Session Item

In this study we aim at developing knowledge-based tools for robust ART planning to detect variations or errors in plan quality for stereotactic pancreatic ART. We developed volume-based dosimetric identifiers which can be used to identify dosimetric deviations of ART plans from the simulation plan.

57 patients treated using 0.35T MRgRT for pancreas cancer were included in this retrospective study. All patients received 50 Gy in 5 fractions. A total of 278 treatment fractions were adapted. The critical gastrointestinal luminal organs, namely stomach, duodenum, small bowel, and large bowel (SDSL) were prioritized over target coverage with the dose constraint of V36 < 0.5 cc. PTV-OPT was generated by subtracting the SDSL plus a 5mm margin from the PTV to allow for appropriate dose gradient from the prescription dose to the SDSL dose-limiting objective. Dosimetric information were retrieved from the treatment reports for both the initial and ART plans using an in-house Matlab program. Several metrics that potentially can identify poor quality plans were calculated. These metrics include V95% for target coverage for PTV and PTV-OPT, the ratio of D95% to D5% for PTV and PTV-OPT, the volume ratio between PTV-OPT and PTV, and MU per fraction. We calculated the difference of these values between each adaptive plan and the simulation plan. Simulation plans were used as the gold standard in this study based on the assumption that the simulation plans are of standard quality due to extended time for planning. We then calculated 95% confidence interval (CI) for the patient population. We evaluated whether variations in metrics for each adapted fraction exceeded the 95% CI, which were flagged for retrospective investigation to identify the cause of high variation in their dosimetric indices such as contouring or planning mistakes and determine sensitivity (Fig. 1). The same number of non-flagged ART plans were also evaluated for comparison and specificity of the tool.

During the retrospective investigation of flagged adaptive plans, a list of causes for deviations were determined. These causes include excessive normalization, inappropriate alignment (Fig. 2), large anatomical changes, etc. For those plans not flagged by our tool, adaptive plans that were found to be low quality were the plans that started with the low quality simulation plan or were under-contoured. We estimated the sensitivity (true positive) and specificity (true negative) of our method to be 84% and 76%, respectively.

We developed dosimetric indicators for ART planning QA that can identify the low quality adaptive plans for stereotactic pancreatic ART. These metrics can be used to identify deviations or actual errors during the online adaptive process, which could be useful as an ART clinical trial QA tool, reference tool for new users, and improve overall ART quality at an institution.