Related Experiment Video
Updated: Oct 23, 2025
10:30
Whole Body and Regional Quantification of Active Human Brown Adipose Tissue Using 18F-FDG PET/CT
Published on: April 1, 2019
9.1K
Low-count whole-body PET with deep learning in a multicenter and externally validated study.
Akshay S Chaudhari1,2,3, Erik Mittra4, Guido A Davidzon5
1Department of Radiology, Stanford University, Palo Alto, CA, USA. akshaysc@stanford.edu.
NPJ Digital Medicine
|August 24, 2021
Summary
Deep learning enhances positron emission tomography (PET) scans, maintaining diagnostic quality with reduced radiation dose or scan time. This AI approach improves safety and lowers costs for PET imaging.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Positron emission tomography (PET) imaging use is limited by cost and radiation.
- Reducing scan time or radiotracer dose degrades image quality.
- Deep-learning reconstruction shows promise but lacks real-world clinical validation.
Purpose of the Study:
- Evaluate a deep-learning algorithm for enhancing low-count PET scans.
- Assess performance in a multi-center, multi-vendor clinical oncology setting.
- Determine generalizability across different PET scanners and institutions.
Main Methods:
- Applied deep learning to fourfold reduced-count whole-body PET scans.
- Evaluated image quality and diagnostic confidence with blinded readers.
- Assessed lesion detection, SUV quantification, and interscan agreement.
- Validated the method on an external dataset from multiple institutions and scanner types.
Main Results:
- Low-count-enhanced PET scans were non-inferior to standard scans in diagnostic image quality and confidence.
- Achieved high sensitivity (0.94) and specificity (0.98) for lesion detection.
- SUV quantification was comparable, with high correlation (CCC ≥ 0.94).
- Interscan variations were lower than intra- and inter-reader variations.
Conclusions:
- Deep learning restores diagnostic quality and SUV accuracy for reduced-count PET scans.
- The method generalizes across multiple institutions and scanner types.
- Enables potential reduction in PET radiation dose or exam duration, improving safety and cost-effectiveness.
