Abstract

Info

Author(s):: YANG Zhihui1; 2; LI Hao1; 2; LI Fengsen1; 2; 1. School of Medical Information Engineering, Gansu University of Chinese Medicine, Lanzhou 730000, China 2. Gansu Provincial Hospital (First Clinical Medical School, Gansu University of Chinese Medicine), Lanzhou 730000, China

Keywords:: Keywords: magnetic resonance imaging brain synthetic computed tomography U-Net

Abstract:: Abstract: Objective To propose an improved U-Net approach for generating synthetic computed tomography (sCT) from brain magnetic resonance imaging (MRI), achieving MRI-to-sCT translation, and further compare the brain sCT generation performance of the improved model with that of the traditional U-Net model. Methods The SynthRAD2023 dataset for CT image generation in radiotherapy planning was adopted in the study. The brain MRI and CT images of 120 patients were selected as the training set, while those from 39 patients were used as the test set. A traditional U-Net model and an improved U-Net model were constructed, and their generated sCT were evaluated regarding image quality and intensity values. Results For the cases in the test set, the average absolute error relative to ground truth CT reached (100.28±49.67) HU for the traditional U-Net model and (94.87±42.50) HU for the improved model. The corresponding peak signal-to-noise ratios were (26.43±2.70) dB and (26.85±2.73) dB, and the structural similarities were 0.808±0.092 and 0.820±0.085, respectively. Conclusion The proposed improved U-Net model outperforms the traditional U-Net model in terms of accuracy for brain MRI-to-sCT translation.