Automated inline myocardial segmentation of joint T1 and T2 mapping using deep learning
File(s)
Accepted version
Author(s)
Type
Journal Article
Abstract
Purpose:
To develop an artificial intelligence (AI) solution for automated segmentation and analysis of joint cardiac MRI T1 and T2 short-axis mapping.
Materials and Methods:
In this retrospective study, a joint T1 and T2 mapping sequence was used to acquire 4240 maps from 807 patients across 2 hospitals (March-November 2020). 509 maps from 94 consecutive patients were assigned to a holdout testing set. A convolutional neural network was trained to segment the endocardial and epicardial contours using an edge probability estimation approach. Training labels were segmented by an expert cardiologist. Predicted contours were processed to yield mapping values for each of the 16 AHA segments. Network segmentation performance and segment-wise measurements on the testing set were compared with two experts on the holdout testing set. The AI model was fully integrated using Gadgetron inline AI to run on MRI scanners.
Results:
A total of 3899 maps (92%) were deemed artifact-free and suitable for human segmentation. AI segmentation closely matched that of each expert (mean Dice coefficient 0.82 ± [SD] 0.07, 0.86 ± 0.06), comparing favorably with interexpert agreement (0.84 ± 0.06). AI-derived segment-wise values for native T1, postcontrast T1 and T2 mapping correlated with experts (R2 0.96, 0.98, 0.87, respectively versus expert 1; 0.97, 0.99, 0.97 versus expert 2) and fell within the range of interexpert reproducibility (R2 = 0.97, 0.99, 0.90). The AI has since been deployed at two hospitals, enabling automated inline analysis.
Conclusion:
Automated inline analysis of joint T1 and T2 mapping allows accurate segment-wise tissue characterization, with performance equivalent to human experts.
To develop an artificial intelligence (AI) solution for automated segmentation and analysis of joint cardiac MRI T1 and T2 short-axis mapping.
Materials and Methods:
In this retrospective study, a joint T1 and T2 mapping sequence was used to acquire 4240 maps from 807 patients across 2 hospitals (March-November 2020). 509 maps from 94 consecutive patients were assigned to a holdout testing set. A convolutional neural network was trained to segment the endocardial and epicardial contours using an edge probability estimation approach. Training labels were segmented by an expert cardiologist. Predicted contours were processed to yield mapping values for each of the 16 AHA segments. Network segmentation performance and segment-wise measurements on the testing set were compared with two experts on the holdout testing set. The AI model was fully integrated using Gadgetron inline AI to run on MRI scanners.
Results:
A total of 3899 maps (92%) were deemed artifact-free and suitable for human segmentation. AI segmentation closely matched that of each expert (mean Dice coefficient 0.82 ± [SD] 0.07, 0.86 ± 0.06), comparing favorably with interexpert agreement (0.84 ± 0.06). AI-derived segment-wise values for native T1, postcontrast T1 and T2 mapping correlated with experts (R2 0.96, 0.98, 0.87, respectively versus expert 1; 0.97, 0.99, 0.97 versus expert 2) and fell within the range of interexpert reproducibility (R2 = 0.97, 0.99, 0.90). The AI has since been deployed at two hospitals, enabling automated inline analysis.
Conclusion:
Automated inline analysis of joint T1 and T2 mapping allows accurate segment-wise tissue characterization, with performance equivalent to human experts.
Date Issued
2022-11-09
Date Acceptance
2022-10-18
Citation
Radiology: Artificial Intelligence, 2022, 1 (5), pp.1-1
ISSN
2638-6100
Publisher
Radiological Society of North America
Start Page
1
End Page
1
Journal / Book Title
Radiology: Artificial Intelligence
Volume
1
Issue
5
Copyright Statement
© 2022 by the Radiological Society of North America, Inc.
License URL
Identifier
https://pubs.rsna.org/doi/10.1148/ryai.220050
Publication Status
Published
Date Publish Online
2022-11-09