کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

Unveiling Enhanced Image Quality in Sparse-View CT with OSEM- ANLM Algorithm

نویسندگان :

Sayna Jamaati¹ Seyed Abolfazl Hosseini² Mohammad Ghorbanzadeh³ Hossein Arabi⁴

1- Sharif university of technology 2- Sharif university of technology 3- Sharif university of technology 4- Geneva University Hospital

کلمات کلیدی :

CT،sparse-view،image reconstruction،OSEM،Asymptotic Non-Local Means

چکیده :

This research presents the Ordered Subset Expectation Maximization-Asymptotic Non-local Means (OSEM-ANLM) algorithm, a novel imaging reconstruction method aimed at improving Computed Tomography (CT) image quality from sparsely sampled data. The algorithm’s performance is evaluated using a patient’s chest CT scan and a brain-skull image from the Rando phantom, with projection views reduced to 60, 45, and 30 to simulate varying data sparsity levels. Comparisons are made against conventional methods, including the Algebraic Reconstruction Technique (ART), OSEM, and OSEM-Non-Local Means(NLM). Qualitative assessments demonstrate the OSEM-ANLM’s superior ability to preserve anatomical structures while minimizing noise and artifacts, even with fewer projection views. Quantitative metrics, including Peak Signal-to-Noise Ratio (PSNR), Normalized Root Mean Square Error (NRMSE), and Structural Similarity Index (SSIM), further validate its effectiveness. For the chest CT image with 30 views (the lowest number of views with the highest level of artifacts), OSEM-ANLM achieves the highest PSNR (38.2693) and SSIM (0.9797), outperforming ART (24.6231, 0.9466), OSEM (25.1310, 0.9512), and OSEM-NLM (36.4061, 0.9669). Similarly, it yields the lowest NRMSE (0.0128), compared to ART (0.0523), OSEM (0.0484), and OSEM-NLM (0.0170). For the brain-skull image, OSEM-ANLM achieves the highest PSNR (37.6986) and SSIM (0.9898), surpassing ART (21.7716, 0.9443), OSEM (23.2124, 0.9521), and OSEM-NLM (35.9652, 0.9833). It also records the lowest NRMSE (0.0160) compared to ART (0.0599), OSEM (0.0526), and OSEM-NLM (0.0279). These results highlight the proposed method’s superior reconstruction accuracy and image fidelity under sparse sampling conditions.