Dicomflex: A novel framework for efficient deployment of image analysis tools in radiological research.

Abstract

OBJECTIVE:Medical image processing tools in research are often developed from scratch without the use of predefined software structures, which potentially makes them less reliable and difficult to maintain. The objective here was to present and evaluate a novel framework (Dicomflex) for the deployment of tools with a uniform workflow, commonly encountered in medical image analysis. MATERIALS AND METHODS:The object-oriented code was developed using Matlab. Dicomflex applications follow the common workflow of image-slice selection, user interaction, image processing, result visualization and progression to next slice. The framework consists of three important classes that host functionality, two configuration files and a front end that displays images, graphs and resulting data. RESULTS:So far, three different research tools have been created under the new framework. In comparison with previous Matlab analysis tools used at our institution, users of Dicomflex tools subjectively considered the learning phase to be shorter and handling to be simpler and more intuitive. They also highlighted the benefit and comfort of the standardized interface and predefined workflow. The framework-inherent handling of software versions was considered highly beneficial for maintenance as well as data and software management at different project stages. The clear separation of framework-related and unrelated code allows for a fast and more direct design of new tools in well-defined steps. The flexibility of the framework translates to a wide range of image processing tasks, such as segmentation, region-of-interest (ROI) analyses or computation of functional parameter maps, but is limited to 2D datasets. CONCLUSION:Potential medical applications include the assessment of cardiac performance, detection of cerebrovascular disease or characterization of cancerous lesions. Dicomflex tools share a similar workflow and host the pertinent functions only. This may be relevant for many image processing needs in radiological research, where quick software deployment and reliability of results is essential.