(Wednesday, 7/17/2019) 9:30 AM - 10:00 AM
Room: Exhibit Hall | Forum 5
Purpose: To develop prediction models of soft-tissue voxel displacement between planning and treatment due to weight loss and organ shrinkage in head-and-neck (HN) radiotherapy. Such models can potentially be used to predict organ segmentations and dose accumulation in future fractions, based on images of the initial fractions, so as to inform plan adaptation decisions. We report the results of our first step in building baseline prediction models to assist future development of more advanced models.
Methods: Pre-treatment (week zero, w0) and weekly (w1, w2, w3 and w4) T2 fat-suppression MRI scans were acquired over the fractionated treatment course. Weekly images were rigidly registered to w0 to remove setup errors. Deformable registrations between w0 and each of the weekly images was accomplished using publicly available software (Plastimatch). For each voxel at w0, a linear-log model of displacement vectors (DVs) was fitted to w1, w2 and w3 to predict the DV for w4. The predicted voxel displacements derived from the fit were compared to those from w0-to-w4 DVs obtained from deformable registration as a control and mean displacement errors were calculated. In addition, pre-treatment parotid segmentations were propagated to w4 using the predicted DVs and compared to expert manual segmentation on w4 in terms of Dice coefficient and 95% Hausdorff distance. The study used MR data from ten patients enrolled in an IRB-approved protocol.
Results: Mean ± standard deviation error of predicted DVs was 5.3±3.4 mm. DICE coefficient of the predicted parotid segmentations was 0.67±0.11 and 95% Hausdorff distance was 5.2±1.7 mm.
Conclusion: The preliminary findings indicate the feasibility to predict displacement fields of anatomical change caused by response to HN radiotherapy. This is an initial step towards predictive dose accumulation to inform plan adaptation decision making. Further validation with larger patient statistics is warranted
Funding Support, Disclosures, and Conflict of Interest: This research is supported by Varian Medical System
Keywords
Deformation, Dose, Linear Regression Analysis
Taxonomy
IM/TH- Image registration : General (Most aspects)
Contact Email
