Improved low-dose positron emission tomography image reconstruction using deep learned prior

Abstract

Positron emission tomography (PET) is a promising medical imaging technology that provides non-invasive and quantitative measurement of biochemical process in the human bodies. PET image reconstruction is challenging due to the ill-poseness of the inverse problem. With lower statistics caused by the limited detected photons, low-dose PET imaging leads to noisy reconstructed images with much quality degradation. Recently, deep neural networks (DNN) have been widely used in computer vision tasks and attracted growing interests in medical imaging. In this paper, we proposed a maximum a posteriori (MAP) reconstruction algorithm incorporating a convolutional neural network (CNN) representation in the formation of the prior. Rather than using the CNN in post-processing, we embedded the neural network in the reconstruction framework for image representation. Using the simulated data, we first quantitatively evaluated our proposed method in terms of the noise-bias tradeoff, and compared with the filtered maximum likelihood (ML), the conventional MAP, and the CNN post-processing methods. In addition to the simulation experiments, the proposed method was further quantitatively validated on the acquired patient brain and body data with the tradeoff between noise and contrast. The results demonstrated that the proposed CNN-MAP method improved noise-bias tradeoff compared with the filtered ML, the conventional MAP, and the CNN post-processing methods in the simulation study. For the patient study, the CNN-MAP method achieved better noise-contrast tradeoff over the other three methods. The quantitative enhancements indicate the potential value of the proposed CNN-MAP method in low-dose PET imaging.

Artificial intelligence guided enhancement of digital PET: scans as fast as CT?

“…Some approaches for whole-body low-count imaging enhancement do not implement GANs [ 28 , 48 , 49 ]. Compared to the few previous works that use GANs on whole-body PET images, in our study the acquisition time was shorter by about one magnitude, and the number of included patients was significantly larger.…”

“…We therefore selected this model for detailed evaluation. In most previous works, only PET data were used as input [ 27 – 29 , 48 , 49 ]. However, our results are in line with a PET/MRI study that describes benefits by simultaneous input of PET and MRI images for enhancement of ultra-low-dose PET images in children [ 35 ].…”

“…In our study, ExtremePET data were used that were acquired in a separate ultra-short scan, whereas most previous studies use simulated short acquisition time data that are created by undersampling of original list-mode data sets [ 29 , 35 , 48 , 49 ]. Sanaat et al also use separately acquired low-dose PET data [ 27 ].…”

Deep learning-based image reconstruction and post-processing methods in positron emission tomography for low-dose imaging and resolution enhancement

“…The linear fit reduces bias by ensuring mean pixel values within small regions are consistent with the original input. Similar work by Wang et al [ 66 ] used a relative difference in the regularisation with no local linear fitting. Lv et al [ 67 ] developed a formulation which integrated two neural networks into the MAP framework: an initial denoising network trained to map low count images to full count images and a subsequent image enhancement network trained to map reconstructions with low iterations to high iterations.…”

FBP‐CNN: A Direct PET Image Reconstruction Network for Flow Visualization

“…The original images reconstructed using traditional algorithms are used as network inputs to obtain higher-quality PET images through methods such as denoising and super-resolution imaging. [16][17][18][19][20][21] Some researchers have combined neural networks and iterative algorithms that can reconstruct images with more detail. [22,23] Although these methods are easy to implement, the final reconstruction results are vulnerable to traditional algorithms, especially many traditional algorithms that require manual selection of regular terms to ensure the quality of reconstructed images.…”