Enhancing Diagnostic Radiology with Explainable Deep Learning for Multi-Disease Detection

Abstract

The integration of deep learning (DL) and explainable artificial intelligence (XAI) is transforming diagnostic radiology by improving both the accuracy and interpretability of medical image analysis. This study developed three deep learning models a 5-block ResNet, an 8-block DenseNet, and a 5-block bottleneck architecture to detect and classify diseased tissue in medical images related to COVID-19, brain tumours, knee arthritis, and breast cancer. Using large datasets and implemented in MATLAB, the models were rigorously trained and validated to ensure high performance and generalizability. A major focus of this work was the incorporation of XAI techniques to make the models’ decision-making processes transparent and clinically intelligible. By visualizing the regions influencing diagnoses, our approach helps build trust and facilitates collaboration between AI systems and medical practitioners. The results show that the combined DL XAI framework not only achieves high diagnostic accuracy but also provides actionable insights that align with clinical reasoning. This work supports the adoption of AI-assisted diagnostics for earlier and more reliable detection of diseases, ultimately contributing to improved patient care.