Implementación de un microservicio escalable para el análisis automatizado de cáncer cerebral en imágenes de resonancia magnética mediante arquitectura de contenedores.

Abstract

This study proposes and implements a microservices architecture for deploying pre-trained brain tumor segmentation models in medical imaging. While numerous studies focus on the algorithmic performance of deep learning models, a gap remains between their experimental validation and their integration into technically deployable systems. In this context, a microservice was developed to integrate pre-trained artificial intelligence models into the healthcare sector, enabling their execution across different hardware environments. The study focused on evaluating the system’s computational performance, considering metrics such as inference latency, memory consumption, and GPU utilization under different configurations. The results demonstrate a significant reliance on hardware acceleration for models that process three-dimensional images, as well as variations in response times depending on the execution environment. The proposal helps to bridge the gap between experimental models and deployable systems by validating the technical feasibility of integrating them into a modular, decoupled, and potentially scalable architecture. Direct clinical validation was not addressed, limiting the scope to the computational and structural evaluation of the system.