Integrating clinical and ECG data for heart disease prediction: A hybrid deep learning approach based on two modalities with particle swarm optimization

Abstract

Cardiovascular diseases (CVDs) remain a leading global health concern, emphasizing the need for accurate and early diagnostic systems. This study introduces a hybrid deep learning model that leverages dual-modality data by integrating clinical tabular data and ECG images for heart disease prediction. Both datasets comprising clinical features and corresponding ECG images of the same individuals and these datasets are real—time datasets. Feature extraction from ECG images is conducted using a fine-tuned EfficientNetB0 convolutional neural network, while features from the clinical dataset are extracted using a Dense Neural Network (DNN). To enhance the model’s predictive performance and reduce dimensionality, Particle Swarm Optimization (PSO) is employed to select the most relevant features from the combined feature space. The proposed dual-modality model uses a fine-tuned DNN classifier, incorporating dense and dropout layers to prevent overfitting and improve generalizability. Extensive pre-processing techniques, including image augmentation and standardization of clinical features, were applied to ensure data quality. The model achieved an accuracy of 86.13%, precision of 87%, recall of 89%, and an F1-score of 88%, significantly outperforming traditional single-modality models. Additionally, it demonstrated strong discriminative capability with a ROC AUC of 0.93. These results highlight the effectiveness of combining diverse data types and optimizing feature selection using IPSO to support clinical decision-making in heart disease diagnosis.