Skip to content
Open access

Empirical optimization of a hybrid explainable ai framework for transparent and clinically aligned deep learning-based decision support

Oct 2026 · Nature Journal of Emerging Sciences Technologies and Innovations · 0 citations

Abstract

Deep learning (DL) has shown great promise for clinical decision support, particularly for classifying medical images, but its limited transparency may limit clinicians’ understanding and trust. Alternative explainable AI (XAI) techniques include attention mechanisms, Grad-CAM, SHAP and LIME, all of which provide complementary explanations but are considered individually. The work proposes an empirical optimization approach to design a hybrid XAI system that improves the transparency and clinical consistency of deep learning-based clinical decision support. Explores alternative configurations of this framework in similar experimental settings, systematically, with intrinsic attention and post-hoc explanations like Grad-CAM, SHAP and LIME. Optimization takes into account predictive performance, explanation fidelity, stability, consistency, complexity, clarity, clinical alignment and computational efficiency. To test the candidate deep learning and XAI configurations, the study employs a pneumonia task in the classification of chest X-rays. The configurations are compared statistically and evaluated on several dimensions to identify whether there are practical, statistically significant, or numerical differences. This is intended to provide a framework for finding an optimal configuration of explanations that is predictive, reliable, understandable, clinically relevant and computationally tractable. Therefore, the study proposes an empirical perspective on hybrid XAI selection, besides the traditional method-by-method comparison.  

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.