An Integrated Three-Tier Microservice Architecture for Deploying State-of-the-Art Ocular Cancer Detection Models

This paper details the design and implementation of a scalable three-tier microservice system architecture optimizedfor the deployment of deep learning models in real-time ocular cancer screening. The platform decouples the presentation layer (React), the data and routing layer (Node.js), and the specialized machine learning inference service (FastAPI). The system supports a robust multi-stage diagnostic workflow, currently implemented using a ResNet50 - A based transfer learning model trained across two phases to achieve empirically verified perfect classification on the test set. The model, which targets Retinoblastoma (RB) [7], achieved An accuracy, Precision, and Recall of 100% in its independent test set. This microservice architecture has been demonstrated to be critical for the effective and scalable integration of high- performance AI tools into practical clinical environments, especiallycially low-resource settings, by providing robust data persistence, dedicated user management, built-in system traceability, and seamless diagnostic request routing. Index Terms—Microservices, Deep Learning Deployment, Retinoblastoma, ResNet50, Mongoose, FastAPI, FullStack, Scal- ability, Cancer Detection, Transfer Learning.