Related Experiment Video
Updated: Nov 11, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
Automated cartilage segmentation and quantification using 3D ultrashort echo time (UTE) cones MR imaging with deep
Yan-Ping Xue1,2, Hyungseok Jang1, Michal Byra1
1Department of Radiology, University of California San Diego, 9452 Medical Center Drive, La Jolla, CA, 92037, USA.
European Radiology
|March 30, 2021
Summary
This study introduces an automated method using 3D UTE cones MRI and U-Net CNN for cartilage segmentation and quantitative mapping. The approach accurately assesses cartilage in osteoarthritis, offering a reliable tool for clinical evaluation.
Area of Science:
- Biomedical Imaging
- Artificial Intelligence in Medicine
- Osteoarthritis Research
Background:
- Accurate assessment of articular cartilage is crucial for diagnosing and monitoring osteoarthritis (OA).
- Quantitative Magnetic Resonance Imaging (MRI) techniques offer detailed insights into cartilage composition and health.
- Manual segmentation of cartilage in 3D MRI is time-consuming and prone to inter-observer variability.
Purpose of the Study:
- To develop a fully automated method for full-thickness cartilage segmentation using 3D ultrashort echo time (UTE) cones MRI.
- To combine automated segmentation with quantitative mapping of T1, T1ρ, T2*, and macromolecular fraction (MMF) for comprehensive cartilage assessment.
- To validate the accuracy and consistency of the automated approach against manual segmentation and evaluate its ability to detect OA-related changes.
Main Methods:
- Sixty-five participants underwent 3 Tesla (3T) 3D UTE cones MRI scans using T1, T1ρ, T2*, and magnetization transfer sequences.
- A transfer learning-based U-Net convolutional neural network (CNN) model was developed for automated cartilage segmentation.
- Segmentation accuracy was assessed using Dice score and volumetric overlap error (VOE); quantitative parameter consistency was evaluated using Pearson correlation and intraclass correlation coefficients (ICC).
Main Results:
- The U-Net CNN model achieved reliable cartilage segmentation with a mean Dice score of 0.82 and mean VOE of 29.86%.
- High consistency was observed between automatic and manual segmentations for quantitative MRI parameters (Pearson's r: 0.91–0.99, ICC: 0.91–0.96).
- Significant differences in UTE biomarkers (T1, T1ρ, T2*, MMF) were found between OA groups and normal controls, indicating sensitivity to disease.
Conclusions:
- The combination of 3D UTE cones MRI and a U-Net CNN model enables fully automated and comprehensive assessment of articular cartilage.
- This automated approach provides reliable quantitative mapping of cartilage properties, facilitating objective evaluation of OA.
- The developed method holds promise for improving the efficiency and accuracy of cartilage assessment in clinical and research settings.
