Deep progressive learning reconstruction for fast and low-dose whole-body PET scans in an integrated PET/MR system.

Authors

Jie Ding, Yihong Yang, Zhe Wang, Haiyan Wang, Na Qi, Zirong Zhou, Xing Chen, Zhiwen You, Chen Xi, Zheng Qian, Lingzhi Hu, Jianmin Yuan, Jun Zhao

Published in

BMC medical imaging. Jul 15, 2026. Epub Jul 15, 2026.

Abstract

PET/MR combines molecular and functional imaging but faces challenges such as prolonged scans, noise from reduced tracer activity, and suboptimal reconstruction methods (e.g., OSEM). Deep learning techniques, such as deep progressive learning (DPL), show promise in enhancing low-tracer activity PET/CT imaging but remain understudied for PET/MR. As MR acquisition continues to accelerate through technological advances, adaptive PET reconstruction methods will become increasingly important for maintaining overall throughput in integrated PET/MR protocols. This study evaluates DPL on a uPMR790 PET/MR system to determine its potential for reducing injected activity or scan durations in whole-body [¹⁸F]FDG PET/MR while preserving image quality.
In our study, we included 115 patients for whole-body PET/MR examinations, of whom 100 received a full injected activity of [¹⁸F]FDG, and 15 received injected activity with half the activity. The list mode PET data from the patients were reorganized to reconstruct PET raw data for each bed position for 420, 210, 140, and 105 s, simulating 1/n (n = 1, 2, 3, 4) of the acquisition time. PET reconstructions were performed using two different methods: ordered subset expectation maximization (OSEM) and DPL [divided into three levels (1, 2, 3)]. In terms of subjective assessment, we conducted a five-point Likert scale visual analysis. For quantitative assessment, we measured the standardized uptake value (SUV) and assessed image quality using the signal-to-noise ratio (SNR), contrast, contrast-to-noise ratio (CNR), coefficient of variation (COV), and target-to-background ratio (TBR).
According to subjective visual assessment methods, DPL scores were significantly higher than OSEM, with DPL1 achieving the highest scores. The advantage of DPL1 became more pronounced with shorter reconstruction times. In both the full injected activity and half injected activity groups, at one-quarter of the reconstruction time, the average Likert visual scores for OSEM and DPL1 were 2.36 vs. 4.01 in the full injected activity group and 1.23 vs. 3.21 in the half injected activity group, respectively (P < 0.01). Compared to the OSEM group, the DPL1 group exhibited significantly improved image quality with markedly reduced noise; the SNR, CNR, contrast, COV, and TBR values were all significantly higher than those of OSEM. These gains were still observable in obese patients within the DPL1 group. There were no significant differences in the lesion SUVmax and SUVmean measured between the OSEM and DPL1 groups (all P > 0.05); however, lesion volumes measured from DPL1 were smaller than those in the OSEM group.
DPL reconstruction suggests that it may be feasible to reduce the administered activity of [¹⁸F]FDG to half or decrease the scanning time to one-quarter in whole-body PET/MR imaging, while maintaining image quality. This approach is also advantageous for obese patients. However, further clinical task-based validation is needed before routine implementation.

PMID:
42458316
Bibliographic data and abstract were imported from PubMed on 16 Jul 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.