Performance of Prediction Models for Diagnosing Severe Aortic Stenosis Based on Aortic Valve Calcium on Cardiac Computed Tomography: Incorporation of Radiomics and Machine Learning

This study aimed to develop a prediction model for diagnosing severe aortic stenosis (AS) using CT radiomics features of aortic valve calcium (AVC) and machine learning (ML) algorithms. It involved 408 patients who underwent cardiac CT and echocardiographic examinations, divided into training and validation sets. AVC was segmented and 128 radiomics features were extracted. Three ML algorithms (LASSO, random forests, and XGBoost) were used for feature selection, and prediction models were developed using logistic regression, RF, and XGBoost. AVC segmentation was independently performed by two radiologists using AVIEW Research software. The LASSO + XGBoost model showed the highest c-index of 0.921, outperforming models based on AVC volume and score, which had c-indexes of 0.894 and 0.899, respectively. Although the differences were not statistically significant, the radiomics prediction models demonstrated higher discrimination abilities for severe AS. Further investigation is needed to validate the added value of radiomics features.