Radiomics for differential diagnosis of ischemic and dilated cardiomyopathy using non-contrast-enhanced cine cardiac magnetic resonance imaging.

Abstract

Objective: Our study investigated the effectiveness of using radiomics and non-gadolinium cine cardiac magnetic resonance (CMR) imaging for differential diagnosis between ischemic and dilated cardiomyopathy (ICM vs. DCM) and detecting myocardial scar without relying on gadolinium-based contrast agents.

Materials and methods: We retrospectively enrolled 200 patients with age $\ge$ 18 who were diagnosed with ICM and DCM, both with and without myocardial scars, and had complete short-axis cine CMR images without artifacts over the heart area. Radiomic features were extracted from the myocardium and blood pools in short-axis cine CMR images. Feature selection was performed using intraclass correlation coefficient and recursive feature elimination with cross-validation (RFECV). Multivariate analysis with logistic regression was utilized to construct 3 radiomic models (ICM vs DCM, Scar vs NoScar in ICM and in DCM). Model performance was evaluated using area under the receiver operating characteristic (ROC) curve (AUC) and compared with radiologist's performance using cine CMR images.

Results: For each of the three models, 21, 14, and 19 radiomic features were selected. The AUC values of each model were 0.964 ± 0.008, 0.989 ± 0.004, and 0.996 ± 0.004, respectively, in the training set and 0.918 ± 0.040, 0.955 ± 0.045, and 0.935 ± 0.052, respectively, in the test set (p < 0.0001). The radiomic models outperformed the radiologist as shown by the ROC curve.

Conclusions: Radiomics showed promise in differentiating between ICM and DCM and detecting myocardial scar with cine CMR images, which offered an alternative for ICM and DCM differentiation in patients with gadolinium contraindication.