EXPLAINABLE NEURAL NETWORKS: TRANSPARENCY AND TRUST IN MEDICAL DIAGNOSIS WITH RADIOLOGICAL IMAGES: A SYSTEMATIC REVIEW

Hans Paiva Sanchez M.; Eduardo Aldair M.C. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2025) · DOI: 10.18687/laccei2025.1.1.1355

Resumen

Advances in Explainable Artificial Intelligence (XAI) have reshaped the landscape of medical diagnosis by addressing challenges related to the interpretability and trust of deep learning models. This study conducts a Systematic Literature Review (SLR), following PRISMA guidelines and the PICO strategy, to examine how XAI techniques have been applied to the analysis of radiological images, including X-rays, computed tomography (CT), and magnetic resonance imaging (MRI). The review involved searches across Scopus, Redalyc, and SciELO databases, initially identifying 230 articles, of which 40 met the inclusion criteria. The analysis revealed a predominance of post-hoc interpretability techniques such as SHAP and LIME, as well as model-inherent approaches including neural networks with attention mechanisms. These methods have demonstrated significant improvements in diagnostic precision and have enhanced clinical interpretability. However, the review also identified critical gaps in the standardization of evaluation metrics and the usability of generated explanations in clinical practice. This work provides key recommendations to guide future research toward the development of explainability tools tailored to the needs of healthcare professionals, fostering greater trust in AI-supported medical diagnosis.

Instituciones: Universidad Tecnológica del Perú