Related Experiment Video
Updated: Sep 28, 2025
11:09
High-Resolution Cardiac Positron Emission Tomography/Computed Tomography for Small Animals
Published on: December 16, 2022
3.9K
Heart and bladder detection and segmentation on FDG PET/CT by deep learning.
Xiaoyong Wang1, Skander Jemaa2, Jill Fredrickson2
1Genentech, Inc., South San Francisco, CA, USA. wang.xiaoyong@gene.com.
BMC Medical Imaging
|March 31, 2022
Summary
This study introduces an automated deep learning method to segment the heart and bladder in PET/CT scans, reducing false positives in tumor detection caused by physiological noise.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- 18fluorodeoxyglucose (FDG)-PET/CT is crucial for oncology, but physiological uptake in the heart and bladder causes false positives.
- Challenges include patient variability, differing PET/CT appearances, and proximity of tumors to these organs.
Purpose of the Study:
- To develop an automated deep learning approach for segmenting the heart and bladder in whole-body PET/CT scans.
- To eliminate physiological noise from FDG accumulation in the heart and bladder for improved tumor detection.
Main Methods:
- Utilized two separate 3D U-Net models for heart and bladder segmentation.
- Employed multi-modal input, combining PET and CT data for each network.
- Trained models on retrospective clinical trial data (575 PET/CT for heart, 538 for bladder).
Main Results:
- Achieved high accuracy on an independent test set with a Dice Similarity Coefficient (DSC) of 0.96 for heart and 0.95 for bladder.
- Demonstrated excellent spatial agreement with Average Surface Distance (ASD) of 0.44 mm for heart and 0.90 mm for bladder.
Conclusions:
- The proposed deep learning methodology effectively segments heart and bladder on PET/CT.
- This approach can be integrated into FDG-PET/CT processing to reduce noise and enhance automated tumor analysis.
