Abstract
BACKGROUND: Hypertrophic cardiomyopathy (HCM) and left ventricular hypertrophy (LVH) from other causes present similar features on transthoracic echocardiography (TTE), making an accurate differentiation challenging. Recent advancements in radiomics and deep transfer learning (DTL) have shown promise; however, no studies have combined these techniques to diagnose HCM and LVH resulting from other causes. Therefore, we developed a fusion model that integrates radiomic features from the left ventricular myocardium in the four-chamber view of TTE with DTL features to differentiate HCM from other causes of LVH, providing more reliable diagnostic support. METHODS: This multicenter study included 971 patients (303 with HCM, 668 with hypertensive heart disease and uremic cardiomyopathy). Patients from Institution 1 were split into a training set and an internal validation set, while patients from Institution 2 served as an external validation set. Radiomic features were extracted using pyradiomics, and DTL features were obtained via DenseNet121. Features were selected using least absolute shrinkage and selection operator (LASSO) and input into ten machine learning algorithms, with support vector machine (SVM) as the classifier. Model performance was assessed using receiver operating characteristic (ROC) curves and decision curve analysis (DCA) and compared with the diagnostic results of two ultrasound physicians. RESULTS: The fusion model demonstrated excellent diagnostic performance: the area under the curve (AUC) values were 0.966 (training set), 0.945 (internal validation), and 0.934 (external validation), thereby outperforming models that used only radiomic or DTL features. DCA indicated superior clinical effectiveness, surpassing the diagnostic performance of two ultrasound physicians. CONCLUSIONS: A fusion model combining radiomics and DTL features significantly improves the ability to distinguish HCM from other causes of LVH and has strong potential for clinical applications.