Hybrid deep learning framework for cardiovascular risk prediction in post-COVID-19 patients

Cardiovascular complications associated with Post-COVID-19 Patients remain difficult to identify at early stages due to heterogeneous physiological manifestations and the limited integration of imaging and clinical indicators in conventional diagnostic frameworks. To address this challenge, this work proposes a multimodal deep learning framework for cardiovascular disease (CVD) risk prediction by integrating cardiac computed tomography (CT) imaging with structured clinical data. The proposed approach applies Adaptive Bilateral Filtering for image enhancement, Kernel Density Fuzzy C-Means (KDFCM) for myocardial segmentation, and Squeezing Extract Chirplet Transform (SSECT) for extracting multi-scale texture and frequency-based imaging features. Clinical variables are preprocessed through Z-score normalization, interquartile range (IQR)-based outlier removal, and Isolation Forest anomaly detection, followed by Adaptive Starfish Optimization (ASO) for feature selection. Imaging and clinical representations are integrated through feature-level fusion and classified using a Deep Image Recognition-based Generative Adversarial Network (DIR-GAN), where adversarial learning enhances discriminative feature representation and model robustness. Experimental evaluation demonstrates strong predictive performance, achieving 94.98% accuracy, 94.98% sensitivity, 93.38% specificity, 95.43% precision, and 94.74% F1-score. The findings indicate that the proposed framework provides reliable and clinically relevant cardiovascular risk stratification for patients recovering from COVID-19, while demonstrating robustness against heterogeneous multimodal data.