All iLive content is medically reviewed or fact checked to ensure as much factual accuracy as possible.
We have strict sourcing guidelines and only link to reputable media sites, academic research institutions and, whenever possible, medically peer reviewed studies. Note that the numbers in parentheses ([1], [2], etc.) are clickable links to these studies.
If you feel that any of our content is inaccurate, out-of-date, or otherwise questionable, please select it and press Ctrl + Enter.
Facial temperature may predict heart disease with greater accuracy than current methods
Last reviewed: 02.07.2025
In a recent study published in the journal BMJ Health & Care Informatics, researchers assessed the feasibility of using facial infrared thermography (IRT) to predict coronary heart disease (CHD).
CHD is one of the leading causes of death and has a significant global burden. Accurate diagnosis of CHD is important for care and treatment. Currently, pre-test probability (PTP) assessment tools are used to determine the likelihood of CHD in patients. However, these tools have problems with subjectivity, limited generalizability, and moderate accuracy.
Although additional cardiovascular testing (coronary artery calcium score and electrocardiography) or sophisticated clinical models integrating additional laboratory markers and risk factors may improve probability estimation, there are issues related to time efficiency, procedural complexity, and limited availability.
IRT, a non-contact surface temperature detection technology, shows promising results for disease assessment. It can detect inflammation and abnormal blood flow from skin temperature patterns. Studies show associations between IRT information and atherosclerotic cardiovascular disease and related conditions.
In this study, the researchers assessed the feasibility of using facial IRT temperature data to predict CAD. Adults undergoing coronary CT angiography (CCTA) or invasive coronary angiography (ICA) were included in the study. Trained personnel obtained baseline data and performed IRT acquisitions before CCTA or ICA.
Electronic medical records were used to obtain additional information, including blood biochemistry, clinical history, risk factors, and CAD screening results. One IRT image per participant was selected for analysis and processed (uniform resizing, conversion to grayscale, and background cropping).
The team developed an IRT image model using an advanced deep learning algorithm. Two models were developed for comparison: one was a PTP (clinical baseline) model that included age, gender, and symptom characteristics of patients, and the other was a hybrid, combining both IRT and clinical information from the IRT and PTP models, respectively.
Several interpretation analyses were performed, including occlusion experiments, visualization of highlight maps, dose-response analyses, and prediction of surrogate CAD labels. In addition, various IRT table features were extracted from the IRT image, classified at the whole-face and region-of-interest (ROI) level.
Overall, the extracted features were classified into first-order texture, second-order texture, temperature, and fractal analysis features. The XGBoost algorithm integrated these extracted features and evaluated their predictive value for CHD. The researchers evaluated the performance using all features and only temperature features.
A total of 893 adults undergoing CCTA or ICA were screened between September 2021 and February 2023. Of these, 460 participants with a mean age of 58.4 years were included; 27.4% were women and 70% had CAD. Patients with CAD had higher age and prevalence of risk factors compared with patients without CAD. The IRT-image model significantly outperformed the PTP model.
However, the performance of the hybrid and IRT image models was not significantly different. Using only temperature features or all extracted features had superior predictive performance, which was consistent with the IRT image model. At the whole-face level, the overall left-to-right temperature difference had the greatest impact, while at the ROI level, the average temperature of the left jaw had the greatest impact.
Varying levels of performance degradation were observed for the IRT-image model when occluding different ROIs. Occlusion of the upper and lower lip region had the greatest impact. In addition, the IRT-image model performed well in predicting surrogate markers associated with CAD, such as hyperlipidemia, smoking, body mass index, glycated hemoglobin, and inflammation.
The study demonstrated the feasibility of using facial IRT temperature data to predict CAD. The IRT image model outperformed the guideline-recommended PTP model, highlighting its potential in CAD assessment. Furthermore, incorporating clinical information into the IRT image model did not provide additional improvement, suggesting that the extracted IRT information already contained important information related to CAD.
Moreover, the predictive value of the IRT model was confirmed using the interpretable IRT table features, which were relatively consistent with the IRT image model. These features also provided information on important aspects for predicting CHD, such as facial temperature symmetry and distribution unevenness. Further studies with larger samples and diverse populations are needed for validation.