Communication Dans Un Congrès Année : 2025

Addressing Explainability, Transparency and Interpretability Requirements of AI Models for ECG Analysis

Résumé

Increased application of artificial intelligence (AI), particularly deep neural networks, in electrocardiogram (ECG) interpretation has highlighted critical concerns about the opacity of algorithmic decisions. The demand for explainability, interpretability, and transparency in medical AI has driven the development of explainable AI (XAI) techniques. This review aims to provide a comprehensive evaluation of the most relevant and promising XAI methods for ECG analysis, highlighting their methodological diversity and identifying the main challenges that must be addressed for clinical translation. Despite rapid advancements in XAI for ECG interpretation, significant gaps remain in clinical validation, personalization, and ethical governance. The findings underscore the need to integrate explainability with robust performance, transparency, and clinician-in-the-loop approaches. This review serves as a reference for future work that aims to bridge the gap between experimental research and clinically viable AI systems, ultimately promoting safer and more accountable deployment of AI in cardiology.

Fichier principal
Vignette du fichier
387_Preprint.pdf (127.21 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05263972 , version 1 (16-09-2025)

Licence

Identifiants

  • HAL Id : hal-05263972 , version 1

Citer

Durande Kamga Nguifo, Gabriel Tozatto Zago, Rafael Pereira, Rodrigo Varejão Andreão, Anthony Fleury. Addressing Explainability, Transparency and Interpretability Requirements of AI Models for ECG Analysis. 52nd International Computing in Cardiology conference (CinC 2025), Sep 2025, Sao Paulo, Brazil. ⟨hal-05263972⟩
107 Consultations
148 Téléchargements

Partager

  • More