Deformable image registration for adaptive radiation therapy of head and neck cancer: accuracy and precision in the presence of tumor changes.

Marcel Van Herk, Angelo Mencarelli, Simon Robert van Kranen, Olga Hamming-Vrieze, Suzanne van Beek, Coenraad Robert Nico Rasch, Marcel van Herk, Jan-Jakob Sonke

    Research output: Contribution to journalArticlepeer-review

    Abstract

    PURPOSE: To compare deformable image registration (DIR) accuracy and precision for normal and tumor tissues in head and neck cancer patients during the course of radiation therapy (RT). METHODS AND MATERIALS: Thirteen patients with oropharyngeal tumors, who underwent submucosal implantation of small gold markers (average 6, range 4-10) around the tumor and were treated with RT were retrospectively selected. Two observers identified 15 anatomical features (landmarks) representative of normal tissues in the planning computed tomography (pCT) scan and in weekly cone beam CTs (CBCTs). Gold markers were digitally removed after semiautomatic identification in pCTs and CBCTs. Subsequently, landmarks and gold markers on pCT were propagated to CBCTs, using a b-spline-based DIR and, for comparison, rigid registration (RR). To account for observer variability, the pair-wise difference analysis of variance method was applied. DIR accuracy (systematic error) and precision (random error) for landmarks and gold markers were quantified. Time trend of the precisions for RR and DIR over the weekly CBCTs were evaluated. RESULTS: DIR accuracies were submillimeter and similar for normal and tumor tissue. DIR precision (1 SD) on the other hand was significantly different (P
    Original languageEnglish
    JournalInternational journal of radiation oncology, biology, physics
    Volume90
    Issue number3
    DOIs
    Publication statusPublished - 1 Nov 2014

    Fingerprint

    Dive into the research topics of 'Deformable image registration for adaptive radiation therapy of head and neck cancer: accuracy and precision in the presence of tumor changes.'. Together they form a unique fingerprint.

    Cite this