Machine Learning based Myocardial Infarction Risk Stratification as a Diagnostic Aid for Remote Areas with Limited Medical Resources.

Background: This study highlights the vital role of Machine Learning in aiding myocardial infarction (MI) diagnosis, crucial in remote areas with limited medical resources. By leveraging ML algorithms and accessible patient data, it offers a valuable tool for early MI detection and risk assessment in underserved regions, potentially improving patient outcomes and healthcare delivery. Methods: In this case-control study, data from 1,200 individuals (300 MI, 900 non- MI) were collected. Significant variables were identified using correlation. Eight ML models were built based on the patient's historical 24 variables and evaluated using the F1 score, Cohen's Kappa, and AUROC. We also conducted real-time clinical validation to assess the practical applicability of the model. Results: In terms of training time, logistic regression (LR) with L2 regularization, AdaBoost, and XGBoost models showed significantly higher times (410ms, 520ms, and 220ms, respectively). LR had the lowest errors (1.67% training, 1.11% testing) and achieved a high accuracy of 96%, notable precision, recall, and an impressive AUC of 98.87%. In real-time clinical validation, LR and XGBoost performed exceptionally well, boasting F1 scores of 96.27% and 98.70%, respectively, solidifying their effectiveness for predictive accuracy in a clinical setting. Conclusion: In real-time clinical validation, LR and XGBoost based on patient's historical data showcased impressive predictive power, highlighting their potential in clinical settings. These models can be helpful to improve the diagnosis of MI in Remote Areas with Limited Medical Resources. [ABSTRACT FROM AUTHOR]