《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (8): 2660-2667.DOI: 10.11772/j.issn.1001-9081.2025070817

• 多媒体计算与计算机仿真 • 上一篇    

基于多尺度注意力自适应融合的稀疏CT伪影抑制Transformer网络

白杰龙, 方晨韵, 乔志伟()   

  1. 山西大学 计算机与信息技术学院,太原 030006
  • 收稿日期:2025-07-22 修回日期:2025-09-25 接受日期:2025-09-25 发布日期:2025-11-05 出版日期:2026-08-10
  • 通讯作者: 乔志伟
  • 作者简介:白杰龙(2000—),男,山西吕梁人,硕士研究生,主要研究方向:医学图像重建、图像处理
    方晨韵(1997—),女,安徽安庆人,博士研究生,主要研究方向:医学图像重建、图像处理
    乔志伟(1977—),男,山西临汾人,教授,博士,主要研究方向:电子顺磁共振成像、图像重建算法、大规模最优化。
  • 基金资助:
    国家自然科学基金资助项目(62071281);中央引导地方科技发展资金项目(YDZJSX2021A003)

Sparse CT artifact suppresion Transformer network based on multi-scale attention adaptive fusion

Jielong BAI, Chenyun FANG, Zhiwei QIAO()   

  1. School of Computer and Information Technology,Shanxi University,Taiyuan Shanxi 030006,China
  • Received:2025-07-22 Revised:2025-09-25 Accepted:2025-09-25 Online:2025-11-05 Published:2026-08-10
  • Contact: Zhiwei QIAO
  • About author:BAI Jielong, born in 2000, M. S. candidate. His research interests include medical image reconstruction, image processing.
    FANG Chenyun, born in 1997, Ph. D. candidate. Her research interests include medical image reconstruction, image processing.
  • Supported by:
    National Natural Science Foundation of China(62071281);Local Science and Technology Development Fund Project Guided by the Central Government(YDZJSX2021A003)

摘要:

稀疏计算机断层成像(CT)重建能够降低患者的辐射剂量,对临床诊断具有重要意义。在基于深度学习的图像重建任务中,经典的Uformer (U-shape Transformer)、Restormer (Restoration Transformer)和AST (Adaptive Sparse Transformer)等网络未考虑图像的多尺度与方向信息,从而忽略了局部细节和全局结构的平衡,伪影抑制效果有限。针对上述问题,提出一种多尺度注意力自适应融合Transformer(MAAF-Transformer)网络。该网络采用并行注意力融合策略,结合多尺度通道方向感知注意力(MSCDA)模块捕获伪影特征,并利用卷积自适应空间通道注意力(CASCA)模块动态调节权重,然后通过门控深度卷积前馈网络(GDFN)筛选有效信息。实验结果表明,在60个稀疏角度下,MAAF-Transformer相较于经典Uformer在峰值信噪比(PSNR)上提高了0.714 1 dB、结构相似性(SSIM)上提高了0.33%,而均方根误差(RMSE)上降低了7.76%,并在视觉效果上同样表现优异。可见,MAAF-Transformer的稀疏重建精度更高,抑制伪影能力更强。

关键词: 稀疏重建, 计算机断层成像, 多尺度, 注意力机制, 自适应融合

Abstract:

Sparse Computed Tomography (CT) reconstruction can reduce patient radiation dosage and is of great significance for clinical diagnosis. In deep learning-based image reconstruction tasks, classic networks such as Uformer (U-shape Transformer), Restormer (Restoration Transformer), and AST (Adaptive Sparse Transformer) fail to consider multi-scale and directional information in images, neglecting the balance between local details and global structure, resulting in limited artifact suppression. To address this issue, a Multi-scale Attention Adaptive Fusion Transformer (MAAF-Transformer) network was proposed. In this network, a parallel attention fusion strategy was adopted, a Multi-Scale Channel Direction-aware Attention (MSCDA) module was combined to capture artifact features, and a Convolutional Adaptive Spatial Channel Attention (CASCA) module was used to adjust the weights dynamically. Then, effective information was filtered through a Gated Deep-convolutional Feedforward Network (GDFN). Experimental results show that at 60 sparse angles, MAAF-Transformer achieves 0.714 1 dB higher Peak Signal-to-Noise Ratio (PSNR), 0.33% higher Structural SIMilarity (SSIM), and 7.76% lower Root Mean Square Error (RMSE) compared with classic Uformer, with excellent performance in terms of visual effects as well. It can be seen that MAAF-Transformer has higher sparse reconstruction accuracy and stronger artifact suppression capability.

Key words: sparse reconstruction, Computed Tomography (CT), multi-scale, attention mechanism, adaptive fusion

中图分类号: