Computer Vision News - May 2018

CaPTk: Cancer Imaging Phenomics Toolkit by Spyridon Bakas & Sarthak Pati 40 Saturday ISBI DAILY This half-day hands-on tutorial on the Cancer Imaging Phenomics Toolkit (CaPTk) was held during the first day of ISBI 2018 and it showcased the use and expansion mechanism of this newly developed, stand-alone, dynamically growing software platform for radiographic cancer image analysis, currently focused on brain, breast, and lung cancer. CaPTk describes a team effort, is available via a very permissive open source license , and incorporates algorithms enabling quantitative biomedical image analysis. The first part of the tutorial , given by the scientific coordinator of CaPTk ( Spyridon Bakas ) , focused on the use of the diverse algorithms included in the platform, spanning across i) essential pre-processing algorithms (denoising, normalization, etc.), ii) general-purpose and specialized segmentation methods, iii) extraction of imaging features from multi-modal radiographic scans, as well as iv) various specialized applications (i.e., risk estimation for breast cancer, brain tumor recurrence prediction, and an imaging signature of EGFR mutations in glioblastoma). The use of all these applications, which describe contributions of numerous research studies (cited within the documentation of the respective applications), was demonstrated via both the command line interface for batch processing, and the graphical user interface highlighting their potential for use by clinical researchers. The second part of the tutorial , given by the lead developer of CaPTk ( Sarthak Pati ), highlighted different ways that computational scientists can integrate their methods (written in MATLAB, Python, or C++) into the CaPTk platform, while leveraging its existing user interface and infrastructure. The ease to incorporate a developer’s algorithms to CaPTk was highlighted, specially through the predictive algorithms integrated into the platform by Saima Rathore . Examples of currently integrated applications (through code snippets) were also provided to the participants, facilitating potentially easier integration. The primary goal of this part was to get the audience comfortable with navigating through the source code of the platform and provide a starting point for anyone interested in distributing their package to a more non-computational (clinical) audience. Tutorial Spyridon Bakas Sarthak Pati BEST OF