X-ray computed tomography (CT) is vital for medical diagnostics, but frequent radiation exposure raises concerns, driving the adoption of low-dose CT (LDCT) to mitigate risks. However, LDCT often introduces noise, compromising diagnostic accuracy. This paper proposes a pure vision transformer (PViT) for LDCT denoising, enhanced with a gradient–Laplacian attention module (GLAM) to improve edge preservation and fine structural detail reconstruction. The model’s robustness was validated across five diverse datasets (piglet, head, abdomen, chest, thoracic), demonstrating consistent performance in preserving anatomical structures. Extensive ablation studies on attention configurations and loss functions further substantiated the contributions of each module. Quantitative evaluation using PSNR and SSIM, alongside radiologist assessment, confirmed significant noise suppression and sharper anatomical boundaries, particularly in regions with fine details such as organ interfaces and bone structures. Additionally, in benchmark comparisons against state-of-the-art LDCT models (RED-CNN, TED-Net, DSC-GAN, DRL-EMP) and traditional methods (BM3D), the model exhibited lower parameter and stable training performance. These findings highlight the model’s robustness, efficiency, and clinical applicability, making it a promising solution for improving LDCT image quality while maintaining computational efficiency.
1 October 2025 in International Journal of Imaging Systems and Technology
Low‐dose computed tomography (LDCT) is widely used to promote reduction of patient radiation exposure, but the associated increase in image noise poses challenges for diagnostic accuracy. In this study, we propose a Vision Transformer (ViT)‐based denoising framework enhanced with a State Space Optimizing Block (SSOB) to improve both image quality and computational efficiency. The SSOB upgrades the multihead self‐attention mechanism by reducing spatial redundancy and optimizing contextual feature fusion, thereby strengthening the transformer's ability to capture long‐range dependencies and preserve fine anatomical structures under severe noise. Extensive evaluations on randomized and categorized datasets demonstrate that the proposed model consistently outperforms existing state‐of‐the‐art denoising approaches. It achieved the highest average SSIM (up to 6.10% improvement), PSNR values (36.51 ± 0.37 dB on randomized and 36.30 ± 0.36 dB on categorized datasets), and the lowest RMSE, surpassing recent CNN‐transformer‐based denoising hybrid models by approximately 12%. Intensity profile analysis further confirmed its effectiveness, showing sharper edge transitions and more accurate gray‐level distributions across anatomical boundaries, closely aligning with ground truth and retaining subtle diagnostic features often lost in competing models. In addition to improved reconstruction quality, the SSOB‐empowered ViT achieved notable computational gains. It delivered the fastest inference (0.42 s per image), highest throughput (2.38 images/s), lowest GPU memory usage (750 MB), and smallest model size (7.6 MB), alongside one of the shortest training times (6.5 h). Compared to legacy architectures, which required up to 16 h of training and substantially more resources, the proposed model offers both accuracy and deployability. Collectively, these findings establish the SSOB as a key component for efficient transformer‐based LDCT denoising, addressing memory and convergence challenges while preserving global contextual advantages.
Low-Dose Computed Tomography (LDCT) reduces radiation exposure but introduces severe quantum noise and streak artifacts that degrade image quality. To address these challenges, we propose the Physics-Guided Sequential State Space Transformer (), a projection-domain denoising framework that combines sequential state-space modeling with a photon-aware attention mechanism to capture long-range dependencies across projection angles with linear computational complexity. A differentiable Filtered Backprojection (FBP) layer further enforces reconstruction-domain consistency during training. The proposed framework was evaluated on the Mayo Clinic LDCT and Projection Dataset using patient-level dataset partitioning. Experimental results demonstrate that consistently outperforms state-of-the-art methods, including DRL, SADiff, and GEDFormer, across the abdomen, head, and chest datasets. On the abdomen dataset, reached a peak PSNR of 42.40 dB, an SSIM of 0.9020, and the lowest RMSE of 0.0076 across anatomical regions. Statistical analysis using 95% confidence intervals and paired Wilcoxon signed-rank tests confirmed that these improvements were significant (). Furthermore, achieved the lowest reconstruction consistency loss (0.0128 at epoch 50), demonstrating stable convergence and the effectiveness of incorporating acquisition physics into projection-domain LDCT denoising.