Regression model for predicting core body temperature in infrared thermal mass screening

Area of Science:

• Medical Devices
• Biomedical Engineering
• Public Health

Background:

• Fever is a key COVID-19 symptom, leading to widespread fever screening.
• Non-contact infrared (IR) screening methods reduce infection risk but may lack core body temperature accuracy.

Purpose of the Study:

• To develop a trained model for predicting core body temperature using IR-measured facial features.
• To assess the accuracy of this prediction method against a reference standard.

Main Methods:

• Five regression models were evaluated for predicting core body temperature from IR facial measurements.
• The linear regression model was selected based on the lowest root-mean-square error (RMSE).
• Temple and nose regions of interest (ROIs) were identified as optimal input parameters.

Main Results:

• The linear regression model demonstrated the lowest RMSE compared to reference core body temperatures.
• The developed model achieved accuracy comparable to an FDA-approved monitoring system.
• The non-contact method showed a mean standard deviation (SD) of ±0.285°C and mean absolute error (MAE) of 0.240°C.

Conclusions:

• IR temperature data, when processed with a linear regression model, can accurately predict core body temperature.
• This method offers a viable solution for rapid mass screening of potential COVID-19 cases.
• Non-contact temperature measurement using this approach is comparable to existing commercial systems.