Thermal imaging and deep learning-based fit-checking for respiratory protection.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Public Health

Background:

• Proper mask-wearing is crucial for preventing disease transmission.
• Assessing mask fit in real-time remains a challenge.
• Thermal imaging offers a novel approach to detect air leakage.

Purpose of the Study:

• To develop an AI model for real-time mask fit assessment using thermal videos.
• To evaluate the performance of different deep learning models for this task.
• To determine the effectiveness of thermal imaging in identifying correct mask usage.

Main Methods:

• Collected 5000 thermal videos from 50 participants wearing five types of masks incorrectly and correctly.
• Utilized deep learning models, specifically 3D Convolutional Neural Networks (3DCNN) and Convolutional Long Short-Term Memory (ConvLSTM).
• Evaluated model performance using Area Under the Receiver Operating Characteristic Curve (AUROC) for binary and multi-classification tasks.

Main Results:

• The 3DCNN model demonstrated superior performance over ConvLSTM for both binary and multi-classification of mask-wearing methods.
• The highest AUROC achieved was 0.986 for multi-classification using the 3DCNN model.
• All mask types achieved AUROC values greater than 0.9, with KF-AD masks being the best classified.

Conclusions:

• AI-powered thermal imaging provides an effective and accurate method for real-time mask fit-checking.
• The developed model is generalizable across various mask types, offering significant advantages for public and occupational health.
• This technology can enhance safety in high-risk environments and improve healthcare system efficiency.