Click here to

Session: Quality Assurance [Return to Session]

Multi-Institution Plan Complexity Characteristics Based On IROC Phantoms

V Desai1*, S Kry2, (1) UW Health, Madison, WI, (2) MD Anderson Cancer Ctr., Houston, TX


WE-C-TRACK 3-5 (Wednesday, 7/28/2021) 1:00 PM - 2:00 PM [Eastern Time (GMT-4)]

Purpose: To evaluate how much treatment plan variability exists across institutions developing plans to satisfy common objectives.

Methods: IROC phantoms provide consistent anatomy and plan objectives, but the institution is allowed to develop a plan based on their treatment techniques. Four standard IROC phantoms (head-and-neck, prostate, lung, and spine) were evaluated based on 24 complexity metrics for more than 1700 plans developed across the radiotherapy community. Statistical tests were used to assess the spread within, and differences between, the complexity of phantom plans. These metrics were also evaluated on a single institution database containing 627 clinical patient plans over the same 4 anatomical sites. A k-means clustering algorithm was applied to the principal components of the 24 complexity metrics to determine whether any distinct plan clusters existed.

Results: For the 4 IROC phantoms, median complexity results were distinct from one another: the H&N phantom and spine phantom were most complex, the prostate and lung least complex, consistent with the complexity of the phantom anatomy. However, there was also a large spread of results for each phantom in terms of solution complexity. These two trends were also seen for the clinical plans. The k-means algorithm demonstrated that complexity metrics successfully stratified the IROC phantom data into two groups, namely prostate/lung and H&N/spine. However, these clusters demonstrated low inter-cluster variance and high intracluster variance. Additionally, step-and-shoot and sliding window plans were quite distinct in terms of complexity, while VMAT plans were not.

Conclusion: Consistency in planning across the community is relatively low as substantial plan variability was seen for common plan objectives on identical treatment geometries. In addition, plan complexity metrics were not uniquely descriptive of a particular class of plans, even for a single phantom geometry, suggesting they may have limited utility as plan descriptors for QA.



    Not Applicable / None Entered.


    Not Applicable / None Entered.

    Contact Email