Deep Generalized Learning Model for PET Image Reconstruction.

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.

Improved accuracy of the biodistribution and internal radiation dosimetry of 13 N-ammonia using a total-body PET/CT scanner.

BACKGROUND: Conventional short-axis PET typically utilizes multi-bed multi-pass acquisition to produce quantitative whole-body dynamic images and cannot record all the uptake information simultaneously, resulting in errors when fitting the time-activity curves (TACs) and calculating radiation doses. PURPOSE: The aim of this study is to evaluate the 13 N-ammonia biodistribution and the internal radiation doses using a 194 cm long total-body PET/CT scanner (uEXPLORER), and make a comparison with the previous short-axis PET results. METHODS: Ten subjects (age 40-74 years) received 13 N-NH3 injection (418.1-670.81 MBq) and were under a dynamic scan for about 60 min with using a 3-dimensional whole-body protocol. ROIs were drawn visually on 11 major organs (brain, thyroid, gallbladder, heart wall, kidneys, liver, pancreas, spleen, lungs, bone marrow, and urinary bladder content) for each subject. TACs were generated using Pmod and the absorbed radiation doses were calculated using Olinda 2.2. To compare with the conventional PET/CT, five points were sampled on uEXPLORER's TACs to mimic the result of a short-axis PET/CT (15 cm axial FOV, consisted of 9 or 10 bed positions). Then the TACs were obtained using the multi-exponential fitting method, and the residence time and radiation dose were also calculated and compared with uEXPLORER. RESULTS: The highest absorbed organ doses were the pancreas, thyroid, spleen, heart wall, and kidneys for the male. For the female, the first five highest absorbed organ dose coefficients were the pancreas, heart wall, spleen, lungs, and kidneys. The lowest absorbed dose was found in red marrow both for male and female. The simulated short-axis PET can fit TACs well for the gradually-changed uptake organs but typically underestimated for the rapid-uptake organs during the first-10 min, resulting in errors in the calculated radiation dose. CONCLUSION: uEXPLORER PET/CT can measure 13 N-ammonia's TACs simultaneously in all organs of the whole body, which can provide more accurate biodistribution and radiation dose estimation compared with the conventional short-axis scanners.