Cargando...
Fecha
2025-06-07
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editorial
Universidad Nacional de Educación a Distancia (UNED). E.T.S. de Ingeniería Informática
Citas
Resumen
Este trabajo se centra en la aplicación de técnicas de Inteligencia Artificial Explicable (XAI, por sus siglas en inglés) para mejorar la comprensión, confianza y utilidad de los modelos predictivos empleados en la detección temprana del deterioro cognitivo leve (DCL). La explicabilidad es un componente clave para la adopción de sistemas de inteligencia artificial en contextos sensibles como el sanitario, donde no basta con obtener predicciones precisas, sino que resulta fundamental entender cómo y por qué se generan dichas predicciones. A partir de datos clínicos y neuropsicológicos recogidos en el marco del proyecto IDENTIA, se entrenaron modelos de aprendizaje automático y se analizaron sus decisiones mediante diferentes enfoques de XAI, incluyendo métodos de importancia de variables como SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations) y análisis contrafactual. Estas técnicas han permitido generar explicaciones tanto a nivel global (identificación de variables relevantes en el conjunto de datos) como a nivel local (justificación de predicciones individuales).
Además, se ha desarrollado una aplicación web interactiva que integra todas las técnicas XAI implementadas, permitiendo explorar las salidas del modelo de forma visual, ajustable y comprensible, favoreciendo la transparencia y la toma de decisiones. Los resultados muestran que las variables más influyentes para el modelo coinciden con indicadores clínicos validados, como el control atencional y la memoria visual, lo que refuerza la validez del enfoque. Las explicaciones generadas se mostraron coherentes, estables y potencialmente útiles para apoyar el diagnóstico clínico.
En conjunto, este estudio demuestra que la integración de técnicas XAI en entornos sanitarios puede aumentar significativamente la transparencia, la trazabilidad y la confianza en los sistemas de IA, facilitando su adopción responsable en la práctica médica y promoviendo una toma de decisiones informada y ética.
This work focuses on the application of Explainable Artificial Intelligence (XAI) techniques to improve the understanding, trust, and utility of predictive models used for the early detection of mild cognitive impairment (MCI). Explainability is a key component for the adoption of artificial-intelligence systems in sensitive contexts such as healthcare, where precise predictions alone are not sufficient; it is essential to understand how and why those predictions are generated. Based on clinical and neuropsychological data collected within the IDENTIA project, machine-learning models were trained and their decisions analysed using various XAI approaches, including feature-importance methods such as SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), and counterfactual analysis. These techniques made it possible to generate explanations at both a global level (identifying the most relevant variables in the dataset) and a local level (justifying individual predictions). In addition, an interactive web application was developed that integrates all the implemented XAI techniques, allowing users to explore the model outputs in a visual, adjustable, and comprehensible way, thus fostering transparency and decision-making. The results show that the most influential variables for the model coincide with validated clinical indicators, such as attentional control and visual memory, which reinforces the validity of the approach. The generated explanations proved to be coherent, stable, and potentially useful to support clinical diagnosis. Taken together, this study demonstrates that integrating XAI techniques into healthcare environments can significantly increase transparency, traceability, and trust in AI systems, thereby facilitating their responsible adoption in medical practice and promoting informed and ethical decision-making.
This work focuses on the application of Explainable Artificial Intelligence (XAI) techniques to improve the understanding, trust, and utility of predictive models used for the early detection of mild cognitive impairment (MCI). Explainability is a key component for the adoption of artificial-intelligence systems in sensitive contexts such as healthcare, where precise predictions alone are not sufficient; it is essential to understand how and why those predictions are generated. Based on clinical and neuropsychological data collected within the IDENTIA project, machine-learning models were trained and their decisions analysed using various XAI approaches, including feature-importance methods such as SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), and counterfactual analysis. These techniques made it possible to generate explanations at both a global level (identifying the most relevant variables in the dataset) and a local level (justifying individual predictions). In addition, an interactive web application was developed that integrates all the implemented XAI techniques, allowing users to explore the model outputs in a visual, adjustable, and comprehensible way, thus fostering transparency and decision-making. The results show that the most influential variables for the model coincide with validated clinical indicators, such as attentional control and visual memory, which reinforces the validity of the approach. The generated explanations proved to be coherent, stable, and potentially useful to support clinical diagnosis. Taken together, this study demonstrates that integrating XAI techniques into healthcare environments can significantly increase transparency, traceability, and trust in AI systems, thereby facilitating their responsible adoption in medical practice and promoting informed and ethical decision-making.
Descripción
Categorías UNESCO
Palabras clave
Inteligencia Artificial Explicable, Deterioro cognitivo, XAI, SHAP, LIME, explicabilidad, aprendizaje automático, diagnóstico clínico, Explainable Artificial Intelligence, cognitive impairment, XAI, SHAP, LIME, explainability, machine learning, clinical diagnosis
Citación
Clavero Sanz, Adrián. Trabajo fin de Máster: "Explicabilidad de modelospredictivos para deteccióntemprana de deterioro cognitivocon XAI". Universidad Nacional de Educación a Distancia (UNED), 2025
Centro
E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial