Low-count 18F-FDG PET/CT using a deep-learning reconstruction algorithm on a 35-cm axial field-of-view digital PET system: a prospective clinical evaluation
Abstract
Background/Introduction To determine how far injected activity can be reduced in clinical 18F-FDG PET/CT while preserving clinical interpretability using a deep progressive reconstruction (DPR) algorithm on one of Europe’s first 35-cm axial field-of-view (AFOV) digital PET/CT systems. Methods One hundred fifty consecutive patients undergoing vertex-to-mid-thigh 18F-FDG PET/CT were prospectively included. All acquisitions were performed at 3 MBq/kg with an acquisition time of 120 s/bed. List-mode data were retrospectively truncated to simulate shorter acquisition durations of 90, 60, and 30 s/bed (corresponding to 75%, 50%, and 25% of the total counts, respectively). Images were reconstructed using EARL-2–compliant TOF-OSEM and DPR. Two blinded physicians scored image quality on a 5-point Likert scale; clinical interpretability was defined as a Likert score ≥ 3. SUVpeak, SUVmax, signal-to- noise ratio (SNR), and lesion contrast-to-noise ratio (CNR) were measured using propagated volumes of interest. Results DPR yielded higher qualitative image quality than EARL-2 OSEM at all count levels. At 60 s/bed (∼50% counts), DPR remained clinically interpretable, whereas matched-count EARL-2 OSEM showed marked image degradation. Lesion CNR was higher with DPR across all simulated count levels. SUVpeak was largely preserved in reference tissues and increased modestly in lesions, whereas SUVmax showed larger reconstruction-dependent increases. Conclusion This prospective clinical evaluation demonstrates that, on a 35-cm AFOV digital PET/CT system, DPR enables an approximately 50% reduction in effective count statistics (60 s/bed or 1.5 MBq/kg under linear scaling assumptions) without compromising clinical interpretability. Quantitative SUV measurements, particularly SUVmax, remain reconstruction dependent.
Related
- Evaluation of an Artificial Intelligence and Machine Learning Algorithm to Enhance Angiographic Coronary Visibility in Morbid Obesity: A Retrospective Observational Study
- inClinico Agent for clinical trial endpoint prediction and prospective evaluation
- Pulmonary Vein Isolation with Different Single-Shot Ablation Devices in Patients with Atrial Fibrillation: A Prospective Observational Single-Center Clinical Evaluation
- Evaluation of the Efficacy and Safety of Combination Therapy of Vamha and Myrha in the Management of PMOS: An Open-Label, Randomized, Multicentre, Comparative, Prospective Clinical Study
- Real-time Evaluation and Anticipation of Causal disTillation early prediction for cardiac surgery-associated acute kidney injury (REACT-01): study protocol for a single-center prospective randomized clinical trial
BibTeX
@article{atieh2026lowcount,
title = {{Low-count 18F-FDG PET/CT using a deep-learning reconstruction algorithm on a 35-cm axial field-of-view digital PET system: a prospective clinical evaluation}},
author = {Michel Atieh and Kevin Dylan Pushparajah and Jérôme Ghidaglia and Maria-Angéla Castilla-Lièvre and Daniel Deliu and Virginie Massaria and Kevin Harnais and Agnieszka Buffon and Maiwenn Beau and Yannick Jentreau and Natacha da Fonseca and Stephane Onephandara and Widad Makhlouf and Aude Amoin and Christopher Boyeau and Karim En Nourhi and Solène Marie and Emmanuel Durand and Yolaine Suc and Sylvain Sans and Ahmadreza Rezaei and Stewart Young and Florent L Besson},
journal = {EANM Innovation},
year = {2026},
month = sep,
doi = {10.1016/j.eanmi.2026.100280},
url = {https://hal.science/hal-05695203v2},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Low-count 18F-FDG PET/CT using a deep-learning reconstruction algorithm on a 35-cm axial field-of-view digital PET system: a prospective clinical evaluation
AU - Michel Atieh
AU - Kevin Dylan Pushparajah
AU - Jérôme Ghidaglia
AU - Maria-Angéla Castilla-Lièvre
AU - Daniel Deliu
AU - Virginie Massaria
AU - Kevin Harnais
AU - Agnieszka Buffon
AU - Maiwenn Beau
AU - Yannick Jentreau
AU - Natacha da Fonseca
AU - Stephane Onephandara
AU - Widad Makhlouf
AU - Aude Amoin
AU - Christopher Boyeau
AU - Karim En Nourhi
AU - Solène Marie
AU - Emmanuel Durand
AU - Yolaine Suc
AU - Sylvain Sans
AU - Ahmadreza Rezaei
AU - Stewart Young
AU - Florent L Besson
PY - 2026
DA - 2026/09/01
JO - EANM Innovation
DO - 10.1016/j.eanmi.2026.100280
UR - https://hal.science/hal-05695203v2
AB - Background/Introduction To determine how far injected activity can be reduced in clinical 18F-FDG PET/CT while preserving clinical interpretability using a deep progressive reconstruction (DPR) algorithm on one of Europe’s first 35-cm axial field-of-view (AFOV) digital PET/CT systems. Methods One hundred fifty consecutive patients undergoing vertex-to-mid-thigh 18F-FDG PET/CT were prospectively included. All acquisitions were performed at 3 MBq/kg with an acquisition time of 120 s/bed. List-mode data were retrospectively truncated to simulate shorter acquisition durations of 90, 60, and 30 s/bed (corresponding to 75%, 50%, and 25% of the total counts, respectively). Images were reconstructed using EARL-2–compliant TOF-OSEM and DPR. Two blinded physicians scored image quality on a 5-point Likert scale; clinical interpretability was defined as a Likert score ≥ 3. SUVpeak, SUVmax, signal-to- noise ratio (SNR), and lesion contrast-to-noise ratio (CNR) were measured using propagated volumes of interest. Results DPR yielded higher qualitative image quality than EARL-2 OSEM at all count levels. At 60 s/bed (∼50% counts), DPR remained clinically interpretable, whereas matched-count EARL-2 OSEM showed marked image degradation. Lesion CNR was higher with DPR across all simulated count levels. SUVpeak was largely preserved in reference tissues and increased modestly in lesions, whereas SUVmax showed larger reconstruction-dependent increases. Conclusion This prospective clinical evaluation demonstrates that, on a 35-cm AFOV digital PET/CT system, DPR enables an approximately 50% reduction in effective count statistics (60 s/bed or 1.5 MBq/kg under linear scaling assumptions) without compromising clinical interpretability. Quantitative SUV measurements, particularly SUVmax, remain reconstruction dependent.
SN - 3051-2913
ER -