Edge-Based Real-Time Medical Image Processing and Classification Using a Lightweight Explainable Deep Learning Framework

Abstract

The rapid advancement of Artificial Intelligence (AI) has significantly improved medical image processing applications in healthcare. Still, there are difficulties in deploying deep learning in the field due to high computational overhead, latency, and limited explainability. To address the stated challenges, the framework under discussion comprises two stages: image preprocessing and a lightweight CNN with depth-wise separable convolutions for effective feature extraction and classification. Grad-CAM and saliency map visualizations provide a visual explanation of the output. This framework is intended for edge deployment and may be used on platforms such as the Raspberry Pi 4, the Jetson Nano, and TinyML. Moreover, the design is aimed at ensuring real-time operation on resource-constrained edge devices while preserving the patients' data privacy. Experimental validation was performed on benchmark medical image datasets typically used in healthcare image classification research. Results indicate that the lightweight model demonstrates comparable classification quality (~90–91% accuracy) but has fewer parameters (2.1M parameters, 0.6 GFLOPs) and faster inference time (~95 ms). Moreover, statistical analysis demonstrates the model's reliability and consistency, while latency testing demonstrates its ability to support real-time image processing. At the same time, adding explainability options to the model does not affect its performance and provides valuable visual explanations for decisions. Overall, the proposed framework has demonstrated that the advantages of accuracy, efficiency, and explainability are combined, providing a promising alternative for real-time medical imaging.