Title: Deep Learning-Based Multi-Modal Medical Image Classification For Skin Lesion, Chest X-Ray, And Retinal Scan Detection In Low-Resource Healthcare Settings In Tanzania
Authors: Johnson George Mlelwa,Lawi Augustino Kihumbu,Alfa Edward Chengula,Carlos Ngolongolo,Catherine Peter Swai,Baraka Mwagala,Jeza Tunje
Volume: 10
Issue: 4
Pages: 140-146
Publication Date: 2026/04/28
Abstract:
Medical image interpretation remains one of the most critical yet challenging aspects of modern diagnostic medicine in Tanzania and sub-Saharan Africa. The increasing burden of non-communicable diseases has dramatically raised demand for accurate and timely analysis across three key imaging modalities: dermoscopic images for skin lesion classification, chest X-rays for pulmonary and cardiac pathology detection, and retinal fundus photographs for diabetic retinopathy grading. This demand occurs against a backdrop of severe specialist shortages, inadequate diagnostic infrastructure, and limited financial resources. This review paper analyzes the current state of deep learning-based medical image classification applicable to Tanzania's low-resource healthcare context, synthesizing literature on convolutional neural network architectures, transfer learning strategies, explainable AI techniques, and deployment frameworks. Drawing from global benchmark datasets (HAM10000, CheXpert, APTOS 2019), African healthcare studies, and international deep learning research, the paper identifies a critical gap: no unified, locally deployable, multi-modal solution with integrated explainability currently exists for Tanzanian or East African clinical environments. The proposed multi-modal deep learning system addresses these shortcomings by offering a web-based, explainable AI-enabled platform that classifies skin lesions, chest X-ray pathologies, and diabetic retinopathy simultaneously, generates Grad-CAM visual heatmaps for clinical transparency, and is deployable offline on standard hardware. This integrated approach promises enhanced diagnostic accuracy, reduced specialist workload, improved clinical trust through explainability, and alignment with Tanzania's digital health transformation goals.