Zhang Qiyang, Hu Yingying, Zhao Yumo, Cheng Jing, Fan Wei, Hu Debin, Shi Fuxiao, Cao Shuangliang, Zhou Yun, Yang Yongfeng, Liu Xin, Zheng Hairong, Liang Dong, Hu Zhanli
IEEE Trans Med Imaging. 2024 Jan;43(1):122-134. doi: 10.1109/TMI.2023.3293836. Epub 2024 Jan 2.
Low-count positron emission tomography (PET) imaging is challenging because of the ill-posedness of this inverse problem. Previous studies have demonstrated that deep learning (DL) holds promise for achieving improved low-count PET image quality. However, almost all data-driven DL methods suffer from fine structure degradation and blurring effects after denoising. Incorporating DL into the traditional iterative optimization model can effectively improve its image quality and recover fine structures, but little research has considered the full relaxation of the model, resulting in the performance of this hybrid model not being sufficiently exploited. In this paper, we propose a learning framework that deeply integrates DL and an alternating direction of multipliers method (ADMM)-based iterative optimization model. The innovative feature of this method is that we break the inherent forms of the fidelity operators and use neural networks to process them. The regularization term is deeply generalized. The proposed method is evaluated on simulated data and real data. Both the qualitative and quantitative results show that our proposed neural network method can outperform partial operator expansion-based neural network methods, neural network denoising methods and traditional methods.
低计数正电子发射断层扫描（PET）成像具有挑战性，因为这个逆问题是不适定的。先前的研究表明，深度学习（DL）有望提高低计数PET图像质量。然而，几乎所有数据驱动的DL方法在去噪后都会出现精细结构退化和模糊效应。将DL纳入传统的迭代优化模型可以有效提高其图像质量并恢复精细结构，但很少有研究考虑模型的完全松弛，导致这种混合模型的性能没有得到充分利用。在本文中，我们提出了一个深度学习与基于交替方向乘子法（ADMM）的迭代优化模型深度集成的学习框架。该方法的创新之处在于我们打破了保真度算子的固有形式，并使用神经网络对其进行处理。正则化项得到了深度推广。所提出的方法在模拟数据和真实数据上进行了评估。定性和定量结果均表明，我们提出的神经网络方法优于基于部分算子扩展的神经网络方法、神经网络去噪方法和传统方法。
