Paul K. Gerke - Building and Replacing On-prem Deep Learning Infrastructure for Medical Image Analysis

Abstract

In 2015, the Diagnostic Image Analysis Group (DIAG) of the Radiology Department, Radboudumc Nijmegen adopted deep learning for diagnostic image analysis research. This required building custom, on-prem computing infrastructure centered around NIVIDA GPUs. This on-prem solution is effective for experimentation but scaling and maintaining the system remains a challenge.

To address these challenges, we have pivoted away from the on-prem solution towards an in-house developed, large open-source platform called Grand Challenge. This change solves our specific scaling and maintenance issues, and also enhances the visibility of research output in the field in general.

During the presentation I showcase the specific setups and discuss different hardware and software problems that we encountered.

Biography

Professional software developer trying to cover the in the entire software and hardware stack “from wire to website”. Working professionally at the Diagnostic Image Analysis Group (Radboudumc Nijmegen) for 10 years after finishing my Master of Science in Artificial Intelligence at the Radboud University Nijmegen.

Spreker