A Real-Time Computer-Aided Diagnosis System for Coronary Heart Disease Prediction Using Clinical Information.

Abstract

Background: It is important to establish a coronary heart disease (CHD) prediction model with high efficiency and precision for early diagnosis of CHD using clinical information. While existing deep learning-based CHD prediction models possess the limitations of large datasets and long training time, existing machine learning-based CHD prediction models have the limitations of low accuracy and robustness, which are unsuitable for clinical application. This study aimed to design a fast and high-precision intelligent model using clinical information to predict CHD.

Methods: Five public datasets, including 303, 293, 303, 200, and 123 patients with 55, 14, 14, 14, and 14 attributes, respectively, were used for model training and testing. After data preprocessing, the singular value decomposition method was utilized to extract features to build the CHD prediction model. Then, the CHD prediction model was established using the 5-fold cross-validation method with a multilayer perceptron approach.

Results: Results show that the established model performs better on the total dataset than the other models we built in this study. This machine learning-based CHD prediction model achieved an improved area under the curve (AUC) of 99.10%, with 96.63% accuracy, 96.50% precision, 97.4% recall, and 97.0% F ₁-score on the total dataset.

Conclusions: This high precision and efficiency achieved by the proposed model on different datasets would be significant for the prediction of CHD for medical and clinical diagnosis purposes.