B-Spline Registration Based on New Concept of an Intelligent Masking Procedure and GPU Computations for the Head and Neck Adaptive Tomotherapy

Abstract

The deformable image registration (DIR) procedure has been optimized for helical tomotherapy. The data on registration shifts obtained on matching planning image with pre-treatment megavoltage CT are used in our software for acceleration of the first step (rigid registration) of the DIR procedure and for implementation of the B-Spline algorithm with intelligent masking. Priorities of the masks were automatically calculated based on disagreement detected during rigid registration. Evaluation tasks included: (a) comparison of accuracy and rate for schemes of pre-registered and non-registered images; (b) qualification of the effectiveness of the intelligent masking process, and (c) determination of acceleration of achievable with GPU computing. A specially designed head and neck phantom used for evaluation included structures with controlled changes of position, volume, density, and shape. Re-contouring procedures were performed with an Adaptive Planning software (Tomotherapy Inc.). No statistical difference was observed in accuracy of DIR based on structure position match on the tomotherapy unit and non pre-registered images (p > 0.7). Using pre-registered data reduces the total time required for execution of the elastic registration procedure by 5%. These data are also necessary for intelligent masking procedure during B-Spine registration. Intelligent masking procedure increases accuracy of the registration for a masked structure (p < 0.04) without decreasing the accuracy in non-masked tissues and additionally reduces the total time by 13%. GPU computations speed up procedure 30 times. GPU computing of the DIR in current status of our investigation could be realized in a relatively short time after pre-treatment imaging. The proposed approach can be used in the routine assessment of anatomic changes occurring in healthy tissue during the course of radiotherapy. Further developments will be concentrated on the full integration of DIR computations in the imaging and treatment process of helical tomotherapy.