Informatics for High-Throughput and Distributed Analysis of Medical Images

Abstract

Clinical research with medical imaging typically involves large-scale data analysis with interdependent software toolsets tied together in a processing workflow. Grid queuing software has become more advanced to meet the needs of different architectures, providing support for general grid engine access has become a difficult problem. Innovation is also constrained by the interoperability of workflows and available processing resources. In an effort to remove these constraints, we improve upon the Java Image Science Toolkit (JIST) by demonstrating: (1) a customizable cluster access controller that is designed for batch operations, (2) a centralized agent that predicts memory and run-time constraints given a module and its inputs in order to more efficiently schedule and process batch operations, and (3) a prototype of a high-throughput bundled resource imaging system for integrated medical imaging analysis which directly interfaces with the MRI scanner. We review the challenges and opportunities involved in integrating across computation resources in an efficient and intuitive manner.