Abstract
In clinical examinations and diagnoses, low-dose computed tomography (LDCT)is crucial for minimizing health risks compared with normal-dose computedtomography (NDCT). However, reducing the radiation dose compromises thesignal-to-noise ratio, leading to degraded quality of CT images. To addressthis, we analyze LDCT denoising task based on experimental results from thefrequency perspective, and then introduce a novel self-supervised CT imagedenoising method called WIA-LD2ND, only using NDCT data. The proposed WIA-LD2NDcomprises two modules: Wavelet-based Image Alignment (WIA) and Frequency-AwareMulti-scale Loss (FAM). First, WIA is introduced to align NDCT with LDCT bymainly adding noise to the high-frequency components, which is the maindifference between LDCT and NDCT. Second, to better capture high-frequencycomponents and detailed information, Frequency-Aware Multi-scale Loss (FAM) isproposed by effectively utilizing multi-scale feature space. Extensiveexperiments on two public LDCT denoising datasets demonstrate that ourWIA-LD2ND, only uses NDCT, outperforms existing several state-of-the-artweakly-supervised and self-supervised methods.