Architecture and Infrastructure of an Intelligent Disease Diagnosis System Based on Medical Image Analysis

This paper presents the architectural and infrastructural solutions of an intelligent disease diagnosis system based on deep medical image analysis. A comparative analysis of CNN architectures (ResNet, DenseNet, EfficientNet) and transformer-based models (ViT, TransUNet) is conducted. Key mathematical formulations - EfficientNet compound scaling, ViT self-attention, Dice Loss, and Grad-CAM - are provided as formal equations. A microservice infrastructure with an XAI module is proposed. Validation on CheXPert and Synapse yielded AUC = 0.912 and Dice Score = 0.772; Grad-CAM expert agreement reached 84.5%.