Comparison of Random Forest and Artificial Neural Network Models to Evaluate Diagnostic Factors in the Necessity to Perform Angiography

Coronary Artery Disease (CAD) is the most common type of cardiovascular disorders. Despite being costly and invasive, coronary angiography is a reliable method for diagnosing CAD. Therefore, it is crucial to use non-invasive methods to screen candidates for angiography to accelerate the process of decision-making. Two powerful Machine Learning (ML) methods are Random Forest (RF) and Artificial Neural Network (ANN).

The present study aimed to compare RF and ANN to define the most important features for positive CAD results and predict the need for angiography as a screening method.

This cross-sectional study was performed on 1128 patients referred for angiography. The data were divided into test and train sets. The models (RF and ANN) were fitted with the angiographic outcome variable (positive or negative) as the dependent variable and five features as predictors. Then, the performances of the models were compared by considering the Area Under the Rock Curve (AUC). All statistical analyses were done using the R software, version 4.1.2.

Out of the 1128 patients, 752 (66.7%) had positive angiography results. The AUC values were 0.75 and 0.52 for the test data set in ANN and RF models, respectively.

Fasting Blood Sugar (FBS), gender, age, Body Mass Index (BMI), and smoking habit were important in predicting the results of an angiography for CAD. Applying these factors in ML approaches can be considered a screen for angiography to accelerate the process of diagnosis.