Wasserstein generative adversarial network with gradient penalty-enhanced one-dimensional convolutional neural network for early cardiovascular disease risk prediction with limited pulse wave samples

Early cardiovascular disease (CVD) risk assessment based on pulse wave analysis is challenged by data scarcity, severe class imbalance, and the difficulty of modeling continuous pathological progression with discrete classification schemes. To learn data distributions from limited samples and to generate high-fidelity synthetic pulse waves for data augmentation, this study proposes a Wasserstein generative adversarial network with gradient penalty (WGAN-GP) enhanced one-dimensional convolutional neural network (1D-CNN) framework for pulse-wave-based CVD risk grading. Subjects are stratified into four clinically meaningful stages—healthy controls, risk exposure, sub-clinical damage, and confirmed lesion—based on vascular functional and structural indicators, enabling a physiologically interpretable pathological continuum. The proposed framework demonstrates strong classification performance under extreme class imbalance. By augmenting limited real pulse-wave samples into a substantially larger training set, the framework achieved an overall accuracy of 92.28% and a recall of 85.33% for severe lesion cases, representing improvements of 8.54% and 24.00%, respectively, compared with the baseline model trained on the original unaugmented dataset. Moreover, the framework may also be beneficial in scenarios with relatively sufficient data, where generative augmentation can further enrich feature variation and potentially improve classification performance. These results indicate that the proposed method provides a robust and generalizable engineering solution for early CVD risk screening.