Scholarlink

Background <p>Cardiovascular diseases (CVDs) continue to be the primary cause of worldwide death, and their prompt identification by electrocardiography (ECG) is impeded by signal variability, noise, and the restricted interpretability of current AI models. Current machine-learning methods are often weak, poorly optimised, or unclear in clinical contexts, which makes them less useful in real-world diagnostic scenarios.</p> Methods <p>This paper suggests an interpretable ECG-based diagnostic framework that incorporates a Three-Parent Genetic Algorithm (3PGA) for feature optimisation and a Gradient Boosting classifier for multi-class cardiovascular disease (CVD) diagnosis. Nineteen clinically relevant ECG and demographic variables were derived from 24,894 real-time ECG recordings obtained from various units of SKIMS Hospital. They used Min–Max normalisation, 3PGA-driven feature selection, and Gradient Boosting classification. We employed accuracy, precision, recall, F1-score, a confusion matrix, ROC curves, and Explainability tools, including SHAP, LIME, and PFI, to check the models.</p> Results <p>The Hybrid 3PGA Gradient Boosting method performed very well, achieving an accuracy of 99.70%, precision of 99.71%, recall of 99.70%, and an F1-score of 99.70%. The confusion matrix exhibited almost complete diagonal dominance and very few misclassifications. The ROC curves indicated that the AUC was 1.00 for all six CVD classes. SHAP and LIME analysis verified that the model was based on physiologically and therapeutically significant ECG characteristics.</p> Conclusion <p>The proposed hybrid optimisation approach provides a highly accurate, straightforward, and computationally efficient method for finding multi-class CVD. Its excellent ability to forecast and facilitate clear decision-making suggests that it may be a good fit for real-time clinical procedures and wearable ECG monitoring devices.</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

Interpretable machine learning framework for detection of cardiovascular diseases from real-time ECG signals

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Imteyaz Hussain Khan,
  • Amar Singh,
  • Hilal Ahmed Rather

摘要

Background

Cardiovascular diseases (CVDs) continue to be the primary cause of worldwide death, and their prompt identification by electrocardiography (ECG) is impeded by signal variability, noise, and the restricted interpretability of current AI models. Current machine-learning methods are often weak, poorly optimised, or unclear in clinical contexts, which makes them less useful in real-world diagnostic scenarios.

Methods

This paper suggests an interpretable ECG-based diagnostic framework that incorporates a Three-Parent Genetic Algorithm (3PGA) for feature optimisation and a Gradient Boosting classifier for multi-class cardiovascular disease (CVD) diagnosis. Nineteen clinically relevant ECG and demographic variables were derived from 24,894 real-time ECG recordings obtained from various units of SKIMS Hospital. They used Min–Max normalisation, 3PGA-driven feature selection, and Gradient Boosting classification. We employed accuracy, precision, recall, F1-score, a confusion matrix, ROC curves, and Explainability tools, including SHAP, LIME, and PFI, to check the models.

Results

The Hybrid 3PGA Gradient Boosting method performed very well, achieving an accuracy of 99.70%, precision of 99.71%, recall of 99.70%, and an F1-score of 99.70%. The confusion matrix exhibited almost complete diagonal dominance and very few misclassifications. The ROC curves indicated that the AUC was 1.00 for all six CVD classes. SHAP and LIME analysis verified that the model was based on physiologically and therapeutically significant ECG characteristics.

Conclusion

The proposed hybrid optimisation approach provides a highly accurate, straightforward, and computationally efficient method for finding multi-class CVD. Its excellent ability to forecast and facilitate clear decision-making suggests that it may be a good fit for real-time clinical procedures and wearable ECG monitoring devices.