Abstract

Info

Title:: Super-resolution reconstruction algorithm of CBCT image based on neural network learning

Author(s):: DENG Chunyan; LU Jiayang; HUANG Baotian; Department of Radiation Oncology, Cancer Hospital of Shantou University Medical College, Shantou 515000, China

Keywords:: Keywords: cone-beam computed tomography convolutional neural network denoise super-resolution construction

Abstract:: Abstract: A reconstruction method based on super-resolution convolutional neural network (SRCNN) is proposed to solve the problem of poor cone-beam computed tomography (CBCT) image quality, thereby improving the resolution of CBCT image. The CBCT images of head and neck, pelvic cavity and thorax were researched. Firstly, image noises were removed by non-local means method, and then super-resolution reconstruction is carried out by bicubic interpolation (BIC) and SRCNN, separately. The results show that both BIC method and SRCNN method can improve the resolution of CBCT image. The peak signal-to-noise ratio (PSNR) obtained by SRCNN method is higher than that obtained by BIC method, but the differences in structural similarity (SSIM) and feature similarity (FSIM) between SRCNN method and BIC method are trivial. The analysis on PSNR and FSIM shows that SRCNN method has more remarkable effect on the improvement of pelvic CBCT image, and the effects on the improvements of head and neck CBCT image and thoracic CBCT image are similar.