Premium
SU‐E‐J‐108: Template Matching Based On Multiple Templates Can Improve the Tumor Tracking Performance When There Is Large Tumor Deformation
Author(s) -
Shi X,
Lin J,
Diwanji T,
Mooney K,
D' Souza W,
Mistry N
Publication year - 2014
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4888160
Subject(s) - template matching , template , tracking (education) , computer science , artificial intelligence , matching (statistics) , computer vision , deformation (meteorology) , rotation (mathematics) , ground truth , nuclear medicine , biomedical engineering , pattern recognition (psychology) , medicine , materials science , image (mathematics) , pathology , psychology , pedagogy , programming language , composite material
Purpose: Recently, template matching has been shown to be able to track tumor motion on cine‐MRI images. However, artifacts such as deformation, rotation, and/or out‐of‐plane movement could seriously degrade the performance of this technique. In this work, we demonstrate the utility of multiple templates derived from different phases of tumor motion in reducing the negative effects of artifacts and improving the accuracy of template matching methods. Methods: Data from 2 patients with large tumors and significant tumor deformation were analyzed from a group of 12 patients from an earlier study. Cine‐MRI (200 frames) imaging was performed while the patients were instructed to breathe normally. Ground truth tumor position was established on each frame manually by a radiation oncologist. Tumor positions were also automatically determined using template matching with either single or multiple (5) templates. The tracking errors, defined as the absolute differences in tumor positions determined by the manual and automated methods, when using either single or multiple templates were compared in both the AP and SI directions, respectively. Results: Using multiple templates reduced the tracking error of template matching. In the SI direction where the tumor movement and deformation were significant, the mean tracking error decreased from 1.94 mm to 0.91 mm (Patient 1) and from 6.61 mm to 2.06 mm (Patient 2). In the AP direction where the tumor movement was small, the reduction of the mean tracking error was significant in Patient 1 (from 3.36 mm to 1.04 mm), but not in Patient 2 (from 3.86 mm to 3.80 mm). Conclusion: This study shows the effectiveness of using multiple templates in improving the performance of template matching when artifacts like large tumor deformation or out‐of‐plane motion exists. Accurate tumor tracking capabilities can be integrated with MRI guided radiation therapy systems. This work was supported in part by grants from NIH/NCI CA 124766 and Varian Medical Systems, Palo Alto, CA.