Exploring XNAT to Foster Development and Testing of Image Processing Methods for Clinical Settings: Preliminary Results

Resumo

Medical imaging has evolved greatly in the past two decades and provides a rich set of data on the anatomy and function of important organs such as the heart. In this regard, medical image processing has also evolved to take advantage of these data and provide clinicians with methods supporting, for instance, (semi-)automatic identification of regions of interest and objective measures that can inform diagnosis. Recently, methods have also been proposed to perform haemodynamic simulations based on anatomic structures extracted from this data, which consubstantiate a noninvasive alternative to obtain important descriptors, such as the fractional flow
reserve (FFR). The development of such methods is best served by working closely with clinicians, and, after a first validation, if these methods can be experimented with in a wide range of cases, they can more closely inform their evolution and an assessment of their performance. Clinicians often have their diagnosis workflow supported on dedicated workstations, often proprietary, that not only integrate the processing and analysis of the imaging data but also provide visualization features that allow, for instance, visualizing the data from specific angles so as to conform with standard practice. Therefore, when developing new methods, it is important to comply with these basic features to promote easier use and evaluation. However, integrating novel methods in existing workstations is mostly out of question and developing a new framework from scratch to support all the workflow is a major effort. In this work, we argue for a novel approach to supporting the research workflow in image processing and analysis methos and explored the XNAT framework, designed to support clinical studies pipelines, to assess how it can be used to deploy these methods, particularly by its integration with image viewers such as the one provided by the Open Health Image Foundation (OHIF). After its deployment in a virtualized environment and exploration of its main features, a proof-of-concept pipeline developed in Python was successfully integrated to smooth the images and return the processing results to the framework. Overall, mastering the integration with XNAT is complex at
first, but the framework provided a promising response to the intended use.