Can Image Segmentation Evaluation Metrics Be Used for Assessing the Quality of SBRT Plans?
T Arsenault1*, A Amini1, A Baydoun1, B George2, L Bailey2, S Bhat2, R Kashani3, T Podder3, (1) University Hospitals Cleveland Medical Center, Cleveland, OH, (2) Case Western Reserve University, School Of Medicine,(3) Seidman Cancer Center /Uh Cleveland & CWRU, OH, Cleveland, OH
Presentations
MO-E115-IePD-F7-5 (Monday, 7/11/2022) 1:15 PM - 1:45 PM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 7
Purpose: To evaluate the efficacy of common image segmentation scoring metrics (ISSM), the Sorensen-Dice similarity coefficient and the Jaccard Similarity coefficient, in the assessment of quality of prescription isodose coverage of the planning target volume (PTV) for SRS/SBRT treatment plans.
Methods: A model was created in MATLAB 2020b to generate simplified target and prescription isodose line volumes with various simulated overlaps between the two volumes. The simulated volumes were used to calculate four commonly used conformity index (CI) metrics: RTOG, SALT, Lomax and Van’Riet/Paddick, along with two ISSMs: Dice and Jaccard similarity score. Four conventional CI metrics were compared with two segmentation metrics for the simulated volumes. Additionally all six parameters were calculated for 87 clinically used lung SBRT plans.
Results: With four simple, simulated target and isodose line volumes, we were able to demonstrate the potential of using ISSM to assess the quality of isodose line coverage of the target in SRS/SBRT plans. Both the ISSMs performed perfectly for gross misalignments with a score of 0. Dice and Jaccard provide slight differences in case of over coverage of prescription isodose with a score of 0.3363 and 0.2021, while Lomax and Van’Riet CI yielded a score of 0.2026. In 87 Lung SBRT cases, Dice and Jaccard score provide a readily available metric to analyze target coverage. Average CIs across the conventional CI indices were 1.12±0.18, 0.96±0.07, 0.88±0.08, and 0.84±0.10 for RTOG, SALT, Lomax and Van’Riet/Paddick, respectively. In comparison, the average CIs for Dice and Jaccard similarity were 0.91±0.07 and 0.85±0.10, respectively.
Conclusion: CI provides a quantitative metrics to facilitate the evaluation of treatment plan quality for SBRT. Through this work, we are able to demonstrate the potential of using commonly segmentation similarity metrics in place of manual calculation of CI, to evaluate the treatment plan for SRS/SBRT cases.
Keywords
Segmentation, Stereotactic Radiosurgery, Dosimetry
Taxonomy
TH- External Beam- Photons: extracranial stereotactic/SBRT
Contact Email