Automated Deep Learning Based Cardiac Quantification in Hypertrophic Cardiomyopathy: A Comparative Study with Manual Segmentation

Abstract

Background: Hypertrophic CardioMyopathy (HCM) is the most prevalent inherited cardiac disorder, where accurate assessment of Left Ventricular (LV) function and Mitral Regurgitation (MR) is crucial. Cardiac Magnetic Resonance (CMR) imaging is considered the gold standard for evaluating these parameters. Recently, Deep Learning (DL) algorithms have emerged to automate cardiac quantification, but their performance in complex pathologies such as HCM still requires validation.
Purpose: To compare the performance of a fully automated deep learning-based cardiac segmentation software (SW 2) (SuiteHEART) with conventional manual segmentation (SW 1) (syngo.Via) for quantifying crucial cardiac parameters in patients with HCM.
Materials and Methods: In this prospective study, 25 consecutive adult patients (mean age 49±12 years) with HCM referred for CMR at our institute were included. CMR examinations were performed by using a 3.0 Tesla scanner (Siemens Vida). The key parameters assessed included Left Ventricular Ejection Fraction (LVEF), End-Diastolic Volume (LVEDV), Stroke Volume (LVSV), Aortic Forward Flow (AoF), Mitral Regurgitation (MR), and Pressure Gradient (PG) across the LVOT. Manual and automated segmentations were performed by using syngo.Via (SW 1) and SuiteHEART software (SW 2), respectively. Statistical analysis included paired t-tests, linear regression, and Bland–Altman analysis.
Results: There was a strong correlation between DL-based and manual measurements for LVEF (r=0.91), LVEDV (r=0.89), LVSV (r=0.87), AoF (r=0.86), MR (r=0.84), and PG (r=0.81) (all p<0.001). Bland–Altman analysis demonstrated acceptable limits of agreement, with no significant bias. Automated segmentation significantly reduced post-processing time compared to manual methods (p<0.001).
Conclusion: Fully automated DL-based cardiac quantification provides accurate and reproducible assessment of the LV function, MR, and flow parameters in HCM patients, closely matching manual segmentation results. Incorporation of DL algorithms can substantially streamline the clinical workflow, although careful validation remains necessary in structurally complex cases such as HCM.