Combining the Segment Anything Model with a deep learning classifier produced an efficient pipeline for classifying MRI ovarian lesions, improving segmentation speed and matching radiologists’ classification performance.
27 October 2023 in European Journal of Nuclear Medicine and Molecular Imaging
Purpose: Differentiating lymph node metastases (LNM) from peripheral ganglia by physiological prostate-specific membrane antigen (PSMA) uptake is challenging. Two tracers (68Ga-PSMA-11 and 18F-fluorodeoxyglucose [FDG]) metabolic uptake patterns were evaluated by positron emission tomography-computed tomography (PET-CT), searching for differences that could tell ganglia from LNM.Methods: Dual 68Ga-PSMA-11 and 18F-FDG PET-CT data of 138 prostate cancer patients acquired from June 2018 to December 2019 were retrospectively evaluated. Ganglia and LNM with PSMA-11 uptake above local background were analyzed by the location and PSMA-11-PET and FDG-PET maximum standardized uptake value (SUVmax).Results: PSMA-11-positive ganglia (n = 381) and LNM (n = 83) were identified in 138 and 58 patients, respectively. The LNM SUVmax of PSMA-11-PET (16.4 ± 14.8 vs 2.3 ± 0.7, P < 0.001) and FDG-PET (3.3 ± 3.2 vs 1.5 ± 0.5, P < 0.001) were higher than in ganglia. The probabilities of being an LNM in the low-potential (PSMA-11-PET SUVmax of <4.1 and FDG-PET SUVmax of 4.1 and FDG-PET SUVmax of <2.05, or PSMA-11-PET SUVmax of 2.05), and high-potential (PSMA-11-PET SUVmax of >4.1 and FDG-PET SUVmax of >2.05) groups were 0.9% (3/334), 44.6% (37/83), and 91.5% (43/47), respectively (P < 0.001). The cervical and coeliac ganglia had higher PSMA-11 and FDG uptake than the sacral ganglia (P < 0.001 for all). LNM PSMA-11 and FDG uptake was similar in these three locations.Conclusion: The FDG-PET and PSMA-11-PET SUVmax, especially when combined, could well differentiate LNM from ganglia. The tracers uptake differed between cervical/coeliac and sacral ganglia, so the lesion location should be considered during image assessment.
7 May 2025 in IEEE Journal of Biomedical and Health Informatics
Fast PET imaging is clinically important for reducing motion artifacts and improving patient comfort. While recent diffusion-based deep learning methods have shown promise, they often fail to capture the true PET degradation process, suffer from accumulated inference errors, introduce artifacts, and require extensive reconstruction iterations. To address these challenges, we propose a novel multistage diffusion framework tailored for fast PET imaging. At the coarse level, we design a multistage structure to approximate the temporal non-linear PET degradation process in a data-driven manner, using paired PET images collected under different acquisition duration. A Phase Error Correction Network (PECNet) ensures consistency across stages by correcting accumulated deviations. At the fine level, we introduce a deterministic cold diffusion mechanism, which simulates intra-stage degradation through interpolation between known acquisition durations—significantly reducing reconstruction iterations to as few as 10. Evaluations on [68Ga]FAPI and [18F]FDG PET datasets demonstrate the superiority of our approach, achieving peak PSNRs of 36.2 dB and 39.0 dB, respectively, with average SSIMs over 0.97. Our framework offers high-fidelity PET imaging with fewer iterations, making it practical for accelerated clinical imaging.