Related Experiment Video
Updated: Aug 12, 2025
3D Whole-heart Myocardial Tissue Analysis
Published on: April 12, 2017
Automated Inline Myocardial Segmentation of Joint T1 and T2 Mapping Using Deep Learning
James P Howard1, Kelvin Chow1, Liza Chacko1
1National Heart and Lung Institute, Imperial College London, Du Cane Rd, B Block, 2nd Floor, Hammersmith Hospital, London W12 0HS, England (J.P.H., G.D.C.); National Amyloidosis Centre, Division of Medicine, University College London, London, England (L.C., M.F.); Cardiovascular MR R&D, Siemens Medical Solutions USA, Chicago, Ill (K.C., L.C., M.F.); and Medical Signal and Image Processing Program, National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Md (P.K., H.X.).
An artificial intelligence (AI) solution automates cardiac MRI T1 and T2 mapping analysis. This AI achieves expert-level accuracy in segment-wise tissue characterization, improving cardiac MRI workflows.
Area of Science:
- Cardiovascular Imaging
- Artificial Intelligence in Medicine
- Medical Image Analysis
Background:
- Cardiac magnetic resonance imaging (MRI) provides vital information on myocardial tissue characteristics.
- T1 and T2 mapping sequences are crucial for quantitative assessment of myocardial tissue.
- Automated analysis of these complex datasets is needed to improve efficiency and consistency.
Purpose of the Study:
- To develop and validate an artificial intelligence (AI) solution for automated segmentation and analysis of joint cardiac MRI short-axis T1 and T2 mapping.
- To assess the performance of the AI model compared to expert human analysis.
Main Methods:
- A convolutional neural network was trained on 4240 cardiac MRI T1 and T2 maps from 807 patients.
- The AI model used an edge probability estimation approach for segmenting endocardial and epicardial contours.
- Performance was evaluated on a holdout set of 509 maps from 94 patients, comparing AI segmentation and measurements against two expert cardiologists.
Main Results:
- AI segmentation demonstrated high agreement with expert segmentation (Dice coefficients of 0.82-0.86) and interexpert agreement (0.84).
- AI-derived segment-wise mapping values for native T1, postcontrast T1, and T2 showed strong correlations with expert values (R² = 0.96-0.99).
- The AI model's performance was equivalent to that of human experts in segment-wise tissue characterization.
Conclusions:
- Automated inline analysis of joint T1 and T2 mapping using AI enables accurate segment-wise tissue characterization.
- The developed AI solution offers expert-level performance, enhancing the efficiency and reliability of cardiac MRI analysis.
- The AI model has been successfully deployed in clinical settings, facilitating automated analysis directly on MRI scanners.
