(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: test feasibility of semi-automatic contouring for prostate cancer patients based on random forest classifier and active contour algorithm implemented in open source application called ITK-SNAP.
Methods: the advent of MRI linear accelerators systems and rapid improvement of onboard imaging systems in radiotherapy clinics, daily adaptation of radiotherapy plan is becoming reality. Technological advancement has enabled inter-institutional and inter-disciplinary studies as well. Crucial to the success of the above goals is reproducible and quick contouring to minimize user bias. Automation of contouring and dose in real time has been an ongoing effort by companies and academic institutions with little success on this front. In this study, we have shown the feasibility of semi-automatic contouring based on Random forest based classifier and active snakes algorithm implemented in opensource software application ITK-SNAP. In ITK-SNAP, random forest based classifier is trained based on manual definition of tissue of interest to generate foreground class contour for organ of interest and rest of the surrounding tissue as background class. In the second stage user initializes (seed point) active contour inside the region of interest and subsequently contour evolves filling the entire three-dimensional ROI.
Results: generated bladder and rectum contours for ten randomly picked prostate cancer patients were very comparable to clinically drawn contours for radiotherapy planning. Average dice similarity coefficient for 80% of the contours generated were >0.93 with 20% of the cases needing minor editing to improve the dice coefficient.
Conclusion: study demonstrates that semi-automated 3D contouring is possible reliably enough to replace subjective and tedious manual contouring. Tools like ITK-SNAP are freely available to all and easy to learn. This enables for more robust contouring for inter-institutional and inter-disciplinary studies which in turn enable radiotherapy clinics to implement daily adaptation of radiotherapy plan and near real time dose tracking
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
