Generative Adversarial Network-based Deep Learning Framework for Cardiovascular Disease Risk Prediction

Early risk assessment is essential since cardiovascular disease (CVD) is a significant healthcare burden. Earlier assessment methods either employed machine learning (ML) paradigms or “conventional CVD risk calculators (CCVRC)”. These methods are haphazard, unreliable, incompletely automated, and subject to variability. The Generative Adversarial Network (GAN) paradigm is thus introduced. Data on 1700 individuals with carotid ultrasonography and matching coronary angiography scores (CAS), the gold standard for determining the degree of coronary artery stenosis. 52 factors in all were utilized, and they were grouped to form three sub-groups, namely (i) officebased, (ii) lab-based, and (iii) carotid ultrasound imaging phenotypes. The imbalanced cohort of the risk classes was handled by using the technique called as synthetic minority over-sampling technique (SMOTE). Using 5-fold crossvalidation, the GAN model’s performance was calculated. Benchmarking of GAN was done against LSTM and RNN. The accuracy and AUC (p=0.001) pairings for the GAN model were 93.00% and 0.953, compared to 85.80% and 0.920 for the LSTM and 82.50% and 0.881 for the RNN, respectively. The online system needs 1 s. GAN algorithms are an effective paradigm for predicting the risk along with the groups of coronary artery disease (CAD).