Abstract
Introduction: Heart disease is a prevalent global health challenge, necessitating early detection for improved patient outcomes. This study aims to develop an innovative heart disease prediction method using end-to-end deep learning, integrating self-attention mechanisms and generative adversarial networks to enhance predictive accuracy and efficiency in healthcare. Methods: We constructed an end-to-end model capable of processing diverse cardiac health data, including electrocardiograms, clinical data, and medical images. Self-attention mechanisms were incorporated to capture data correlations and dependencies, improving the extraction of latent features. Additionally, generative adversarial networks were employed to synthesize supplementary cardiac health data, augmenting the training dataset. Experiments were conducted using publicly available heart disease datasets for training, validation, and testing. Multiple evaluation metrics, including accuracy, recall, and F1-score, were employed to assess model performance. Results: Our model consistently outperformed traditional methods, achieving accuracy rates exceeding 95% on multiple datasets. Notably, the recall metric demonstrated the model's effectiveness in identifying heart disease patients, with rates exceeding 90%. The comprehensive F1-score also indicated exceptional performance, achieving optimal results. Discussion: This research highlights the potential of end-to-end deep learning with self-attention mechanisms in heart disease prediction. The model's consistent success across diverse datasets offers new possibilities for early diagnosis and intervention, ultimately enhancing patients' quality of life and health. These findings hold significant clinical application prospects and promise substantial advancements in the healthcare field.