Automated lung tumor detection in fluoroscopic images for breathing motion mitigation in proton therapy

Details

Supervisors

Lee, John Aldo

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil biomédical

Abstract

Tumor motion is one of the major issues of radiotherapy that can lead to discrepancies between planned dose and dose absorbed by the patient. Due to the proton beam physics, proton therapy is particularly sensitive to this source of geometrical uncertainties. In the specific case of lung cancer, tumor motions induced by breathing can be substantial, which greatly complicate the treatment by proton therapy. However, motion mitigation techniques offer the opportunity to deal with intrafractional motions and ensure conformity in the delivered dose distribution profile. Two of these techniques, beam gating and beam tracking, rely on a real-time monitoring of the tumor. For this purpose, dedicated image processing algorithms must be designed in order to detect lung tumors in images acquired during the treatment. The purpose of this thesis is to investigate methods for lung tumor detection in fluoroscopic images in order to mitigate tumor motion in proton therapy. Here, detection has to be taken in the broad sense of the term and, depending on the strategy used to mitigate motion, information about either the contour or the location of the tumor center can be computed. Both segmentation and tracking methods are then under scope of this work. A segmentation tool based on graph cut as well as a kernel mean-shift tracking method are investigated. These two approaches have been tested on digitally reconstructed radiographs after application of a contrast enhancement method implementing the contrast limited adaptive histogram equalization algorithm.