广东工业大学学报

• •    

结合GAN损失与预训练模型的半监督SPECT重建方法

徐锦华, 李斯   

  1. 广东工业大学 计算机学院, 广东 广州 510006
  • 收稿日期:2023-11-19 出版日期:2024-09-27 发布日期:2024-09-27
  • 通信作者: 李斯(1985–) ，男，副教授，博士，主要研究方向为医学影像、最优化理论等，E-mail: sili@gdut.edu.cn
  • 作者简介:徐锦华(1999–) ，男，硕士研究生，主要研究方向为深度学习、医学影像，E-mail:2112105008@mail2.gdut.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(11771464)；广东省自然科学基金资助项目(2022A1515012379)

Combining GAN loss with pre-trained model for semi-supervised SPECT reconstruction

Xu Jin-hua, Li Si   

  1. School of Computer Science and Technology, Guangdong University of Technology, Guangzhou 510006, China
  • Received:2023-11-19 Online:2024-09-27 Published:2024-09-27

HTML

PDF (PC)

21

摘要： 单光子发射计算机断层扫描(Single-Photon Emission Computerized Tomography，SPECT)检查中所使用的放射性示踪剂会对人体造成辐射损害，因此，低剂量SPECT在核医学成像中受到广泛关注。在低剂量成像条件下，投影数据受到严重的噪声污染。已有大量研究工作探索基于全监督的深度学习重建方法以抑制图像噪声。全监督方法所得的图像质量取决于标签的数量与质量。然而，用作监督标签的正常剂量图像在临床上难以获取。为克服上述困难，本文提出一种带有GAN(Generative Adversarial Network)损失的预训练平均教师方法，以实现低剂量SPECT重建。所提方法为平均教师模型引入基于Swin-Conv-Unet的预训练模型，以提高未标记训练数据的可靠性。教师模型通过一致性正则化监督学生模型；预训练模型使用少量标记数据进行训练，并通过GAN损失提高监督可靠性。数值实验验证所提方法在抑制噪声和保持特征方面的性能，相较于平均教师方法，论文重建方法所得图像的SSIM提高了2%，RMSE降低了9%，PSNR提高了0.77 dB。数据集由SIMIND仿真软件使用XCAT数字模体生成。

关键词: 半监督学习, SPECT重建, 平均教师, 预训练模型

Abstract: The radioactive tracers used in Single-Photon Emission computerized Tomography (SPECT) scans can cause radiation exposure to the human body. Therefore, low-dose SPECT has attracted widespread attention in nuclear medicine imaging. Under low-dose imaging conditions, projection data is heavily contaminated by severe noise. There has been a significant amount of studies that explore fully supervised deep learning reconstruction methods to suppress image noise. The quality of images obtained by fully supervised methods depend on the quantity and quality of the labels. However, it is challenging to obtain the normal-dose images with supervised labels in clinical practice. To overcome the challenge, we propose a pre-trained mean teacher method with GAN loss to achieve low-dose SPECT reconstruction. The proposed method introduces a Swin-Conv-Unet-based pre-trained model into the mean teacher model to enhance the reliability of unlabeled training data. The teacher model supervises the student model through consistency regularization; the pre-trained model is trained with a small amount of labeled data and enhances the supervision reliability through GAN loss. Numerical experiments validate the performance of the proposed method in noise suppression and feature preservation. When compared with the mean teacher method, the SSIM of the reconstructed images is increased by 2%, the RMSE is reduced by 9%, and the PSNR is increased by 0.77 dB. The dataset is generated by the SIMIND simulation software using XCAT digital phantoms.

Key words: Semi-supervised learning, SPECT reconstruction, Mean teacher, Pre-trained model

中图分类号: 

  • TP391
[1] ZHANG J H, LI S, KROL A, et al. Infimal convolution-based regularization for SPECT reconstruction [J]. Medical Physics, 2018, 45: 5397-5410.
[2] KROL A, LI S, SHEN L X, et al. Preconditioned alternating projection algorithms for maximum a posteriori ECT reconstruction [J]. Inverse Problems, 2012, 28: 115005.
[3] JIANG Y, LI S, XU Y S. A higher-order polynomial method for SPECT reconstruction [J]. IEEE Transactions on Medical Imaging, 2018, 38: 1271-1283.
[4] RAVISHANKAR S, YE J C, FESSLER J A. Image reconstruction: from sparsity to data-adaptive methods and machine learning [J]. Proceedings of the IEEE, 2019, 108: 86-109.
[5] CHEN X, ZHOU B, XIE H, et al. DuDoSS: deep-learning-based dual-domain sinogram synthesis from sparsely sampled projections of cardiac SPECT [J]. Medical Physics, 2023, 50: 89-103.
[6] LI S, PENG L, LI F, et al. Low-dose sinogram restoration enabled by conditional GAN with cross-domain regularization in SPECT imaging [J]. Mathematical Biosciences and Engineering, 2023, 20: 9728-9758.
[7] 叶文权, 李斯, 凌捷. 基于多级残差U-Net的稀疏SPECT图像重建[J]. 广东工业大学学报, 2023, 40(1): 61-67.
YE W Q, LI S, LING J. Sparse-view SPECT image reconstruction based on multilevel-residual U-Net [J]. Journal of Guangdong University of Technology, 2023, 40(1): 61-67.
[8] DONG X, VEKHANDE S, CAO G. Sinogram interpolation for sparse-view micro-CT with deep learning neural network[C]//CHEN G H. Medical Imaging 2019: Physics of Medical Imaging. San Diego: SPIE, 2019, 10948: 692-698.
[9] YUAN H, JIA J, ZHU Z. SIPID: a deep learning framework for sinogram interpolation and image denoising in low-dose CT reconstruction[C]//AMINI A. 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) . Washington: IEEE, 2018: 1521-1524.
[10] SHAO W, ROWE S P, DU Y. SPECTnet: a deep learning neural network for SPECT image reconstruction [J]. Annals of Translational Medicine, 2021, 9(9): 819.
[11] LI Z, DEWARAJA Y K, FESSLER J A. Training end-to-end unrolled iterative neural networks for SPECT image reconstruction [J]. IEEE Transactions on Radiation and Plasma Medical Sciences, 2023, 7(4): 410-420.
[12] SHAO W, POMPER M G, DU Y. A learned reconstruction network for SPECT imaging [J]. IEEE Transactions on Radiation and Plasma Medical Sciences, 2020, 5(1): 26-34.
[13] HÄGGSTRÖM I, SCHMIDTLEIN C R, CAMPANELLA G, et al. DeepPET: a deep encoder–decoder network for directly solving the PET image reconstruction inverse problem [J]. Medical Image Analysis, 2019, 54: 253-262.
[14] ZHU B, LIU J Z, CAULEY S F, et al. Image reconstruction by domain-transform manifold learning [J]. Nature, 2018, 555: 487-492.
[15] PRIBANIĆ I, SIMIĆ S D, TANKOVIĆ N, et al. Reduction of SPECT acquisition time using deep learning: a phantom study [J]. Physica Medica, 2023, 111: 102615.
[16] 夏皓, 蔡念, 王平, 等. 基于多分辨率学习卷积神经网络的磁共振图像超分辨率重建[J]. 广东工业大学学报, 2020, 37(06): 26-31.
XIA H, CAI N, WANG P, et al. Magnetic resonance image super-resolution via multi-resolution learning [J]. Journal of Guangdong University of Technology, 2020, 37(06): 26-31.
[17] ZHANG Z, LIANG X, DONG X, et al. A sparse-view CT reconstruction method based on combination of DenseNet and deconvolution [J]. IEEE Transactions on Medical Imaging, 2018, 37(6): 1407-1417.
[18] XIE H, THORN S, LIU Y H, et al. Deep-learning-based few-angle cardiac SPECT reconstruction using transformer [J]. IEEE Transactions on Radiation and Plasma Medical Sciences, 2022, 7(1): 33-40.
[19] SHAO W, LEUNG K, POMPER M, et al. SPECT image reconstruction by a learnt neural network [J]. Journal of Nuclear Medicine, 2020, 61: 1478.
[20] GU J, YE J C. AdaIN-based tunable CycleGAN for efficient unsupervised low-dose CT denoising [J]. IEEE Transactions on Computational Imaging, 2021, 7: 73-85.
[21] TARVAINEN A, VALPOLA H. Mean teachers are better role models: weight-averaged consistency targets improve semi-supervised deep learning results[C]//GUYON I. Advances in Neural Information Processing Systems 30. Long Beach: NeurIPS, 2017, 30.
[22] YU L, WANG S, LI X, et al. Uncertainty-aware self-ensembling model for semi-supervised 3D left atrium segmentation[C]//SHEN D G. Medical Image Computing and Computer Assisted Intervention–MICCAI 2019. Shenzhen: Springer, 2019(11765) : 605-613.
[23] CHEPLYGINA V, DE BRUIJNE M, PLUIM J P W. Not-so-supervised: a survey of semi-supervised, multi-instance, and transfer learning in medical image analysis [J]. Medical Image Analysis, 2019, 54: 280-296.
[24] ZHANG K, LI Y, LIANG J, et al. Practical blind image denoising via Swin-Conv-UNet and data synthesis[J]. Machine Intelligence Research, 2023: 1-14.
[25] 甘孟坤, 曾安, 张小波. 基于Swin-Unet的主动脉再缩窄预测研究[J]. 广东工业大学学报, 2023, 40(5): 34-40.
GAN M K, ZENG A, ZHANG X B. Aortic re-coarctation prediction research based on Swin-Unet [J]. Journal of Guangdong University of Technology, 2023, 40(5): 34-40.
[26] RONNEBERGER O, FISCHER P, BROX T. U-net: convolutional networks for biomedical image segmentation[C]// NAVAB N. Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015. Munich: Springer, 2015: 234-241.
[27] SEGARS W P, BOND J, FRUSH J, et al. Population of anatomically variable 4D XCAT adult phantoms for imaging research and optimization [J]. Medical Physics, 2013, 40(4): 043701.
[1] 张灵, 李荣臻, 郑苏. 融合标签语义嵌入和图卷积的短文本特征扩展及分类方法[J]. 广东工业大学学报, 2024, 41(01): 69-78.
[2] 柴文光, 罗崇熙. 半监督遥感图像建筑物变化检测算法[J]. 广东工业大学学报, 2024, 41(0): 0-.
[3] 叶文权, 李斯, 凌捷. 基于多级残差U-Net的稀疏SPECT图像重建[J]. 广东工业大学学报, 2023, 40(01): 61-67.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
版权所有 © 2017 《广东工业大学学报》编辑部
地址：广州市东风东路729号广东工业大学 邮编：510090
电话：020-37626653,020-37626139,020-37626396
邮箱：xbzrb@gdut.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发