Noise reduction with cross-tracer and cross-protocol deep transfer learning for low-dose PET

Abstract

Previous studies have demonstrated the feasibility of reducing noise with deep learning-based methods for low-dose fluorodeoxyglucose (FDG) positron emission tomography (PET). This work aimed to investigate the feasibility of noise reduction for tracers without sufficient training datasets using a deep transfer learning approach, which can utilize existing networks trained by the widely available FDG datasets. In this study, the deep transfer learning strategy based on a fully 3D patch-based U-Net was investigated on a 18 F-fluoromisonidazole ( 18 F-FMISO) dataset using single-bed scanning and a 68 Ga-DOTATATE dataset using whole-body scanning. The datasets of 18 F-FDG by single-bed scanning and whole-body scanning were used to obtain pre-trained U-Nets separately for subsequent cross-tracer and cross-protocol transfer learning. The full-dose PET images were used as the labels while low-dose PET images from 10% counts were used as the inputs. Three types of U-Nets were obtained: a U-Net trained by FDG dataset, a pre-trained FDG U-Net fine-tuned by another less-available tracer (FMISO/DOATATE), and a U-Net completely trained by a large number of less-available tracer datasets (FMISO/DOATATE), used as the reference U-Net. The denoising performance of the three types of U-Nets was evaluated on single-bed 18 F-FMISO and whole-body 68 Ga-DOTATATE separately and compared using normalized root-mean-square error (NRMSE), signal-to-noise ratio (SNR), and relative bias of region of interest (ROI). For cross-tracer transfer learning, all the U-Nets provided denoised images with similar quality for both tracers. There was no significant difference in terms of NRMSE and SNR when comparing the former two U-Nets with the reference U-Net. The ROI biases for these U-Nets were similar. For cross-tracer and cross-protocol transfer learning, the pre-trained single-bed FDG U-Net fine-tuned by whole-body DOTATATE data provided the most consistent images with the reference U-Net. Fine-tuning significantly reduced the NRMSE and the ROI bias and improved the SNR when comparing the fine-tuned U-Net with the U-Net trained by single-bed FDG only (NRMSE: 96.3% ± 21.1% versus 120.6% ± 18.5%, ROI bias: −10.5% ± 13.0% versus −14.7% ± 6.4%, SNR: 4.2 ± 1.4 versus 3.9 ± 1.6, for fine-tuned U-Net and the U-Net trained by single-bed FDG, respectively, with p < 0.01 in all cases). This work demonstrated that it is feasible to utilize existing networks well-trained by FDG datasets to reduce the noise for other less-available tracers and other scanning protocols by using the fine-tuning strategy.

Generation of synthetic PET images of synaptic density and amyloid from 18F‐FDG images using deep learning

“…Visually, the predicted images appear to have less noise than the true images, and the predicted images are smoother than the true images. The smooth effect was also observed in other image denoising studies using the U‐Net model 32 33 …”

“…The smooth effect was also observed in other image denoising studies using the U-Net model. 32 The U-Net model minimized the squared error between the predicted image and the true image, which may cause a smooth effect during prediction. 33 The mean NRMSE, SSIM, and Pearson coefficient across all AD and CN subjects for the 18 F-FDG SUVR → 11 C-UCB-J SUVR network and 18 F-FDG K i Ratio → 11 C-UCB-J SUVR network are listed in Table 4.…”

“…The patch size of 32 × 32 × 32 was applied in the 11 C-UCB-J image generation, while 64 × 64 × 64 was applied in the 11 C-PiB SUVR image generation. We chose the patch size based on a pilot study 31,32 (methods in Supplemental Materials), where we found the performances of 32 × 32 × 32 and 64 × 64 × 64 patches were similar for 11 C-UCB-J, while 64 × 64 × 64 patches outperformed 32 × 32 × 32 patches for 11 C-PiB image generation(Figure S1). The Adam optimizer was applied with an initial learning rate of 0.001 and the learning decay rate of 0.96 by each epoch.…”

Generation of parametric Ki images for FDG PET using two 5‐min scans

“…This higher noise level in the dual‐time‐point K i images may degrade the lesion detectability when compared to the standard K i images. In future studies, we will investigate and implement state‐of‐the‐art denoising methods, such as anatomical‐guided median non‐local mean filter 56 and deep learning based denoizing methods 57,58 on the dual‐time‐point K i images to improve image quality and noise. The lesion detectability will be further investigated on the denoized dual‐time‐point K i images.…”

Applications of artificial intelligence and deep learning in molecular imaging and radiotherapy

“…So far, most of the studies were conducted using conventional tracers, such as 18 F-FDG. It appears that neural networks developed/trained using specific tracers and/or acquisition protocols are applicable to other radiotracers and protocols but further research is required to support this hypothesis (Liu et al 2019c ).…”