Premium
SU‐E‐T‐572: A Plan Quality Metric for Evaluating Knowledge‐Based Treatment Plans
Author(s) -
Chanyavanich V,
Lo J,
Das S
Publication year - 2012
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4735661
Subject(s) - metric (unit) , similarity (geometry) , plan (archaeology) , computer science , quality (philosophy) , radiation treatment planning , data mining , a priori and a posteriori , artificial intelligence , medical physics , mathematics , medicine , radiology , image (mathematics) , operations management , radiation therapy , geography , philosophy , archaeology , epistemology , economics
Purpose: In prostate IMRT treatment planning, the variation in patient anatomy makes it difficult to estimate a priori the potentially achievable extent of dose reduction possible to the rectum and bladder. We developed a mutual information‐based framework to estimate the achievable plan quality for a new patient, prior to any treatment planning or optimization. Methods: The knowledge‐base consists of 250 retrospective prostate IMRT plans. Using these prior plans, twenty query cases were each matched with five cases from the database. We propose a simple DVH plan quality metric (PQ) based on the weighted‐sum of the areas under the curve (AUC) of the PTV, rectum and bladder. We evaluate the plan quality of knowledge‐based generated plans, and established a correlation between the plan quality and case similarity. Results: The introduced plan quality metric correlates well (r2 = 0.8) with the mutual similarity between cases. A matched case with high anatomical similarity can be used to produce a new high quality plan. Not surprisingly, a poorly matched case with low degree of anatomical similarity tends to produce a low quality plan, since the adapted fluences from a dissimilar case cannot be modified sufficiently to yield acceptable PTV coverage. Conclusions: The plan quality metric is well‐correlated to the degree of anatomical similarity between a new query case and matched cases. Further work will investigate how to apply this metric to further stratify and select cases for knowledge‐based planning.