Sistema de inteligencia artificial para identificar infección y cuantificar invasión parasitaria por Toxoplasma gondii en fibroblastos de ratón.

Saved in:

Bibliographic Details
Title:	Sistema de inteligencia artificial para identificar infección y cuantificar invasión parasitaria por Toxoplasma gondii en fibroblastos de ratón.
Alternate Title:	Artificial intelligence system for identifying infection and quantifying parasitic invasion by Toxoplasma gondii in mouse fibroblasts.
Authors:	González-Garay, Alejandro¹ pegasso100@gmail.com, Ortiz-Alegría, Luz Belinda², Cañedo-Solares, Irma², Cisneros-Tecuchillo, Montserrat², Luna-Pastén, Héctor², Vargas-Villavicencio, José Antonio², Valenzuela-Moreno, Luis Fernando², Farfán-Moreno, José Eduardo³, CedilloPeláez, Carlos², Caballero-Ortega, Heriberto²
Source:	Acta Pediatrica de Mexico. 2026 Supplement, Vol. 47, pS60-S61. 2p.
Subjects:	TOXOPLASMA gondii, FIBROBLASTS, MICROSCOPY, REGRESSION analysis, PARASITIC diseases, IMAGE recognition (Computer vision), ARTIFICIAL intelligence, ARTIFICIAL neural networks
Abstract (English):	BACKGROUND: Microscopic evaluation of T. gondii invasion usually requires manual inspection, which can be time-consuming, subjective, and poorly scalable. In this context, AI models represent an alternative for automating the identification of infected images and quantifying relevant findings. OBJECTIVE: To develop a cascaded AI model for evaluating parasitic invasion by T. gondii in mouse fibroblasts. METHODS: A two-step AI system was developed from microscopic images of mouse fibroblast cultures exposed to T. gondii (strain ME49-1 genotypes 1,8,28; strain ME49- 1 genotypes 1,344 y 345), observed between 0 to 24 hours, and linked to a CSV file containing reference variables. The dataset was expanded using six geometric transformations. In step 1, a cascaded classifier based on a dense neural network with cross-validation was developed, consisting of a sensitive screening phase followed by a specific confirmatory phase. Each image was represented using embeddings from four backbones and K-means cluster encoding. In step 2, regression models were implemented to estimate the number of cells, infected cells, vacuoles, parasites, and parasites per vacuole. In addition, an interpretability analysis using pseudo-Grad-CAM was performed, and an alternative network was trained to learn maps of cells, vacuoles, and parasites. RESULTS: From 905 original images, the dataset was expanded to 5,430 images, with an infection prevalence of 17.7%. The cascaded classifier achieved a sensitivity of 0.94, specificity of 0.89, precision of 0.66, an AUC-ROC of 0.725, and an F1-score of 0.77. In step 2, the regression models reported a macro MAE of 1.114. CONCLUSIONS: The system showed that a sequential strategy consisting of sensitive screening followed by specific confirmation improves specificity without an apparent loss of sensitivity. Step 2 has a functional architecture, although it still requires optimization. [ABSTRACT FROM AUTHOR]
Abstract (Spanish):	ANTECEDENTES: La evaluación microscópica de la invasión por T. gondii requiere varios evaluadores y es poco escalable, implicando tiempo prolongado de ejecución. En este contexto, los modelos de IA representan una alternativa para automatizar la identificación de células infectadas y cuantificar hallazgos relevantes a gran escala. OBJETIVO: Desarrollar un modelo de IA en cascada para evaluar la invasión parasitaria por T. gondii en fibroblastos de ratón. MÉTODOS: Se desarrolló un sistema de IA de 2 pasos a partir de imágenes microscópicas de cultivos de fibroblastos de ratón expuestos a T. gondii (genotipos #1 -cepa testigo-, #8, #28, #344 y #345) observados entre 0 y 24 horas, y vinculados a un archivo CSV con variables de referencia. Los datos se ampliaron mediante seis transformaciones geométricas. Paso 1, se desarrolló un clasificador en cascada basado en una red neuronal densa con validación cruzada, integrada por una fase de tamizaje sensible y otra confirmatoria especifica. Cada imagen se representó mediante embeddings de 4 backbones y codificación por clústeres K- means. Paso 2, se implementaron modelos de regresión para estimar número de células, células infectadas, vacuolas, parásitos y parásitos por vacuola. Además, se realizó un análisis de interpretabilidad con pseudo-Grad-CAM y se entrenó una red alterna para aprender mapas de células, vacuolas y parásitos. RESULTADOS: A partir de 905 imágenes originales, el conjunto se amplió a 5,430 imágenes, con una prevalencia de infección de 17.7%. El clasificador en cascada alcanzó una sensibilidad de 0.94, especificidad de 0.89, precisión de 0.66, AUC-ROC de 0.725 y F1-score de 0.77. En el paso 2, los modelos de regresión reportaron un MAE macro de 1.114. CONCLUSIONES: El sistema mostró que una estrategia secuencial de tamizaje sensible seguida de confirmación específica mejora la especificidad sin pérdida aparente de sensibilidad. El paso 2 cuenta con una arquitectura funcional, aunque aún requiere optimización. [ABSTRACT FROM AUTHOR]
	Copyright of Acta Pediatrica de Mexico is the property of Instituto Nacional de Pediatria (INP) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	MedicLatina

Description
Abstract:	BACKGROUND: Microscopic evaluation of T. gondii invasion usually requires manual inspection, which can be time-consuming, subjective, and poorly scalable. In this context, AI models represent an alternative for automating the identification of infected images and quantifying relevant findings. OBJECTIVE: To develop a cascaded AI model for evaluating parasitic invasion by T. gondii in mouse fibroblasts. METHODS: A two-step AI system was developed from microscopic images of mouse fibroblast cultures exposed to T. gondii (strain ME49-1 genotypes 1,8,28; strain ME49- 1 genotypes 1,344 y 345), observed between 0 to 24 hours, and linked to a CSV file containing reference variables. The dataset was expanded using six geometric transformations. In step 1, a cascaded classifier based on a dense neural network with cross-validation was developed, consisting of a sensitive screening phase followed by a specific confirmatory phase. Each image was represented using embeddings from four backbones and K-means cluster encoding. In step 2, regression models were implemented to estimate the number of cells, infected cells, vacuoles, parasites, and parasites per vacuole. In addition, an interpretability analysis using pseudo-Grad-CAM was performed, and an alternative network was trained to learn maps of cells, vacuoles, and parasites. RESULTS: From 905 original images, the dataset was expanded to 5,430 images, with an infection prevalence of 17.7%. The cascaded classifier achieved a sensitivity of 0.94, specificity of 0.89, precision of 0.66, an AUC-ROC of 0.725, and an F1-score of 0.77. In step 2, the regression models reported a macro MAE of 1.114. CONCLUSIONS: The system showed that a sequential strategy consisting of sensitive screening followed by specific confirmation improves specificity without an apparent loss of sensitivity. Step 2 has a functional architecture, although it still requires optimization. [ABSTRACT FROM AUTHOR]
ISSN:	01862391
DOI:	10.18233/apm.v47i1S.3441