Article Impact Level: HIGH Data Quality: STRONG Summary of Frontiers in Cardiovascular Medicine, 11, 1384977. https://doi.org/10.3389/fcvm.2024.1384977 Dr. Mengyao Duan et al.
Points
- Researchers in China used advanced imaging and machine learning techniques to investigate the diagnostic value of tongue images for coronary artery disease (CAD) in 684 patients across four hospitals.
- Various models, such as Decision Trees, Random Forests, Logistic Regression, Support Vector Machines, and XGBoost, were applied, with XGBoost selected for the final CAD diagnostic model.
- Tongue images alone classified CAD patients with 67% accuracy and an AUC of 0.690, while a model based on traditional risk factors achieved a 73% accuracy and an AUC of 0.763.
- Incorporating tongue image features improved the model’s performance, boosting accuracy to 76%, recall to 85%, and AUC to 0.786, demonstrating the added value of tongue imaging.
- Tongue imaging is a promising, noninvasive, and cost-effective method for CAD diagnosis. It could potentially serve as new biomarkers and risk indicators for large-scale screenings.
Summary
In a pioneering study aimed at advancing coronary artery disease (CAD) diagnostics, researchers recruited 684 patients from four hospitals in China to ascertain the diagnostic value of tongue images. By collecting baseline information and standardized tongue images, the team utilized advanced imaging techniques such as DeepLabV3+ for segmentation and Resnet-18 for feature extraction. The study employed various machine learning models, including Decision Trees (DT), Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), and XGBoost to evaluate the integration of tongue image features with traditional risk factors in CAD diagnostic models. The comparative analysis focused on accuracy, precision, recall, F1-score, area under the precision-recall curve (AUPR), and area under the curve (AUC).
The results demonstrated that tongue images could effectively classify patients with CAD, showing a standalone accuracy (ACC) of 0.670 and an AUC of 0.690. Among the algorithms tested, XGBoost was selected for developing the final CAD diagnostic model. Initially, a model based solely on traditional risk factors achieved an ACC of 0.730, a precision of 0.811, and an AUC of 0.763. However, incorporating tongue image features significantly improved the model’s performance, achieving an ACC of 0.760, precision of 0.773, recall of 0.850, and an AUC of 0.786.
The study concludes that tongue imaging is a viable method for enhancing CAD diagnosis, supporting the feasibility of this non-invasive, simple, and cost-effective approach for large-scale screenings. The integration of tongue image features improved the diagnostic accuracy and suggested their potential as new biomarkers and risk indicators for CAD, offering a promising avenue for clinical application and further research in the field.
Link to the article: https://www.frontiersin.org/journals/cardiovascular-medicine/articles/10.3389/fcvm.2024.1384977/full
References Duan, M., Mao, B., Li, Z., Wang, C., Hu, Z., Guan, J., & Li, F. (2024). Feasibility of tongue image detection for coronary artery disease: Based on deep learning. Frontiers in Cardiovascular Medicine, 11, 1384977. https://doi.org/10.3389/fcvm.2024.1384977
