A Multimodal Diagnostic Model for CHD Utilizing ECG and Clinical Data

ABSTRACT

Coronary heart disease (CHD) remains prevalent worldwide, and accurate diagnosis is crucial for subsequent medical intervention. This study proposes a multimodal fusion model (SRNXFM), based on Squeeze‐and‐Excitation ResNet with Next (SEResNeXt), utilizing 12‐lead electrocardiogram (ECG) and actual clinical data to facilitate early diagnosis of CHD. This study is divided into three sections: research on the clinical data model, the ECG model and the multimodal fusion model. For the clinical data model, five machine learning algorithms and four deep learning algorithms were used to analyse this study. For the ECG model, this study proposes a SEResNeXt‐LSTM framework that inputs the features extracted by SEResNeXt into LSTM for the diagnosis of CHD. For the multimodal fusion model, this study proposes the SRNXFM model, a dual‐input framework based on feature fusion. It integrates ECG and clinical data from the same patient to diagnose CHD. Experimental results demonstrate that the model achieved an AUC value of 0.9485, significantly better than the values based solely on clinical data (best AUC: 0.8493) or ECG data (best AUC: 0.8581). Furthermore, the model attained an overall accuracy of 0.9529, repreenting improvements of 10.76% and 12.98% over clinical data (0.8603) and ECG (0.8434) models, respectively. It also outperformed the single‐modality approaches across other key metrics, including precision (0.9001), recall (0.9978) and F1‐score (0.9311). To improve the interpretability of proposed model, the contribution ranking of risk factor characteristics is also studied. This provides valuable insights into the prevention and treatment of CHD and supports clinicians in making accurate diagnoses.