Exploiting GPUs on distributed infrastructures for medical imaging applications with VIP and DIRAC

GPU usage has become mandatory for the processing of (3D) medical data, as well as for efficient machine learning approaches such as deep learning. In this contribution, we present how VIP and DIRAC can be leveraged to run medical image processing applications on distributed computing resources equipped with GPUs.VIP (Virtual Imaging Platform) is a web portal for the simulation and processing of massive data in medical imaging. VIP users can access applications as a service and significant amounts of computing resources and storage with no required technical skills beyond the use of a web browser. VIP relies on the DIRAC (Distributed Infrastructure with Remote Agent Control) interware for scheduling tasks for execution on distributed infrastructures such as grid, clouds, and local clusters. New applications are regularly integrated into VIP/DIRAC. They all have their own requirements, among which GPU usage is more and more frequent.This contribution will give an overview of the targeted medical applications and their requirements, as well as technical insights on how VIP and DIRAC allow end users to efficiently exploit GPU resources with no specific knowledge about the underlying distributed infrastructure.