Heterogeneity biomarkers on cardiovascular magnetic resonance imaging – a novel paradigm to measure structural and functional variation in precision cardiology
K Hesse, M Khanji, A Chahal, S Chadalavada, D Condurache, L Szabo, Z Raisi-Estabragh, S Petersen, N AungAbstract
Background
Pathological cardiac remodelling is heterogeneous, especially in the early stages of disease progression. Yet, despite the availability of segmental information embedded within standard cardiovascular magnetic resonance (CMR) imaging, clinical reports rarely incorporate formal assessment of structural or functional heterogeneity. We hypothesised that heterogeneity biomarkers are measurable and reproducible to facilitate earlier diagnosis, improved prognostication and more targeted treatment.
Purpose
To assess the prognostic value and derive reference ranges for 3 heterogeneity biomarkers (Figure 1): A) mean absolute deviation of maximum segmental wall thickness (MadWT); B) circumferential strain coefficient of variation (CoVCS); and C) left ventricular (LV) mechanical synchrony index (TTPSNR).
Methods
Deep learning-derived CMR segmentation and feature tracking calculated biomarkers from two overlapping cohorts, Healthy Hearts Consortium (HHC, n=8,443) and UK Biobank (UKBB, n=60,746). Normal, mild, moderate and severe biomarker categories were defined by 95%, 99.73% and 99.99% prediction intervals in HHC (Figure 1). Kaplan-Meier survival curves and Cox proportional hazards regression analysed unadjusted and adjusted associations with cardiovascular (CV) outcomes, including major adverse CV events (MACE), heart failure (HF), arrhythmia (atrial fibrillation [AF] and ventricular arrhythmia [VA]) and mortality in UKBB. Covariables were age, sex and ethnicity.
Results
Over a median follow-up of 5.7 years, risk of adverse outcomes increased in stepwise fashion across severity strata for MadWT, CoVCS and LV TTPSNR (Figure 1). Figure 2 presents other notable findings: A) In the top 1% most physically active, MadWT was the only biomarker to differentiate between hypertensive and non-hypertensive (P<0.05). B) When LVEF ≥50% and ≥3 of the following present age >65 years, obesity, hypertension, diabetes, AF and chronic kidney disease, then CoVCS was more predictive of HF than global longitudinal strain (HR 1.37 [1.14-1.65] vs. HR 0.68 [0.53-0.86] per 1 standard deviation change). C) LV TTPSNR ≤5.5 ms and either QRS duration ≥94 ms or QTc interval ≥428 ms conferred greater risk of the composite endpoint, HF, VA and CV death than either alone (P<0.0001).
Conclusions
Heterogeneity biomarkers are sensitive to early pathological signals by measuring disease regionality. Although findings require validation in other healthy and diseased cohorts, the biomarkers may be highly relevant in clinical decision pathways. MadWT may facilitate differentiation between physiological and pathological causes of unexplained LV hypertrophy. Mechanical synchrony index assessment may complement electrical dyssynchrony criteria for cardiac resynchronisation therapy. CoVCS may provide important regionality assessment in HF with preserved ejection fraction. The biomarkers can be readily integrated into deep learning-assisted CMR analysis workflows.Figure 1Figure 2