Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fever Detection with Infrared Thermography: Enhancing Accuracy through Machine Learning Techniques

Parsa Razmara, Tina Khezresmaeilzadeh, B. Keith Jenkins

TL;DR

This study tackles fever detection with Infrared Thermography by integrating machine learning and carefully engineered features to estimate oral temperature from non-contact measurements. Through a rigorous workflow—from data cleaning and feature extraction to correlation-based selection and SBS—the authors demonstrate that a final engineered feature set, combined with both traditional regression and CNN-based models, achieves state-of-the-art accuracy (RMSE as low as $0.2223$ with a 1D CNN and $0.2296$ with Binning). The results surpass prior single-feature approaches and highlight the value of combining physiological knowledge with statistical validation for robust non-contact diagnostics. The approach is positioned for generalization to other remote sensing biomedical tasks, potentially improving public health screening and non-invasive parameter estimation.

Abstract

The COVID-19 pandemic has underscored the necessity for advanced diagnostic tools in global health systems. Infrared Thermography (IRT) has proven to be a crucial non-contact method for measuring body temperature, vital for identifying febrile conditions associated with infectious diseases like COVID-19. Traditional non-contact infrared thermometers (NCITs) often exhibit significant variability in readings. To address this, we integrated machine learning algorithms with IRT to enhance the accuracy and reliability of temperature measurements. Our study systematically evaluated various regression models using heuristic feature engineering techniques, focusing on features' physiological relevance and statistical significance. The Convolutional Neural Network (CNN) model, utilizing these techniques, achieved the lowest RMSE of 0.2223, demonstrating superior performance compared to results reported in previous literature. Among non-neural network models, the Binning method achieved the best performance with an RMSE of 0.2296. Our findings highlight the potential of combining advanced feature engineering with machine learning to improve diagnostic tools' effectiveness, with implications extending to other non-contact or remote sensing biomedical applications. This paper offers a comprehensive analysis of these methodologies, providing a foundation for future research in the field of non-invasive medical diagnostics.

Fever Detection with Infrared Thermography: Enhancing Accuracy through Machine Learning Techniques

TL;DR

This study tackles fever detection with Infrared Thermography by integrating machine learning and carefully engineered features to estimate oral temperature from non-contact measurements. Through a rigorous workflow—from data cleaning and feature extraction to correlation-based selection and SBS—the authors demonstrate that a final engineered feature set, combined with both traditional regression and CNN-based models, achieves state-of-the-art accuracy (RMSE as low as with a 1D CNN and with Binning). The results surpass prior single-feature approaches and highlight the value of combining physiological knowledge with statistical validation for robust non-contact diagnostics. The approach is positioned for generalization to other remote sensing biomedical tasks, potentially improving public health screening and non-invasive parameter estimation.

Abstract

The COVID-19 pandemic has underscored the necessity for advanced diagnostic tools in global health systems. Infrared Thermography (IRT) has proven to be a crucial non-contact method for measuring body temperature, vital for identifying febrile conditions associated with infectious diseases like COVID-19. Traditional non-contact infrared thermometers (NCITs) often exhibit significant variability in readings. To address this, we integrated machine learning algorithms with IRT to enhance the accuracy and reliability of temperature measurements. Our study systematically evaluated various regression models using heuristic feature engineering techniques, focusing on features' physiological relevance and statistical significance. The Convolutional Neural Network (CNN) model, utilizing these techniques, achieved the lowest RMSE of 0.2223, demonstrating superior performance compared to results reported in previous literature. Among non-neural network models, the Binning method achieved the best performance with an RMSE of 0.2296. Our findings highlight the potential of combining advanced feature engineering with machine learning to improve diagnostic tools' effectiveness, with implications extending to other non-contact or remote sensing biomedical applications. This paper offers a comprehensive analysis of these methodologies, providing a foundation for future research in the field of non-invasive medical diagnostics.
Paper Structure (37 sections, 2 figures, 6 tables)

This paper contains 37 sections, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset
  4. Data Allocation
  5. Validation Set and Hyperparameter Tuning
  6. Preprocessing
  7. Handling Missing Data
  8. Feature Transformation for Temperature Readings
  9. Standardization of Numerical Features
  10. Feature Extraction
  11. Incorporating New Features with Polynomial Interaction Terms
  12. Enhancing Model Predictions with Replicated Feature Engineering
  13. Encoding and Developing New Features from Categorical Variables
  14. Feature Selection
  15. Biological and Physiological Analysis
  16. ...and 22 more sections

Figures (2)

  • Figure 1: Overview of the system workflow, illustrating stages from data preprocessing to model training and evaluation. The process includes initial data sanitization, feature encoding, polynomial and categorical feature engineering, biological and technical analysis for feature selection, and model training with performance evaluation.
  • Figure 2: RMSE Variation with Repetitions of Feature $T\_Max\_1$: Shows the relationship between RMSE and the number of repetitions for $T\_Max\_1$. A minimum RMSE is observed at five repetitions, beyond which additional repetitions do not improve model performance.