저선량 치과 콘빔CT 영상에서 잡음 감소와 구조적 선명도 보존을 위한 해부학적 구조 인지형 U-Net 기반 잡음 제거 방법

초록

Low-dose dental cone-beam computed tomography (CBCT) imaging is widely used to reduce radiation exposure; however, increased image noise and structural sharpness degradation remain major limitations. In this study, a anatomical structure-aware U-Net-based denoising framework was proposed for simultaneous noise suppression and anatomical structure preservation in low-dose dental CBCT images. The proposed method separated input low-dose CBCT images into bone, soft-tissue, and air regions using intensity-based segmentation and regionprops-based refinement. Subsequently, tissue-specific adaptive denoising was performed using region-wise U-Net processing. To minimize discontinuity and step artifacts at tissue boundaries, Gaussian alpha blending-based reconstruction was additionally applied. The denoising performance was quantitatively evaluated using contrast-to-noise ratio, coefficient of variation, edge preservation index, and noise power spectrum. Experimental results demonstrated that the proposed method effectively reduced noise while preserving cortical boundaries and trabecular bone structures compared with conventional BM3D and U-Net methods. In particular, the proposed framework maintained high-frequency spectral characteristics and achieved superior edge preservation performance. Therefore, the proposed anatomical structure-aware denoising framework may provide an effective approach for improving image quality in low-dose dental CBCT imaging.

키워드