Abstract
BACKGROUND: Epicardial adipose tissue (EAT) radiomics derived from cardiac computed tomography (CT) images may provide insights into EAT characteristics, which can further predict regression of left ventricular mass index (LVMI) after transcatheter aortic valve replacement (TAVR). This study aimed to develop and validate a radiomics nomogram based on pre-procedural EAT CT to predict inadequate LVMI regression following TAVR. METHODS: Inadequate LVMI regression was defined as ΔLVMI% < 15% at one-year post TAVR. Radiomics features from pre-procedural CT images were selected mainly by least absolute shrinkage and selection operator algorithm. The patients were randomly divided into the training and validation cohorts to establish and evaluate three feature classifier models based on the selected features, using which the Radiomics scores (Radscores) were then calculated. A radiomics nomogram was constructed using independent risk factors and further assessed using area under the curve, calibration curve, and decision curve analysis. RESULTS: A total of 144 consecutive TAVR patients (42 patients with inadequate and 102 patients with adequate LVMI regression) were randomly assigned to the training and validation cohorts (116 patients and 28 patients, respectively). A total of 1130 radiomics features from each patient yielded 6 features for the Radscore construction after selection, with logistic regression and support vector machine models favored. Subsequently, a nomogram based solely on the Radscore was constructed, with an area under the curve of 0.743 in the validation cohort, along with favorable decision curve analysis and calibration curves. CONCLUSIONS: The developed radiomics nomogram, serving as a non-invasive tool, achieved satisfactory preoperative prediction of inadequate LVMI regression in TAVR patients, thereby facilitating clinical management.