Program Information
MR-Based Atlas for Autosegmentation of Target and Normal Structure for MR-Only Planning in Prostate
N Tyagi1*, S Fontenla2 , A Apte3 , K Ostergren4 , C Tomer5 , M Hunt6 , M Zelefsky7 , A Fontenlla8 , (1) Memorial Sloan-Kettering Cancer Center, New York, NY, (2) Memorial Sloan-Kettering Cancer Centre, New York, NY, (3) Memorial Sloan-Kettering Cancer Center, New York, NY, (4) MIM Softwrae, Cleaveland, OH, (5) Memorial Sloan-Kettering Cancer Centre, New York, NY, (6) Memorial Sloan-Kettering Cancer Center, New York, NY, (7) Memorial Sloan Kettering Cancer Center, New York, New York, (8) Memorial Sloan-Kettering Cancer Center, New York, NY
Presentations
SU-H2-GePD-J(A)-6 (Sunday, July 30, 2017) 3:30 PM - 4:00 PM Room: Joint Imaging-Therapy ePoster Lounge - A
Purpose: To develop an MRI atlas based auto-segmentation of target and normal structures for MR-only planning of prostate cancer patients undergoing external beam radiotherapy
Methods: Twenty-five intact prostate cancer patients treated using an MR-only simulation and planning workflow at our institution were selected for this study. All patients were scanned with a sequence protocol consisting of small FOV T2w axial and large FOV 3D fast-field echo based mDIXON MRI. While T2w MRI was used for soft-tissue contouring, mDIXON MRI was used to generate synthetic CT using a commercial synthetic CT software called MRCAT (MR for Calculating Attenuation). Twenty patients were used to build three atlases in MIMTM using T2w MRI, mDIXON in-phase MRI and MRCAT synthetic-CT. Both MR images were bias-field corrected before adding to the atlas. Target and normal structures including prostate, bladder, rectum and penile bulb were segmented on T2w MRI by two expert physicians. Femurs were segmented on the mDIXON MRI and MRCAT synthetic CT. Five remaining patients were used to test atlas-based automatic segmentation using a free form intensity-based deformable registration algorithm. These structures were created using either a majority vote or STAPLE algorithm using a multi-pass atlas of closest matches. The atlas generated contours were analyzed with respect to the manual drawn contours using dice similarity coefficient (DSC).
Results: Femurs performed better on MRCAT synthetic CT as compared mDIXON in-phase MRI with DSC of 0.96±0.01. The DSC for bladder, rectum, penile bulb and prostate was 0.92±0.04, 0.8±0.06, 0.85±0.04 and 0.83±0.09 respectively. Femurs were generated using 5 subjects in the atlas whereas bladder, rectum, penile bulb and prostate were generated using all the subjects in the atlas.
Conclusion: MR atlas-based auto-segmentation gives reasonable accuracy for clinical use. Adding more patients to the atlas will further improve the atlas-based segmentation accuracy.
Contact Email: