Validation and evaluation of medical image processing procedures: Requirements for a Grid-enabled platform

A major problem in the validation of image processing software and procedures is testing them on an adequate number of clinical studies. The purpose of this work is the creation of software tools to enable a large database of images to be available to a research network for the purpose of validating and intercomparing a variety of different image processing procedures. Given the importance of the initiative to create Grid-enabled software, using for example the Globus toolkit as part of the next generation Internet, this aim here was to design a system for collecting and making available a large number of clinically validated studies for the purpose of software validation. Such a database must include, in addition to DICOM and other format images, the related clinical information, while respecting ethical and data protection regulations. initial work has been to develop a distributed database rather than a server to manage distributed applications. The users of the system being prototyped (MIAS-Grid) have access via a server to four linked databases containing 1) user access information, 2) anonymized patient details, 3) clinical information and 4)access information to the corresponding images. The prototype is in the process of being populated with liver, colon and lung CT/MR and US Studies. A related project IXI is being used notably for validation of brain analysis. This leads to the longer term aim of creating a Grid-enabled workstation for prototyping new procedures. The key target applications are non-rigid registration, segmentation and active appearance models. Such a database is essential for validation of both existing and novel image processing procedures and can only be created by a consortium. The key problem encountered is the entry of adequate validated clinical information which is essential in this clinical trial type of validation methodology. (c) 2005 CARS & Elsevier B.V. All rights reserved.