๐ First Visit: Redirecting automatically to original journal site...
โธ๏ธ STOP REDIRECT & STAY ON PAGESlice Consistent Three Dimensional Modeling for CT Reconstruction | Multidisciplinary Research in Computing Information Systems
Abstract
Abstract
Most CT denoising methods operate on individual slices and do not explicitly exploit inter-slice continuity. Based on hybrid convolutionโattention designs exemplified by CTLformer, this paper proposes a slice-consistent reconstruction approach that extends convolutional and self-attention operations to three-dimensional contexts. Local convolutions capture intra-slice texture, while attention mechanisms aggregate information across adjacent slices. The model is evaluated on volumetric CT datasets containing 7,200 3D scans and approximately 180,000 slices. Compared with 2D CNNs, 3D CNNs, and transformer-based volumetric models, the proposed method improves volumetric PSNR by 1.0โ1.6 dB and reduces inter-slice intensity discontinuities
Most CT denoising methods operate on individual slices and do not explicitly exploit inter-slice continuity. Based on hybrid convolutionโattention designs exemplified by CTLformer, this paper proposes a slice-consistent reconstruction approach that extends convolutional and self-attention operations to three-dimensional contexts. Local convolutions capture intra-slice texture, while attention mechanisms aggregate information across adjacent slices. The model is evaluated on volumetric CT datasets containing 7,200 3D scans and approximately 180,000 slices. Compared with 2D CNNs, 3D CNNs, and transformer-based volumetric models, the proposed method improves volumetric PSNR by 1.0โ1.6 dB and reduces inter-slice intensity discontinuities