Torso synthetic CT generation by integrating deep learning and segmentation for FDG-PET/MR attenuation correction.
Jin Uk Heo1,2, Shujin Sun3, Robert S Jones1,4
1Department of Radiology, Case Western Reserve University, Cleveland, OH, 44106, United States of America.
Biomedical Physics & Engineering Express
|October 16, 2025
Summary
This study introduces a novel deep learning method for accurate attenuation correction in PET/MR imaging. The method generates synthetic CT from MR data, improving SUV accuracy and outperforming previous techniques.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Radiology
Background:
- Positron Emission Tomography/Magnetic Resonance (PET/MR) offers advantages over PET/CT, including simultaneous acquisition and superior soft tissue contrast.
- Accurate attenuation correction (AC) for PET/MR is challenging because MR signals do not directly correlate with attenuation.
- Deep learning offers a promising approach to generate synthetic CT (sCT) from MR data for AC.
Purpose of the Study:
- To develop and validate a novel deep learning method for accurate AC in torso FDG-PET/MR imaging.
- To generate an AC map for the entire torso using Dixon MR images, merging deep learning with threshold-based segmentation.
- To compare the performance of the novel method against previously published AC techniques.
Main Methods:
- Utilized 29 prospectively collected, paired FDG-PET/CT and MR datasets for training and validation.
- Employed a U-net Residual Network conditional Generative Adversarial Network integrated with tissue segmentation (URcGANmod) for sCT generation from Dixon MR data.
- Focused on torso AC (base of skull to mid-thigh) and compared performance to four existing methods.
Main Results:
- The URcGANmod generated accurate torso sCT with a mean absolute difference of 32 ± 4 HU per voxel.
- When applied for AC, SUV differences were within 4.4% compared to CT-based AC, with excellent reproducibility (<3.5% standard deviation).
- The novel method demonstrated superior accuracy and precision for MR-based AC (MRAC) compared to four other methods.
Conclusions:
- Combining deep learning and segmentation significantly enhances MRAC accuracy in torso FDG-PET/MR.
- The proposed method achieves less than 4.4% SUV error, improving quantitative accuracy throughout the torso.
- The method's accuracy and precision warrant further investigation in multicenter trials for quantitative longitudinal studies.
Related Concept Videos
01:10Computed Tomography
Tomography refers to imaging by sections. Computed tomography (CT) is a non-invasive imaging technique that uses computers to analyze several cross-sectional X-rays to reveal minute details about structures in the body.
The technique was invented in the 1970s and is based on the principle that as X-rays pass through the body, they are absorbed or reflected at different levels. In the technique, a patient lies on a motorized platform while a computerized axial tomography (CAT) scanner rotates...
The technique was invented in the 1970s and is based on the principle that as X-rays pass through the body, they are absorbed or reflected at different levels. In the technique, a patient lies on a motorized platform while a computerized axial tomography (CAT) scanner rotates...
01:29Positron Emission Tomography
Positron emission tomography (PET) is a medical imaging technique involving radiopharmaceuticals — substances that emit short-lived radiation. Although the first PET scanner was introduced in 1961, it took 15 more years before radiopharmaceuticals were combined with the technique and revolutionized its potential.
One of the main requirements of a PET scan is a positron-emitting radioisotope, which is produced in a cyclotron and then attached to a substance used by the part of the body being...
One of the main requirements of a PET scan is a positron-emitting radioisotope, which is produced in a cyclotron and then attached to a substance used by the part of the body being...