Predictive Model Based on Texture Analysis of Noncontrast Cardiac Magnetic Resonance Images for the Prognostic Evaluation of Cardiac Amyloidosis.

Objectives: We aimed to develop a predictive model based on textural features of noncontrast cardiac magnetic resonance (CMR) imaging for risk stratification toward adverse events in patients with cardiac amyloidosis (CA).

Methods: A cohort of 78 patients with CA was grouped into training (n = 54) and validation (n = 24) sets at a ratio of 7:3. A total of 275 textural features were extracted from the CMR images. MaZda and a support vector machine (SVM) were used for feature selection and model construction. An SVM model incorporating radiological and textural features was built to predict endpoint events by evaluating the area under the curve.

Results: In the entire cohort, 52 patients experienced major adverse cardiovascular events and 26 patients did not. By combining 2 radiological features and 8 texture features, extracted from cine and T2-weighted imaging images, the SVM model achieved area under the curves of the receiver operating characteristic and precision-recall curves of 0.930 and 0.962 in the training cohort and that of 0.867 and 0.941 in the validated cohort, respectively. The Kaplan-Meier curve of this SVM model criterion significantly stratified the CA outcomes (log-rank test, P < 0.0001).

Conclusions: The SVM model based on radiological and textural features derived from noncontrast CMR images can be a reliable biomarker for adverse events prognostication in patients with CA.