MLOps for Medical Imaging: A Cloud-Edge architecture for Experiment Tracking and Model Evaluation

Artificial intelligence (AI) is increasingly transforming the healthcare sector by enabling advanced automated decision-making. However, implementing Machine Learning (ML) systems in medical imaging applications presents significant challenges, such as ensuring consistent results across different environments, maintaining comprehensive records of experimental processes, and establishing reliable methods for model comparison and evaluation. These issues become more evident as medical AI shifts towards a hybrid Cloud-Edge Continuum (CEC), adding complexity to workload orchestration and performance monitoring across diverse infrastructures. To address these challenges, this work introduces a modular Edge Micro Data Centre (EMDC) architecture that supports a scalable experiment tracking framework designed for AI model deployments in healthcare. The approach extends MLflow with a domain-specific module for medical image segmentation, incorporating several clinically relevant performance metrics. The EMDC architecture enables distributed processing and performance monitoring within a hybrid CEC environment. By aligning MLOps principles with CEC infrastructure, the proposed system facilitates reliable and scalable AI integration into the healthcare sector.