Program Information

Automatic Segmentation of Multiple Pelvic Organs On a Web-Based Image Processing and Plan Evaluation Platform (WIPPEP)

X Chai*, L Liu , L Xing , Stanford Univ School of Medicine, Stanford, CA

Presentations

TU-F-BRF-1 Tuesday 4:30PM - 6:00PM Room: Ballroom F

Purpose:To present a novel automatic segmentation method for multiple pelvic organs and integrate into the web-based image processing and plan evaluation platform (WIPPEP) to provide easy access through a web browser and leverage the power of cloud for faster computation .

Methods:This segmentation method uses principal component analysis to model organ shapes and hierarchical clustering to model the appearance of each organ surface patch. Coarse and refined segmentation are done iteratively for each surface patch to find the best reference patch profile via a hierarchical cluster decision tree. Final segmentation is obtained using the generated global reference profile. The user chooses the CT image data to be segmented from the image server; the image volume is then visualized in the web browser and an HTTP POST request is sent to the computation server by clicking a button. When the computation server receives the segmentation request, it would then send an HTTP GET request to the image server to download the image volumes, call the segmentation program, and eventually send the segmentation results back to the web server. The segmentation results are transferred in JSON format and quickly passed to the canvas element in the web browser to display the contours.

Results:For 5 test patients, the average DICE overlaps for prostate, bladder and rectum were 0.75, 0.78 and 0.57, respectively. With the platform running on hospital intranet, it took 20 seconds to load CT image volume to the web-browser and 80 seconds to receive segmentation results from the computation server.

Conclusion:This system allows users to access image data and send segmentation requests from a web browser, perform segmentation on a computation server, and quickly visualize resulting segmentation contours on a web browser. The software prototype achieves auto-segmentation of prostate, bladder and rectum with moderate accuracy in a minute's timeframe.

Contact Email: