Click here to


Are you sure ?

Yes, do it No, cancel

Plan Quality Assessment for Rectal Cancer Patients Using Prediction of Organ-At-Risk Dose Metrics

A Vaniqui*, R Canters, F Vaassen, C Hazelaar, I Lubken, K Kremer, C Wolfs, W Van Elmpt, Maastro, Maastricht, NL,


(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

With the introduction of automatic planning, assessment of plan quality of individual plans becomes more difficult. We developed an automatic anatomy-based plan quality method that predicts the organ-at-risk (OAR) dose. This method was applied to rectal cancer patients and compared patient cohorts of automatically generated treatment plans with manual treatment planning.

196 rectal cancer patients treated according to institutional guidelines in 2018 with a VMAT protocol (25x2Gy) were included. A dose prediction model based on the overlap-volume histogram (OVH) concept (Petit and van Elmpt, R&O-2015) was trained on a cohort of 22 patients. From this cohort, the mean dose relative to the distance from the planning target volume (PTV) to the two relevant OARs (bowel bag and bladder) was calculated. From this dose-distance curve DVH prediction parameters for the selected OAR was derived solely based on the anatomical position of the PTV and OAR. This model was subsequently evaluated on two validation cohorts: 1) 93 patients treated using a manually optimized treatment plan, 2) 95 patients automatically planned (RapidPlanTM, Varian Medical Systems). OAR dose differences for the PTV coverage, bowel bag and bladder were compared to the predicted dose levels from the model.

For cohort 1 differences between predicted and achieved doses to the bowel bag and bladder (normalized to the mean PTV dose) were 0.5±2.9% and -0.2±2.1%, respectively, and for cohort 2, the differences for bowel bag and bladder mean doses were of -0.2 ± 2.1% and -0.4 ± 0.6%.

An independent patient-specific plan QA method for rectal cancer patients was validated in two clinical cohorts. Dose values were accurately predicted for both OARs. Manual treatment planning showed larger variation and sub-optimal plan quality compared to RapidPlanTM generated treatment plans. This QA method can be used clinically to flag outliers in plan quality.


Treatment Verification, Quality Assurance, Radiation Therapy


TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

Contact Email