Why the study?
Accurate LV myocardium delineation is essential for cardiac MRE analysis but currently relies on manual annotation and additional structural MRI, leaving it uncertain whether native cardiac MRE data alone suffice for reliable automated segmentation.
Population
16 healthy male volunteers
Comparison
nnU-Net v2 and MedSAM frameworks vs manual annotation benchmarked against inter-reader agreement
Design
Deep learning model development and validation study
Key result
Deep learning models (nnU-Net and MedSAM) achieved accurate left ventricular myocardium segmentation from native cardiac MRE data with Dice scores of 0.82, matching inter-reader agreement (0.79).
Authors
Loading...
Supports feasibility of automated LV segmentation from native MRE alone; leaves open clinical validation before workflow adoption.
Absolute Event Rate: 0.82% vs 0.79%
Deep learning models can accurately segment the left ventricular myocardium directly from native cardiac MRE data, achieving performance comparable to human readers.
Atamaniuk et al. (2026) studied Healthy (n=16). Deep learning-based segmentation (nnU-Net v2 and MedSAM) vs. Manual annotation (inter-reader agreement) was evaluated on Dice score for LV myocardium segmentation. Deep learning models (nnU-Net and MedSAM) achieved accurate left ventricular myocardium segmentation from native cardiac MRE data with Dice scores of 0.82, matching inter-reader agreement (0.79).