Related Experiment Video
Updated: Jan 15, 2026
05:41
A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
Published on: February 6, 2020
9.8K
PET Head Motion Estimation Using Supervised Deep Learning With Attention.
IEEE Transactions on Medical Imaging
|October 13, 2025
Summary
Deep learning head motion correction (DL-HMC++) accurately predicts head movement in brain PET scans using raw data. This method significantly reduces motion artifacts, improving image quality and quantitative analysis for neurological disorder diagnosis.
Area of Science:
- Medical Imaging
- Neuroscience
- Artificial Intelligence
Background:
- Head movement in brain PET imaging causes artifacts and quantification errors.
- Hardware-based motion tracking (HMT) is not always practical in clinical settings.
Purpose of the Study:
- To develop and evaluate a deep learning approach (DL-HMC++) for head motion correction in PET imaging.
- To enable accurate quantitative analysis and diagnosis of neurological disorders.
Main Methods:
- A deep learning model (DL-HMC++) was trained using supervised learning on dynamic PET scans with HMT data.
- The model predicts rigid head motion from one-second 3D PET raw data.
- Evaluation was performed on two PET scanners and four radiotracers.
Main Results:
- DL-HMC++ outperformed existing data-driven motion estimation methods.
- Generated motion-free images with clear brain structure delineation and minimal artifacts.
- Quantitative analysis showed minimal differences compared to gold-standard HMT (1.2±0.5% on HRRT, 0.5±0.2% on mCT).
Conclusions:
- DL-HMC++ offers effective and generalizable data-driven PET head motion correction.
- This approach removes the need for HMT, making motion correction more accessible in clinical practice.
- DL-HMC++ has the potential to improve diagnostic accuracy for neurological conditions.
