Deep match: A zero-shot framework for improved fiducial-free respiratory motion tracking

Abstract

Background and purpose: Motion management is essential to reduce normal tissue exposure and maintain adequate tumor dose in lung stereotactic body radiation therapy (SBRT). Lung SBRT using an articulated robotic arm allows dynamic tracking during radiation dose delivery. Two stereoscopic X-ray tracking modes are available - fiducial-based and fiducial-free tracking. Although X-ray detection of implanted fiducials is robust, the implantation procedure is invasive and inapplicable to some patients and tumor locations. Fiducial-free tracking relies on tumor contrast, which challenges the existing tracking algorithms for small (e.g., <15 mm) and/or tumors obscured by overlapping anatomies. To markedly improve the performance of fiducial-free tracking, we proposed a deep learning-based template matching algorithm - Deep Match.

Method: Deep Match consists of four self-definable stages - training-free feature extractor, similarity measurements for location proposal, local refinements, and uncertainty level prediction for constructing a more trustworthy and versatile pipeline. Deep Match was validated on a 10 (38 fractions; 2661 images) patient cohort whose lung tumor was trackable on one X-ray view, while the second view did not offer sufficient conspicuity for tumor tracking using existing methods. The patient cohort was stratified into subgroups based on tumor sizes (<10 mm, 10-15 mm, and >15 mm) and tumor locations (with/without thoracic anatomy overlapping).

Results: On X-ray views that conventional methods failed to track the lung tumor, Deep Match achieved robust performance as evidenced by >80 % 3 mm-Hit (detection within 3 mm superior/inferior margin from ground truth) for 70 % of patients and <3 mm superior/inferior distance (SID) ∼1 mm standard deviation for all the patients.

Conclusion: Deep Match is a zero-shot learning network that explores the intrinsic neural network benefits without training on patient data. With Deep Match, fiducial-free tracking can be extended to more patients with small tumors and with tumors obscured by overlapping anatomy.

Keywords: Computer vision; CyberKnife; Deep learning; SBRT; Template matching; Xsight Lung Tracking.