Spatiotemporal Deep Learning for Scar Screening in CMR: Toward Selective Use of Gadolinium.
Authors
Affiliations (9)
Affiliations (9)
- Departments of Medicine (Cardiovascular Division) and.
- Division of Cardiology, Weill Cornell Medicine, New York, NY, USA.
- Hypertrophic Cardiomyopathy Center, Lahey Medical Center, Boston, MA, USA.
- Department of Medicine, Boston Medical Center, Boston, MA, USA.
- Department of Medicine (Cardiology Division), Duke University, Durham, NC, USA.
- Departments of Medicine (Cardiovascular Division) and; Radiology, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, MA, USA.
- Department of Medicine (Cardiovascular Division), Brigham and Women's Hospital and Harvard Medical School, Boston, MA, USA.
- Department of Medicine (Section of Cardiovascular Medicine), Boston Medical Center and Boston University Chobanian & Avedisian School of Medicine, Boston, MA, USA.
- Departments of Medicine (Cardiovascular Division) and. Electronic address: [email protected].
Abstract
Late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) is the gold standard for assessing myocardial scar. However, a substantial proportion of patients referred for CMR scar assessment are ultimately found to have no evidence of myocardial scarring. To develop and evaluate a deep learning (DL) model capable of identifying patients without myocardial scar using cine imaging alone, thereby obviating the need for contrast administration and LGE imaging in these individuals. We developed a novel spatiotemporal DL architecture to identify patients unlikely to have myocardial scar using contrast-free cine images. The model was trained on short-axis cine images from a consecutive cohort of 3,000 patients (1,753 males; mean age 54 ± 18 years) undergoing CMR for evaluation of known or suspected cardiovascular disease, using 1.5 and 3T scanners from Siemens and GE. External validation was performed in an independent multicenter cohort of 1,792 patients where images were acquired on 1.5T and 3T scanners from Siemens and Philips. The architecture utilizes factorized convolutions to extract spatial and temporal features and incorporates residual attention mechanisms to emphasize features most predictive of scar presence on LGE imaging. To evaluate the incremental value of incorporating temporal information, a second model was developed that excluded the temporal kernel from the DL architecture. Both models were trained and optimized using the same training dataset and were evaluated based on similar internal and external testing cohorts. Model performance in identifying patients without myocardial scar was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. The spatiotemporal model had a higher AUC compared to the spatial model in the internal (0.79±0.02 vs. 0.70±0.05, p<0.05, n=500) and external cohort (0.78 vs. 0.64, p<0.001, n=1792). The spatiotemporal model correctly identified 64% and 52% of patients without scar in the internal and external test sets, while maintaining a high sensitivity (86% and 82%). Incorporating temporal information from cine images using an end-to-end spatiotemporal DL architecture enables non-contrast screening for myocardial scar.