VFSS 영상 잡음 저감을 위한 적응형 잡음분산 조건부 평균화 및 경계보존 디테일 보상 기법

초록

In this study, an adaptive noise variance conditioned averaging (NVCA)-based denoising framework was proposed to effectively reduce signal-dependent quantum noise in low-dose videofluoroscopic swallowing study (VFSS) images while preserving swallowing-related anatomical structures and motion information. To model the Poissonian-Gaussian mixed noise characteristics of low-dose fluoroscopy, a signal-dependent noise model was employed, and adaptive threshold maps were generated using spatial edge maps and temporal motion maps. Subsequently, weighted averaging based on spatio-temporal neighborhood accumulation was performed for noise suppression, followed by edge-preserving detail compensation to restore structural sharpness. The proposed method was evaluated using simulated low-dose VFSS datasets and compared with block-matching and 3D filtering (BM3D) and conventional NVCA methods. Quantitative evaluation demonstrated that the proposed method achieved the highest performance with an SSIM of 0.85±0.08 and an EPI of 0.81±0.07. In computation time analysis, the proposed framework required 12.40 sec for processing 246 frames, demonstrating approximately 41-fold faster performance than BM3D. Enlarged visual comparison further confirmed improved preservation of bolus boundaries and swallowing-related anatomical structures. Therefore, the proposed framework may serve as an effective image processing approach for reducing the trade-off between radiation dose reduction and image quality preservation in low-dose VFSS applications.

키워드