Clinical implementation of dose-volume histogram predictions for organs-at-risk in IMRT planning

17th International Conference on the Use of Computers in Radiation Therapy

物理学;计算机科学

Moore, K.L.^1 ; Appenzoller, L.M.^2 ; Tan, J.^2 ; Michalski, J.M.^2 ; Thorstad, W.L.^2 ; Mutic, S.^2

Department of Radiation Medicine and Applied Sciences, University of California, San Diego, CA 92093, United States^1

Department of Radiation Oncology, Washington University in St. Louis, St. Louis, MO 63110, United States^2

关键词: Anatomical variability;    Automated systems;    Dose distributions;    Dose-volume histograms;    Predictive models;    Quantitative assessments;    Three-dimensional dose distribution;    Treatment planning systems;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/489/1/012055/pdf
DOI  :  10.1088/1742-6596/489/1/012055

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

True quality control (QC) of the planning process requires quantitative assessments of treatment plan quality itself, and QC in IMRT has been stymied by intra-patient anatomical variability and inherently complex three-dimensional dose distributions. In this work we describe the development of an automated system to reduce clinical IMRT planning variability and improve plan quality using mathematical models that predict achievable OAR DVHs based on individual patient anatomy. These models rely on the correlation of expected dose to the minimum distance from a voxel to the PTV surface, whereby a three-parameter probability distribution function (PDF) was used to model iso-distance OAR subvolume dose distributions. DVH models were obtained by fitting the evolution of the PDF with distance. Initial validation on clinical cohorts of 40 prostate and 24 head-and-neck plans demonstrated highly accurate model-based predictions for achievable DVHs in rectum, bladder, and parotid glands. By quantifying the integrated difference between candidate DVHs and predicted DVHs, the models correctly identified plans with under-spared OARs, validated by replanning all cases and correlating any realized improvements against the predicted gains. Clinical implementation of these predictive models was demonstrated in the PINNACLE treatment planning system by use of existing margin expansion utilities and the scripting functionality inherent to the system. To maintain independence from specific planning software, a system was developed in MATLAB to directly process DICOM-RT data. Both model training and patient-specific analyses were demonstrated with significant computational accelerations from parallelization.