Scholarlink

Objective <p>Current multimodal models often experience performance degradation when faced with incomplete data and lack effective fusion strategies for diverse data types, such as imaging, clinical, and electrophysiological data. This study aims to develop a novel model to improve the accuracy of early diagnosis of coronary artery disease.</p> Materials and methods <p>Clinical data, laboratory test results, coronary CTA images, and ECG data from consecutive patients who underwent coronary CTA examinations at the center between February 2022 and August 2023 were collected. Features were extracted using convolutional neural networks, and a multimodal prediction model was developed using a weighted fusion strategy. An internal validation was performed with a data split ratio of 7:1.5:1.5, and an independent external validation was conducted using an external dataset. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC-ROC), and model interpretability was assessed using the SHAP algorithm. Diagnostic consistency was evaluated using the Kappa coefficient.</p> Result <p>Compared to traditional diagnostic methods, the model has increased the diagnosis rate of coronary heart disease by 10.43% points in absolute terms and by 44.0% in relative terms. This improvement has greatly enhanced the ability to identify coronary heart disease at an early stage. For diagnosing cardiovascular diseases, the performance of the multimodal model is significantly superior to that of single imaging or clinical indicators. The area under the curve (AUC) was 0.89 in internal validation, and the model also demonstrated excellent performance in the external validation dataset, with an AUC of 0.83.</p> Conclusion <p>This model, by integrating three types of data for the first time and tolerating up to 50% missing data, significantly improves the detection rate of coronary artery disease and enhances diagnostic sensitivity and accuracy, thereby providing robust support for early clinical intervention and precise treatment.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multimodal deep learning with missing data robustness for enhanced early diagnosis of coronary artery disease using CCTA, clinical, and ECG data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月13日

查看全文
  • YongBo Tu,
  • XiaoRong Yang,
  • Fang Liu,
  • RuYue Ai,
  • Yu Luo,
  • Ran ChunYan,
  • GuangXu Yang,
  • TiJiang Zhang,
  • Lin Jiang

摘要

Objective

Current multimodal models often experience performance degradation when faced with incomplete data and lack effective fusion strategies for diverse data types, such as imaging, clinical, and electrophysiological data. This study aims to develop a novel model to improve the accuracy of early diagnosis of coronary artery disease.

Materials and methods

Clinical data, laboratory test results, coronary CTA images, and ECG data from consecutive patients who underwent coronary CTA examinations at the center between February 2022 and August 2023 were collected. Features were extracted using convolutional neural networks, and a multimodal prediction model was developed using a weighted fusion strategy. An internal validation was performed with a data split ratio of 7:1.5:1.5, and an independent external validation was conducted using an external dataset. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC-ROC), and model interpretability was assessed using the SHAP algorithm. Diagnostic consistency was evaluated using the Kappa coefficient.

Result

Compared to traditional diagnostic methods, the model has increased the diagnosis rate of coronary heart disease by 10.43% points in absolute terms and by 44.0% in relative terms. This improvement has greatly enhanced the ability to identify coronary heart disease at an early stage. For diagnosing cardiovascular diseases, the performance of the multimodal model is significantly superior to that of single imaging or clinical indicators. The area under the curve (AUC) was 0.89 in internal validation, and the model also demonstrated excellent performance in the external validation dataset, with an AUC of 0.83.

Conclusion

This model, by integrating three types of data for the first time and tolerating up to 50% missing data, significantly improves the detection rate of coronary artery disease and enhances diagnostic sensitivity and accuracy, thereby providing robust support for early clinical intervention and precise treatment.