Predicting angiographic coronary artery disease using machine learning and high-frequency QRS

Area of Science:

• Cardiology
• Medical Diagnostics
• Machine Learning

Background:

• Exercise stress electrocardiography (ECG) is vital for diagnosing stable coronary artery disease (CAD).
• Current diagnostic accuracy requires improvement for better patient outcomes.
• High-frequency QRS (HF QRS) analysis offers a potential avenue for enhanced diagnostic capabilities.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting angiographic coronary artery disease (CAD).
• To utilize HF QRS analysis of cycling exercise ECG data for improved CAD detection.
• To identify key clinical predictors of CAD through advanced data analysis.

Main Methods:

• Prospective study involving 140 inpatients and 59 healthy volunteers undergoing exercise ECG.
• Coronary angiography used as the gold standard for determining coronary heart disease (CHD) presence.
• Automated, blinded HF QRS analysis performed, followed by multifactorial retrospective analysis to build machine learning models (XGBoost, Logistic Regression, LightGBM, RandomForest, ANN, SVM).

Main Results:

• The coronary group exhibited higher prevalence of male gender, age, BMI, hypertension, and diabetes compared to the non-coronary group.
• Elevated lipid levels, longer QRS duration during exercise, more positive leads, and significant HF QRS changes were observed in the coronary group.
• Machine learning models were constructed using clinical factors and HF QRS conclusions.

Conclusions:

• Male gender, advanced age, hypertension, diabetes mellitus, and positive exercise stress test HF QRS findings are indicative of high CHD risk.
• Logistic Regression model demonstrated superior performance in predicting CHD.
• A validated column line graph was developed for assessing individual CHD risk.