Efficient parameter estimation for anatomy deformation models used in 4D-CT

Abstract

A critical feature of radiation therapy for cancerous tumors located in the thorax and abdomen is addressing tumor motion due to breathing. To achieve this goal, a CT study (ordinarily denoted as 4D-CT) showing tumor loca-tion, size, and shape against time is essential. Several 4D-CT reconstruction methods have been proposed that employ anatomy deformation models. The proposed method estimates temporal parameters for these models using an ap-proach that does not require markers or manual designation of landmark anatomical features. A neural network is trained to estimate the parameters based on simple statistical features of the CT projections. The proposed method achieves an average estimation error of less than 0.02 seconds, corresponding to a spatial error of less than 1.3 mm. The accuracy of the proposed method is evaluated in the presence of several limiting constraints such as computational complexity and noise.