Motion Artifacts in PET Scans: How AI Can Help
"New research reveals how data-driven AI can detect and compensate for patient movement during coronary PET imaging, leading to more accurate diagnoses."
Medical imaging, particularly Positron Emission Tomography (PET), relies on patients remaining still for extended periods. However, even slight movements can blur images, creating artifacts that compromise the accuracy of diagnoses. This is especially critical in coronary imaging, where precise detection of micro-calcification is crucial.
Coronary PET imaging often requires scans lasting up to 30 minutes, increasing the likelihood of patient movement. These movements, referred to as gross patient motion (GPM), can significantly degrade image quality and affect quantitative assessments. Consequently, there is a growing need for methods that can automatically detect and compensate for GPM during PET scans.
A recent study published in the Journal of Nuclear Medicine investigates a novel data-driven approach for detecting and correcting patient motion during coronary 18F-NaF PET imaging. This method leverages artificial intelligence to analyze data and compensate for movement, leading to more accurate and reliable results. Let's explore this technology and what it means for the future of cardiac PET imaging.
How Motion Correction Affects PET Images
A 2025 study reported that data-driven motion-corrected PET reconstructions produced higher standardized uptake values (SUVs) and/or contrast-to-noise ratios for tracer-avid lesions than ungated or end-expiratory-gated reconstructions. The corrected images were of similar or superior quality and sometimes made additional tracer-avid lesions identifiable. A separate study describes how respiratory motion correction differs from total-body dynamic motion correction: respiration is mostly periodic, allowing a limited set of motion fields to be estimated from a static scan using existing vendor software.
AI to the Rescue: How Motion Detection Works
The cornerstone of this innovative approach is a data-driven algorithm designed to track patient movement by monitoring the center-of-mass (CoM) of count rates within the PET data. This technique analyzes data every 200 milliseconds, enabling it to capture even subtle shifts in patient position. The study focused on two primary types of patient motion:
- Sudden Repositioning (SR): Abrupt movements exceeding 0.5 mm within 3 seconds.
- General Repositioning (GR): Gradual drifts greater than 0.3 mm over 15 seconds or more.
Triple-Gated PET and Motion Validation
Research on coronary 18F-NaF PET has examined combining cardiac-motion correction with end-diastolic imaging and correction for gross patient motion, while noting that earlier approaches did not account for respiratory motion or test scan-rescan reproducibility. A separate anatomical validation study compared automatic respiratory-motion correction in coronary PET using FusionQuant with expert measurements from four-dimensional CT. It reported similar respiratory motion in two directions and in 3D between the CT measurements and the automatic PET correction.
Motion Correction Still Has Technical Limits
A review of PET motion correction in integrated PET/MR identifies challenges in modeling aperiodic or unpredictable motion and achieving adequate temporal resolution for dynamic kinetic modeling. It also notes that MR availability may be limited during diagnostic acquisitions. In whole-body PET, motion can misalign transmission and emission scans, causing incorrect attenuation-correction factors to be applied. Long PET neuroimaging acquisitions also remain vulnerable to motion artifacts that compromise quantitative accuracy.
Comparing PET Motion-Correction Approaches
A 2012 comparison paper describes three motion-correction methods and distinguishes approaches that model rigid-body motion from those that model non-rigid-body motion. It notes that motion degrades image quality and introduces bias, a concern for high-resolution scanners including the mCT, mMR, and HRRT. Another study states that evidence has been insufficient to determine which of the two main PET motion-correction approaches is better in practice, and sets out to compare their convergence and image properties. The supplied sources therefore describe methods and comparison criteria but do not establish a single best approach.
The Future of Cardiac PET Imaging
The study's findings suggest that GPM is a common occurrence during coronary 18F-NaF PET imaging and can significantly affect the accuracy of quantitative assessments. By implementing an automated motion compensation technique, clinicians can potentially improve diagnostic accuracy and reduce the risk of misdiagnosis.
Deep Learning for Brain PET Motion
Head movement during long PET scan sessions degrades reconstruction quality and introduces artifacts that can limit clinical diagnosis and treatment. The cited Cross-Attention for Improved Motion Correction in Brain PET paper describes recent deep learning-based work on this problem. The available source snippet does not report performance figures or clinical validation results, so it supports the motivation and research direction rather than a conclusion about effectiveness.
PET Market and Technology Outlook
IMV's 226-page PET Imaging Market Outlook Report 2026 covers scan volumes, the PET/CT installed base, radiopharmaceutical trends, theranostics, AI adoption, and purchase plans through 2028. A 2025 overview describes PET technology advancements as supporting more accurate diagnoses and targeted treatment plans. These sources point to continued attention to AI adoption and molecular-imaging technology, but the supplied material gives no quantified forecast for motion-correction adoption or its future clinical impact.
Clinical Examples of Motion Correction
A 2015 clinical PET/MR report presented four case examples using 18F-FDG and 68Ga-DOTATATE, with improved image sharpness and fewer respiratory artifacts in scans involving pancreatic, liver, and lung lesions, as well as cardiac scans. A 2022 study of data-driven rigid motion correction for brain PET reported that head motion can reduce diagnostic value and impair quantitation; the approach was described as producing motion estimates with accuracy below 1 mm and temporal resolution of at least 1 Hz. Another study introduced an elastic motion-correction technique based on mass preservation for cardiac and respiratory motion, which its authors identify as contributors to inaccurate PET quantification and lesion characterization.
This AI-driven approach offers a practical solution for addressing motion-related artifacts in PET imaging. It doesn't require any additional hardware or modifications to existing scanning protocols, making it easily adaptable to current clinical workflows. The retrospective nature of the method means it can also be applied to previously acquired data, unlocking new insights from past studies.
As AI continues to evolve, its applications in medical imaging will likely expand, leading to more precise, efficient, and personalized healthcare. This study represents a significant step toward harnessing the power of AI to improve the accuracy and reliability of cardiac PET imaging, ultimately benefiting patients and clinicians alike.