Deep learning-based prediction of coronary artery calcium scoring in hemodialysis patients using radial artery calcification.

Abstract

Objective: This study used random forest model to explore the feasibility of radial artery calcification in prediction of coronary artery calcification in hemodialysis patients.

Material and methods: We enrolled hemodialysis patients and performed ultrasound examinations on their radial arteries to evaluate the calcification status using a calcification index. All involved patients received coronary artery computed tomography scans to generate coronary artery calcification scores (CACS). Clinical variables were collected from all patients. We constructed both a random forest model and a logistic regression model to predict CACS. Logistic regression model was used to identify the risk factors of radial artery calcification.

Results: One hundred eighteen patients were included in our analysis. In random forest model, the radial artery calcification index, age, serum C-reactive protein, body mass index (BMI), diabetes, and hypertension history were related to CACS based on the average decrease of the Gini coefficient. The random forest model achieved a sensitivity of 76.9%, specificity of 75.0%, and area under receiver operating characteristic of 0.869, while the logistic regression model achieved a sensitivity of 75.2%, specificity of 68.7%, and area under receiver operating characteristic of 0.742 in prediction of CACS. Sex, BMI index, smoking history, hypertension history, diabetes history, and serum total calcium were all the risk factors related to radial artery calcification.

Conclusions: A random forest model based on radial artery calcification could be used to predict CACS in hemodialysis patients, providing a potential method for rapid screening and prediction of coronary artery calcification.