Wasserstein Deep Convolutional GAN With Growth Optimizer for Multi-Modal Feature Extraction in Cardiovascular Diagnosis.

BACKGROUND: Cardiovascular disease (CVD) diagnosis using multimodal health care data remains a major challenge due to the heterogeneity of clinical and imaging information, along with the instability of generative deep learning models trained on limited medical data sets. METHODS: This research proposes a hybrid multimodal framework termed WDCGAN-COA-CD-MD for integrating Electronic Health Records (EHR), Magnetic Resonance Imaging (MRI), and Single-Photon Emission Computed Tomography (SPECT) data through feature-level fusion. Preprocessing is performed using an Adaptive Distorted Gaussian Matched Filter (ADGMF), while Ternary Pattern and Discrete Wavelet Transform (TP-DWT) techniques are utilized for discriminative feature extraction. A Wasserstein Deep Convolutional Generative Adversarial Network (WDCGAN) is employed for CVD classification. To improve adversarial convergence stability and avoid mode collapse, the Coati Optimization Algorithm (COA) adaptively optimizes generator and discriminator weight parameters using a hybrid adversarial-classification loss-based fitness function. RESULTS: Experiments were conducted on publicly available multimodal cardiovascular data sets using a 70/15/15 train-validation-test split. The proposed WDCGAN-COA-CD-MD framework achieved an accuracy of 91.82% and outperformed baseline models with statistically significant improvements (p < 0.05) in accuracy, sensitivity, specificity, precision, and F1-score. CONCLUSIONS: The integration of multimodal feature fusion, Wasserstein adversarial learning, and adaptive COA-based optimization improves convergence stability, feature representation capability, and diagnostic reliability for CVD prediction.