Differential cardiac remodelling: patterns of cardiomyocyte morphometry in multiple patient conditions.
Abstract
Cardiac remodelling drives heart failure (HF), yet microstructural drivers in human tissue remain poorly defined due to the limited resolution of clinical imaging. This study sought to quantitatively assess cardiomyocyte morphometry and its relationship with clinical phenotypes.
We developed MyoSAM, a deep learning framework for quantitative morphometric characterization of H&E-stained whole-slide images (N = 334 patients). Cardiomyocyte size was quantified using the myocyte maximum inscribed circle (MMIC) radius, along with measures of nuclear size and perinuclear-halo size, interstitial and macrodomain stromal fractions. Independent associations between these histological features and clinical variables were assessed using multivariable regression and patient-level linear mixed-effects models to account for clustering of cellular measurements within individuals. MMIC radius was the strongest histological predictor of left ventricular mass index (LVMI; P < 0.001) and was positively correlated with clinical HF (Pearson's r = 0.77, P < 0.001). In severe coronary artery disease, the association between LVMI and MMIC radius was significantly attenuated (P = 0.001), indicating that myocyte hypertrophy was diminished despite increased mass; instead, remodelling shifted towards greater interstitial stromal expansion (P = 0.025). Males exhibited larger MMIC radii than females (fold-changes 1.08, P = 0.003). Distinct aetiology-specific signatures emerged: anthracycline-treated hearts showed large MMIC radii and perinuclear halos with minimal stroma, while non-ischaemic cardiomyopathy exhibited the greatest nuclear enlargement and stromal macrodomain expansion.
Computational pathology enables high-resolution characterization of myocardial remodelling, revealing distinct microstructural phenotypes across sex, coronary disease burden, and HF aetiology that may improve cardiac phenotyping and risk stratification.