Clinical Report: Manual vs AI-Driven Myocardial Strain Measurement by CMR Feature Tracking

Overview

This study compared manual and AI-based myocardial strain measurement methods using cardiovascular magnetic resonance (CMR) feature tracking in healthy volunteers and patients with various cardiac diseases. AI-driven segmentation demonstrated potential to streamline strain assessment while maintaining accuracy across global and segmental strain metrics.

Background

Myocardial strain quantifies myocardial deformation and can be assessed by CMR through tissue tagging or feature tracking (FT) on cine images. FT allows evaluation of longitudinal, circumferential, and radial strains from routine cine images without special sequences. Despite its clinical utility in conditions like cardiotoxicity, ischemic heart disease, and cardiomyopathies, FT lacks standardization and is influenced by manual contouring variability. AI-based segmentation offers a promising approach to reduce manual bias and improve reproducibility in strain analysis.

Data Highlights

The study included 60 healthy volunteers and 76 patients with left ventricular hypertrophy, aortic stenosis, hypertrophic cardiomyopathy, and chronic myocardial infarction subtypes. CMR was performed at 1.5-T and 3-T scanners with standardized cine imaging protocols. Manual and AI segmentations were performed in end-diastole on short-axis and long-axis views. Strain values for longitudinal, circumferential, and radial directions were derived globally and segmentally using the 17-segment AHA model excluding the apical segment. AI segmentation was based on deep convolutional neural networks trained on diverse datasets including pathological cases.

Key Findings

• AI-based segmentation successfully generated contours for strain analysis comparable to manual segmentation in both healthy and diseased hearts.
• Global and segmental strain values derived from AI contours showed good agreement with manual measurements across longitudinal, circumferential, and radial strain components.
• AI segmentation reduced the time-consuming manual contouring process and minimized operator-dependent variability.
• Proper feature tracking was confirmed by mesh analysis and myocardial point tracking, with AI contours demonstrating reliable tracking performance.
• AI models trained on large, heterogeneous datasets including pathological conditions enhanced generalizability of segmentation results.

Clinical Implications

AI-driven myocardial strain measurement via CMR feature tracking can streamline clinical workflows by reducing manual contouring time and variability. This approach may facilitate standardized strain assessment in routine practice and large-scale studies, improving detection and monitoring of myocardial dysfunction across diverse cardiac diseases. Integration of AI segmentation into clinical CMR analysis has the potential to enhance reproducibility and diagnostic confidence.

Conclusion

AI-based segmentation for myocardial strain measurement by CMR feature tracking offers a reliable and efficient alternative to manual contouring, supporting its adoption in clinical and research settings for comprehensive cardiac function evaluation.

References

1. Original Study -- Evaluation of Manual Versus AI-Driven Myocardial Strain Measurement Through Feature Tracking