Click here to

Session: Therapy General ePoster Viewing [Return to Session]

Development of An Automated Region of Interest Selection Algorithm for Surface Guided Radiation Therapy of Breast Cancer

S Cui*, G Li, H Kuo, B Zhao, A Li, L Cervino, Memorial Sloan Kettering Cancer Center, New York, NY


PO-GePV-T-77 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: The goal was to investigate if we can automate the region of interest (ROI) definition for breast surface guided radiotherapy (SGRT) preparation by proposing and implementing an algorithm.

Methods: Fifty-two whole breast cancer patients previously treated with the SGRT were included as the training dataset. The clinical ROI originally defined by dosimetrists was retrieved from the SGRT system. The patient’s anatomical structures (body and ipsilateral breast) were retrieved from the planning system. Over the patient cohort, the average distances between the clinical ROI boundaries and the ipsilateral breast boundaries in the superior-inferior (S-I) direction were computed. This knowledge served as base rule to create automated ROIs. Next, the automated ROI was created by learning from our clinical protocol (1) sets the ROI boundaries in S-I direction by expanding ipsilateral breast contours, (2) sets the ROI boundaries in cross-sectional direction that stretches from ipsilateral body midline to the contralateral breast midline (approximately), and (3) crops the corner to avoid armpit. For evaluation, fifteen cases were randomly selected from the patient cohort to compare the clinical ROIs and automated ROIs. Specially, a blinded evaluation was performed by three AlignRT experts to rate the acceptability and the quality (1-10 scale). Additionally, the dice similarity coefficient (DSC) was calculated to compare the similarity between the clinical and automated ROIs.

Results: The blinded evaluation results showed that the average ROI acceptability was 13.7/15 (clinical) and 15/15 (automated). The average quality rating was 6.98 ± 2.26 (clinical) and 8.17 ± 1.23 (automated). These results indicated that the ROI acceptability, quality, and consistence can be improved by automation. The DSC result was 0.82 ± 0.04, suggesting automated ROIs had a large overlap with clinically accepted ROIs.

Conclusion: The automated ROI selection algorithm can potentially accelerate clinical workflow and reduce the ROI variation eliminating the inter-observer variability. 


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email