An Architecture of a Web Application for Deploying Machine Learning Models in Healthcare Domain

An important gap exists between advanced Deep Learning (DL) models developed for medical imaging and their insitu implementation in clinical environments. Our research proposes a scalable software architecture for a web-based computer-aided diagnosis (CAD) application that bridges the gap between research and production. The main goals are: (1) creating an intuitive interface for non-technical users and, (2) designating a feedback mechanism to enhance the Machine Learning (ML) models with medical expertise. The architecture assumes a microservices approach, containerization, and current web technologies to enable seamless deployment and user-friendliness. Key components include ONNX-based model conversion for cross- platform compatibility, asynchronous processing based on FastAPI, and a modular data handling system using PostgreSQL. Our proof-of-concept (PoC) shows the system's feasibility by deploying it on the Google Cloud Platform (GCP) and getting initial feedback from clinicians. Early findings indicate potential improvements in model usability and stakeholder engagement. Future actions will involve implementing continuous integration and delivery pipelines to improve system resilience and scalability.