Extracting lung contour deformation features with deep learning for internal target motion tracking: a preliminary study.

Researchers

Journal

Modalities

Models

Abstract

To propose lung contour deformation feature (LCDF) as a surrogate to estimate the thoracic internal target motion, and to report its performance by correlating with the changing body using a cascade ensemble model (CEM). LCDF, correlated to the respiration driver, is employed without patient-specific motion data sampling and additional training before treatment.
Approach: LCDF is extracted by matching lung contours via an encoder-decoder deep learning model. CEM estimates LCDF from the currently-captured body, and then uses the estimated LCDF to track internal target motion. The accuracy of the proposed LCDF and CEM was evaluated using 48 targets’ motion data, and compared with other published methods
Main results: LCDF estimated the internal targets with a localization error of 2.6±1.0mm (average ± standard deviation). CEM reached a localization error of 4.7±0.9mm and a real-time performance of 256.9±6.0ms. Although with no internal anatomy knowledge, they achieved a small accuracy difference (of 0.34~1.10mm for LCDF and of 0.43~1.75mm for CEM at 95% confidence level) with a patient-specific lung biomechanical model and the deformable image registration models.
Significance: The results demonstrated the effectiveness of LCDF and CEM on tracking target motion. LCDF and CEM are non-invasive, and require no patient-specific training before treatment. They show potential for broad applications.Creative Commons Attribution license.