Related Experiment Video
Updated: Aug 17, 2025
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
A Deep Learning Segmentation Pipeline for Cardiac T1 Mapping Using MRI Relaxation-based Synthetic Contrast
Nitish Bhatt1, Venkat Ramanan1, Ady Orbach1
1Faculty of Medicine (N.B.), Department of Medical Imaging (L.G., L.J.J.), Department of Medicine (I.R.), and Department of Medical Biophysics (G.A.W., N.R.G.), University of Toronto, 1 King's College Circle, Toronto, ON, Canada M5S A18; Physical Sciences Platform, Sunnybrook Research Institute, Toronto, Ontario, Canada (N.B., V.R., L.B., M.N., F.G., X.Q., G.A.W., N.R.G.); Schulich Heart Program (V.R., A.O., L.B., M.N., F.G., X.Q., I.R., G.A.W., N.R.G.) and Department of Radiology (L.G., L.J.J.), Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada; and Department of Radiology, St Michael's Hospital, Toronto, Ontario, Canada (L.J.J.).
This study introduces an automated deep learning method for cardiac MRI T1 map segmentation and analysis using synthetic T1-weighted images. The method demonstrates high accuracy and strong correlation with expert analysis for T1 and ECV quantification.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Cardiovascular Imaging
Background:
- Cardiac MRI T1 maps are crucial for tissue characterization.
- Accurate segmentation and analysis of T1 maps are essential for quantitative assessment.
- Current methods can be time-consuming and require expert input.
Purpose of the Study:
- To develop and evaluate an automated deep learning method for cardiac MRI T1 map segmentation and analysis.
- To utilize synthetic T1-weighted images for contrast augmentation in deep learning models.
- To assess the accuracy and generalizability of the automated method.
Main Methods:
- A retrospective study included 100 patients' cardiac MRI scans.
- Synthetic T1-weighted images were generated for contrast augmentation.
- A convolutional neural network was trained for myocardial segmentation and T1/ECV analysis.
- The automated method was validated against expert analysis and an external dataset.
Main Results:
- The automated method achieved high Dice similarity coefficients (DSCs) for myocardial segmentation (internal: 0.81, external: 0.80).
- Automated segmental measurements strongly correlated with expert analysis for T1native (R=0.87), T1post (R=0.91), and ECV (R=0.92).
- Results were comparable to interobserver variability.
Conclusions:
- The developed deep learning method provides accurate automated T1 and ECV analysis for cardiac MRI.
- Synthetic contrast augmentation enhances the performance of automated segmentation and analysis.
- The method shows potential for reliable application across diverse clinical scenarios and imaging parameters.
