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

Abstract

Image processing plays a crucial role in maximising diagnostic quality of positron emission tomography (PET) images. Recently, deep learning methods developed across many fields have shown tremendous potential when applied to medical image enhancement, resulting in a rich and rapidly advancing literature surrounding this subject. This review encapsulates methods for integrating deep learning into PET image reconstruction and post-processing for low-dose imaging and resolution enhancement. A brief introduction to conventional image processing techniques in PET is firstly presented. We then review methods which integrate deep learning into the image reconstruction framework as either deep learning-based regularisation or as a fully data-driven mapping from measured signal to images. Deep learning-based post-processing methods for low-dose imaging, temporal resolution enhancement and spatial resolution enhancement are also reviewed. Finally, the challenges associated with applying deep learning to enhance PET images in the clinical setting are discussed and future research directions to address these challenges are presented.

PET image enhancement using artificial intelligence for better characterization of epilepsy lesions

“…To train our model we had to overcome the limited availability of high quality training data, a common challenge for the deblurring problem ( 53 ) and so we chose to use simulation. We developed a pipeline based on a Monte-Carlo based PET simulator as it can accurately model the PET acquisition process including physical effects resulting in realistic sinograms ( 29 ) that have the same data distribution as the real PET.…”

“…We therefore consider that the clinical data application was successful in illustrating proof-of-concept that a model trained on Monte-Carlo simulated PET data is applicable on real data. Whereas, generalizability to out-of-distribution data is a common critical limiting factor for deep learning-based image processing ( 53 ), in our case this limitation could in principle be overcome by creating more simulations using the Monte-Carlo pipeline with settings tuned ( 29 , 31 ) to simulate different scanners and reconstructions. Nevertheless, a study of generalization exceeds the scope of this manuscript.…”

Image-based estimation of the left ventricular cavity volume using deep learning and Gaussian process with cardio-mechanical applications

“…This dataset contains manual segmentations of different elements within the CMR images, including the myocardium, the LV cavity, and the right ventricle at the end of diastole and the end of systole. The structure of the deep learning model that we have used for segmentation is known as the U-net, which has been used repeatedly for medical image segmentation and resolution enhancement in the literature (Li et al, 2018;Zhou et al, 2018;Ronneberger et al, 2015;Ankenbrand et al, 2021;Pain et al, 2022;Rabbani and Babaei, 2022;Rabbani et al, 2022a). To improve the accuracy of segmentation, as suggested in the literature (Xia and Kulis, 2017), we have used two U-net structures in a sequential manner as depicted in Fig.…”

Artificial intelligence-based 68Ga-DOTATOC PET denoising for optimizing 68Ge/68Ga generator use throughout its lifetime

“…However, no clinical study has demonstrated the value of this software to enhance the quality of low-dose 68 Ga PET images, even though nuclear medicine departments are concerned about this issue. Various other deep learning-based methods have been evaluated for low-dose imaging and resolution enhancement, but none of them are currently validated for clinical use ( 14 ). Denoising techniques for 68 Ga-labeled radiotracers in PET imaging have been explored using both reconstruction-based methods and deep-learning techniques.…”