Anatomical-guided attention enhances unsupervised PET image denoising performance

Although supervised convolutional neural networks (CNNs) often outperform conventional alternatives for denoising positron emission tomography (PET) images, they require many low-and high-quality reference PET image pairs. Herein, we propose an unsupervised 3D PET image denoising method based on an anatomical information-guided attention mechanism. The proposed magnetic resonance-guided deep decoder (MR-GDD) utilizes the spatial details and semantic features of MR-guidance im-age more effectively by introducing encoder-decoder and deep decoder subnetworks. Moreover, the spe-cific shapes and patterns of the guidance image do not affect the denoised PET image, because the guidance image is input to the network through an attention gate. In a Monte Carlo simulation of [F-18]fluoro-2-deoxy-D-glucose (FDG), the proposed method achieved the highest peak signal-to-noise ratio and structural similarity (27.92 +/- 0.44 dB/0.886 +/- 0.007), as compared with Gaussian filtering (26.68 +/- 0.10 dB/0.807 +/- 0.004), image guided filtering (27.40 +/- 0.11 dB/0.849 +/- 0.003), deep image prior (DIP) (24.22 +/- 0.43 dB/0.737 +/- 0.017), and MR-DIP (27.65 +/- 0.42 dB/0.879 +/- 0.007). Furthermore, we experimentally visualized the behavior of the optimization process, which is often unknown in un-supervised CNN-based restoration problems. For preclinical (using [F-18]FDG and [C-11]raclopride) and clin-ical (using [F-18]florbetapir) studies, the proposed method demonstrates state-of-the-art denoising per-formance while retaining spatial resolution and quantitative accuracy, despite using a common network architecture for various noisy PET images with 1/10th of the full counts. These results suggest that the proposed MR-GDD can reduce PET scan times and PET tracer doses considerably without impacting pa-tients. (C) 2021 Elsevier B.V. All rights reserved.