink> <nolink nlid="nl50" bibid="bib17" firstref="ref72"></nolink> <nolink nlid="nl51" bibid="bib29" firstref="ref73"></nolink> <nolink nlid="nl52" bibid="bib65" firstref="ref75"></nolink> <nolink nlid="nl53" bibid="bib13" firstref="ref76"></nolink> <nolink nlid="nl54" bibid="bib51" firstref="ref77"></nolink> <nolink nlid="nl55" bibid="bib38" firstref="ref79"></nolink> <nolink nlid="nl56" bibid="bib12" firstref="ref84"></nolink> <nolink nlid="nl57" bibid="bib16" firstref="ref85"></nolink> <nolink nlid="nl58" bibid="bib64" firstref="ref87"></nolink> <nolink nlid="nl59" bibid="bib31" firstref="ref89"></nolink> <nolink nlid="nl60" bibid="bib47" firstref="ref90"></nolink> <nolink nlid="nl61" bibid="bib53" firstref="ref91"></nolink> <nolink nlid="nl62" bibid="bib90" firstref="ref93"></nolink> <nolink nlid="nl63" bibid="bib46" firstref="ref94"></nolink> <nolink nlid="nl64" bibid="bib68" firstref="ref95"></nolink> <nolink nlid="nl65" bibid="bib15" firstref="ref97"></nolink> <nolink nlid="nl66" bibid="bib83" firstref="ref101"></nolink> <nolink nlid="nl67" bibid="bib78" firstref="ref110"></nolink> <nolink nlid="nl68" bibid="bib25" firstref="ref112"></nolink> <nolink nlid="nl69" bibid="bib56" firstref="ref113"></nolink> <nolink nlid="nl70" bibid="bib39" firstref="ref114"></nolink> <nolink nlid="nl71" bibid="bib76" firstref="ref116"></nolink> <nolink nlid="nl72" bibid="bib88" firstref="ref128"></nolink> <nolink nlid="nl73" bibid="bib63" firstref="ref151"></nolink> <nolink nlid="nl74" bibid="bib81" firstref="ref152"></nolink> <nolink nlid="nl75" bibid="bib50" firstref="ref153"></nolink> <nolink nlid="nl76" bibid="bib28" firstref="ref154"></nolink> <nolink nlid="nl77" bibid="bib94" firstref="ref155"></nolink> <nolink nlid="nl78" bibid="bib14" firstref="ref156"></nolink> <nolink nlid="nl79" bibid="bib52" firstref="ref157"></nolink> <nolink nlid="nl80" bibid="bib23" firstref="ref162"></nolink> <nolink nlid="nl81" bibid="bib62" firstref="ref163"></nolink> <nolink nlid="nl82" bibid="bib67" firstref="ref164"></nolink> <nolink nlid="nl83" bibid="bib70" firstref="ref165"></nolink> <nolink nlid="nl84" bibid="bib71" firstref="ref166"></nolink> <nolink nlid="nl85" bibid="bib91" firstref="ref168"></nolink> <nolink nlid="nl86" bibid="bib86" firstref="ref170"></nolink> <nolink nlid="nl87" bibid="bib27" firstref="ref174"></nolink> <nolink nlid="nl88" bibid="bib49" firstref="ref182"></nolink>
Header DbId: eric
DbLabel: ERIC
An: EJ1464139
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Detecting Autism Spectrum Disorder and Attention Deficit Hyperactivity Disorder Using Multimodal Time-Frequency Analysis with Machine Learning Using the Electroretinogram from Two Flash Strengths
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Sultan+Mohammad+Manjur%22">Sultan Mohammad Manjur</searchLink><br /><searchLink fieldCode="AR" term="%22Luis+Roberto+Mercado+Diaz%22">Luis Roberto Mercado Diaz</searchLink><br /><searchLink fieldCode="AR" term="%22Irene+O%2E+Lee%22">Irene O. Lee</searchLink><br /><searchLink fieldCode="AR" term="%22David+H%2E+Skuse%22">David H. Skuse</searchLink><br /><searchLink fieldCode="AR" term="%22Dorothy+A%2E+Thompson%22">Dorothy A. Thompson</searchLink><br /><searchLink fieldCode="AR" term="%22Fernando+Marmolejos-Ramos%22">Fernando Marmolejos-Ramos</searchLink><br /><searchLink fieldCode="AR" term="%22Paul+A%2E+Constable%22">Paul A. Constable</searchLink><br /><searchLink fieldCode="AR" term="%22Hugo+F%2E+Posada-Quintero%22">Hugo F. Posada-Quintero</searchLink> (ORCID <externalLink term="http://orcid.org/0000-0003-4514-4772">0000-0003-4514-4772</externalLink>)
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="SO" term="%22Journal+of+Autism+and+Developmental+Disorders%22"><i>Journal of Autism and Developmental Disorders</i></searchLink>. 2025 55(4):1365-1378.
– Name: Avail
  Label: Availability
  Group: Avail
  Data: Springer. Available from: Springer Nature. One New York Plaza, Suite 4600, New York, NY 10004. Tel: 800-777-4643; Tel: 212-460-1500; Fax: 212-460-1700; e-mail: customerservice@springernature.com; Web site: https://link.springer.com/
– Name: PeerReviewed
  Label: Peer Reviewed
  Group: SrcInfo
  Data: Y
– Name: Pages
  Label: Page Count
  Group: Src
  Data: 14
– Name: DatePubCY
  Label: Publication Date
  Group: Date
  Data: 2025
– Name: TypeDocument
  Label: Document Type
  Group: TypDoc
  Data: Journal Articles<br />Reports - Research
– Name: Subject
  Label: Descriptors
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Autism+Spectrum+Disorders%22">Autism Spectrum Disorders</searchLink><br /><searchLink fieldCode="DE" term="%22Attention+Deficit+Hyperactivity+Disorder%22">Attention Deficit Hyperactivity Disorder</searchLink><br /><searchLink fieldCode="DE" term="%22Symptoms+%28Individual+Disorders%29%22">Symptoms (Individual Disorders)</searchLink><br /><searchLink fieldCode="DE" term="%22Clinical+Diagnosis%22">Clinical Diagnosis</searchLink><br /><searchLink fieldCode="DE" term="%22Measurement+Equipment%22">Measurement Equipment</searchLink><br /><searchLink fieldCode="DE" term="%22Physiology%22">Physiology</searchLink><br /><searchLink fieldCode="DE" term="%22Screening+Tests%22">Screening Tests</searchLink>
– Name: DOI
  Label: DOI
  Group: ID
  Data: 10.1007/s10803-024-06290-w
– Name: ISSN
  Label: ISSN
  Group: ISSN
  Data: 0162-3257<br />1573-3432
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: Purpose: Autism spectrum disorder (ASD) and attention deficit hyperactivity disorder (ADHD) are conditions that similarly alter cognitive functioning ability and challenge the social interaction, attention, and communication skills of affected individuals. Yet these are distinct neurological conditions that can exhibit diverse characteristics which require different management strategies. It is desirable to develop tools to assist with early distinction so that appropriate early interventions and support may be tailored to an individual's specific requirements. The current diagnostic procedures for ASD and ADHD require a multidisciplinary approach and can be lengthy. This study investigated the potential of electroretinogram (ERG), an eye test measuring retinal responses to light, for rapid screening of ASD and ADHD. Methods: Previous studies identified differences in ERG amplitude between ASD and ADHD, but this study explored time-frequency analysis (TFS) to capture dynamic changes in the signal. ERG data from 286 subjects (146 control, 94 ASD, 46 ADHD) was analyzed using two TFS techniques. Results: Key features were selected, and machine learning models were trained to classify individuals based on their ERG response. The best model achieved 70% overall accuracy in distinguishing control, ASD, and ADHD groups. Conclusion: The ERG to the stronger flash strength provided better separation and the high frequency dynamics (80-300 Hz) were more informative features than lower frequency components. To further improve classification a greater number of different flash strengths may be required along with a discrimination comparison to participants who meet both ASD and ADHD classifications and carry both diagnoses.
– Name: AbstractInfo
  Label: Abstractor
  Group: Ab
  Data: As Provided
– Name: DateEntry
  Label: Entry Date
  Group: Date
  Data: 2025
– Name: AN
  Label: Accession Number
  Group: ID
  Data: EJ1464139
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=eric&AN=EJ1464139
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1007/s10803-024-06290-w
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 14
        StartPage: 1365
    Subjects:
      – SubjectFull: Autism Spectrum Disorders
        Type: general
      – SubjectFull: Attention Deficit Hyperactivity Disorder
        Type: general
      – SubjectFull: Symptoms (Individual Disorders)
        Type: general
      – SubjectFull: Clinical Diagnosis
        Type: general
      – SubjectFull: Measurement Equipment
        Type: general
      – SubjectFull: Physiology
        Type: general
      – SubjectFull: Screening Tests
        Type: general
    Titles:
      – TitleFull: Detecting Autism Spectrum Disorder and Attention Deficit Hyperactivity Disorder Using Multimodal Time-Frequency Analysis with Machine Learning Using the Electroretinogram from Two Flash Strengths
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Sultan Mohammad Manjur
      – PersonEntity:
          Name:
            NameFull: Luis Roberto Mercado Diaz
      – PersonEntity:
          Name:
            NameFull: Irene O. Lee
      – PersonEntity:
          Name:
            NameFull: David H. Skuse
      – PersonEntity:
          Name:
            NameFull: Dorothy A. Thompson
      – PersonEntity:
          Name:
            NameFull: Fernando Marmolejos-Ramos
      – PersonEntity:
          Name:
            NameFull: Paul A. Constable
      – PersonEntity:
          Name:
            NameFull: Hugo F. Posada-Quintero
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 04
              Type: published
              Y: 2025
          Identifiers:
            – Type: issn-print
              Value: 0162-3257
            – Type: issn-electronic
              Value: 1573-3432
          Numbering:
            – Type: volume
              Value: 55
            – Type: issue
              Value: 4
          Titles:
            – TitleFull: Journal of Autism and Developmental Disorders
              Type: main
ResultId 1