Yenny Z Szeto1, Marnix G Witte1, Marcel van Herk1, Jan-Jakob Sonke2. 1. Department of Radiation Oncology, The Netherlands Cancer Institute, Amsterdam, The Netherlands. 2. Department of Radiation Oncology, The Netherlands Cancer Institute, Amsterdam, The Netherlands. Electronic address: j.sonke@nki.nl.
Abstract
PURPOSE: To develop a population based statistical model of the systematic interfraction geometric variations between the planning CT and first treatment week of lung cancer patients for inclusion as uncertainty term in future probabilistic planning. MATERIALS AND METHODS: Deformable image registrations between the planning CT and first week CBCTs of 235 lung cancer patients were used to generate deformation vector fields (DVFs) representing the geometric variations of lung cancer patients. Using a second deformable registration step, the average DVF per patient was mapped to an average patient CT. Subsequently, the dominant modes of systematic geometric variations were extracted using Principal Component Analysis (PCA). For evaluation a leave-one-out cross-validation was performed. RESULTS: The first three PCA components mainly described cranial-caudal, anterior-posterior, and left-right variations, respectively. Fifty and 112 components were needed to describe correspondingly 75% and 90% of the variance. An overall systematic variation of 3.6mm SD was observed and could be described with an accuracy of about 1.0mm with the PCA model. CONCLUSIONS: A PCA based model for systematic geometric variations in the thorax was developed, and its accuracy determined. Such a model can serve as a basis for probability based treatment planning in lung cancer patients.
PURPOSE: To develop a population based statistical model of the systematic interfraction geometric variations between the planning CT and first treatment week of lung cancerpatients for inclusion as uncertainty term in future probabilistic planning. MATERIALS AND METHODS: Deformable image registrations between the planning CT and first week CBCTs of 235 lung cancerpatients were used to generate deformation vector fields (DVFs) representing the geometric variations of lung cancerpatients. Using a second deformable registration step, the average DVF per patient was mapped to an average patient CT. Subsequently, the dominant modes of systematic geometric variations were extracted using Principal Component Analysis (PCA). For evaluation a leave-one-out cross-validation was performed. RESULTS: The first three PCA components mainly described cranial-caudal, anterior-posterior, and left-right variations, respectively. Fifty and 112 components were needed to describe correspondingly 75% and 90% of the variance. An overall systematic variation of 3.6mm SD was observed and could be described with an accuracy of about 1.0mm with the PCA model. CONCLUSIONS: A PCA based model for systematic geometric variations in the thorax was developed, and its accuracy determined. Such a model can serve as a basis for probability based treatment planning in lung cancerpatients.