Feature-tracking-based strain analysis – a comparison of tracking algorithms

Introduction
Optical flow feature-tracking (FT) strain assessment is increasingly being employed scientifically and clinically. Several software packages, employing different algorithms, enable computation of FT-derived strains. The aim of this study is to investigate the impact of the underlying algorithm on the validity and robustness of FT-derived strain results.

Material and methods
CSPAMM and SSFP cine sequences were acquired in 30 subjects (15 patients with aortic stenosis and associated secondary hypertrophic cardiomyopathy, and 15 controls) in identical midventricular short-axis locations. Global peak systolic circumferential strain (PSCS) was calculated using tagging and feature-tracking software with different algorithms (non-rigid, elastic image registration, and blood myocardial border tracing). Intermodality agreement and intra- as well inter-observer variability were assessed.

Results
Intermodality/inter-algorithm comparison for global PSCS using Friedman’s test revealed statistically significant differences (tagging vs. blood myocardial border tracing algorithm). Intermodality assessment revealed the highest correlation between tagging and non-rigid, elastic image registration (r = 0.84), while correlation between tagging and blood myocardial border tracing (r = 0.36) and between the two feature-tracking software packages (r = 0.5) were considerably lower.

Conclusions
The type of algorithm employed during feature-tracking strain assessment has a significant impact on the results. The non-rigid, elastic image registration algorithm produces more precise and reproducible results than the blood myocardium tracing algorithm.