Unsupervised clustering of intra-ventricular haemodynamic forces for the phenotyping of left ventricular function in non-ischaemic left ventricular cardiomyopathy

Cardiac magnetic resonance (CMR) is essential for diagnosing cardiomyopathy, serving as the gold standard for assessing heart chamber volumes and tissue characterization. Haemodynamic forces (HDFs) analysis, a novel approach using standard cine CMR images, estimates energy exchange between the left ventricular (LV) wall and blood. While prior research has focused on peak or mean longitudinal HDF values, this study aims to investigate whether unsupervised clustering of HDF curves can identify clinically significant patterns and stratify cardiovascular (CV) risk in non-ischaemic LV cardiomyopathy (NILVC).

A retrospective cohort of 279 patients with NILVC who underwent cardiac CMR at Vall d’Hebron University Hospital (Barcelona) was examined. Unsupervised clustering of longitudinal and transversal HDF curves was performed using dynamic time warping for dissimilarity measurement and the partitioning around medoids algorithm. Outcomes were defined as a composite of CV mortality, heart failure hospitalization, and ventricular arrhythmias. The median age was 65 (57.0; 74.0) years, with 27.2% females and 35.5% showing late gadolinium enhancement (LGE). Unsupervised clustering identified three distinct clusters, delineating risk groups with worsening LA and LV function, indicating a stepwise increase in CV risk profile. Over a median follow-up of 40 months, 60 patients experienced the composite outcome. After adjusting for LGE, LV ejection fraction (EF), and LV size, Clusters 2 and 3 demonstrated a significantly higher risk of adverse events (both P < 0.05) compared with Cluster 1.

Analysing both longitudinal and transversal HDFs throughout the cardiac cycle enables the identification of distinct phenotypes with prognostic value beyond EF and LGE in NILVC patients.