Explainable Deep Learning Models for Multi-Disease Medical Imaging: Integrating ResNet, DenseNet, and Bottleneck Architectures for Transparent Diagnostic Radiology

Abstract

The advent of deep learning (DL) and explainable artificial intelligence (XAI) has ushered in a transformative era in diagnostic radiology, offering advanced and interpretable solutions for disease detection and classification. This study presents the development and validation of three novel deep learning architectures. a five-block ResNet, an eight-block DenseNet, and a five-block bottleneck model specifically designed to process imaging datasets for conditions such as COVID-19, brain tumors, knee arthritis, and breast cancer. The models were trained and tested using MATLAB on extensive, disease-specific datasets to ensure robustness, accuracy, and generalizability. A key contribution of this research lies in integrating XAI methods with the proposed deep learning frameworks, thereby addressing the critical need for transparency and interpretability in AI-driven medical diagnostics. This integration enables clinicians to better understand and trust the reasoning behind the AI models’ predictions, effectively bridging the gap between computational intelligence and clinical decision-making. The experimental results demonstrate that combining XAI with DL not only enhances diagnostic accuracy and efficiency but also produces outputs consistent with clinical expectations. Moreover this work contributes to advancing diagnostic radiology by introducing interpretable, high-performing AI models capable of early and reliable disease detection, ultimately supporting improved patient outcomes and more informed treatment planning.