Article Impact Level: HIGH Data Quality: STRONG Summary of BMJ Health & Care Informatics, 31(1), e100942. https://doi.org/10.1136/bmjhci-2023-100942 Minghui Kung et al.
Points
- Researchers explored using facial infrared thermography (IRT) combined with machine learning to predict coronary artery disease (CAD) in 460 participants undergoing coronary angiography or CT angiography.
- The developed deep-learning IRT image model showed strong predictive performance with an area under the curve (AUC) of 0.804, outperforming traditional pretest probability (PTP) models (AUC 0.713).
- Interpretable IRT tabular features extracted from images confirmed the high predictive value of IRT information, maintaining a high-performance level (AUC 0.796).
- The IRT model achieved satisfactory performance even using only traditional temperature features (AUC 0.786).
- The study demonstrates facial IRT’s and machine learning’s feasibility and effectiveness for non-invasive CAD prediction, offering a promising alternative to traditional methods.
Summary
In a recent study, researchers investigated the potential of using facial infrared thermography (IRT) combined with machine learning for predicting coronary artery disease (CAD). The study enrolled 460 participants referred for invasive coronary angiography or coronary CT angiography (CCTA) and utilized facial IRT images captured before confirmatory CAD examinations. A deep-learning IRT image model was developed and validated, showcasing remarkable performance with an area under the curve (AUC) of 0.804, significantly outperforming the pretest probability (PTP) models recommended by guidelines (AUC 0.713). Additionally, interpretable IRT tabular features extracted from the images further confirmed the predictive value of IRT information, with a consistently high performance level (AUC 0.796) maintained.
The study’s results highlight the promising role of non-contact facial IRT information in CAD prediction. The comprehensive analysis revealed that the IRT image model and interpretable IRT features provided superior predictive capabilities compared to traditional PTP models. Even with only traditional temperature features, the IRT model maintained a satisfactory performance level (AUC 0.786). This prospective study underscores the feasibility and effectiveness of leveraging facial IRT combined with machine learning techniques for enhancing CAD assessment, offering a potential non-invasive and efficient approach for predicting this cardiovascular condition.
Link to the article: https://informatics.bmj.com/content/31/1/e100942
References Kung, M., Zeng, J., Lin, S., Yu, X., Liu, C., Shi, M., Sun, R., Yuan, S., Lian, X., Su, X., Zhao, Y., Zheng, Z., & Ji, X. (2024). Prediction of coronary artery disease based on facial temperature information captured by non-contact infrared thermography. BMJ Health & Care Informatics, 31(1), e100942. https://doi.org/10.1136/bmjhci-2023-100942
