Enhancing cardiac amyloidosis and hypertrophic cardiomyopathy differentiation using radiomics-based machine learning on cardiac MRI

Background Cardiac amyloidosis (CA) and hypertrophic cardiomyopathy (HCM) present with overlapping myocardial native T1 values on cardiac magnetic resonance imaging (CMR), which can make differentiation challenging. A precise and noninvasive diagnosis of CA is crucial for early intervention, as delayed diagnosis can lead to a poor prognosis. Machine learning has emerged as a promising tool to enhance diagnostic accuracy in cardiology, offering an innovative approach to overcoming these challenges. Objective This study aimed to differentiate CA from HCM using radiomics-based machine learning applied to native T1 mapping on CMR, with the goal of improving diagnostic accuracy and enabling early intervention. Methods The study population comprised 35 patients (11 with CA and 24 with HCM). Native T1 mapping on CMR was utilized to extract 137 radiomics features, and feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression. Six machine learning models were employed: support vector machines (SVM), decision trees, random forests, logistic regression, Naïve Bayes, and k-nearest neighbors (k-NN). Model performance was evaluated using leave-one-out cross-validation. The results showed that SVM achieved the highest accuracy (0.92), followed by logistic regression (0.88) and Naïve Bayes (0.84). The SVM also demonstrated superior performance with a precision, recall, and F1 score of 0.91, and an AUC of 0.86. Decision trees and random forests showed lower accuracy (0.76). These findings highlight the effectiveness of the SVM in differentiating CA from HCM using radiomics-based features. Conclusion Radiomics-based machine learning significantly improves CA and HCM differentiation using native T1 mapping of CMR. The SVM demonstrated the highest accuracy, highlighting its potential as a noninvasive and effective diagnostic tool. Further validation in larger cohorts is required.