Estimation of Total Hemoglobin (SpHb) from Facial Videos Using 3D Convolutional Neural Network-Based Regression.

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Computer Vision

Background:

• Hemoglobin measurement is crucial for diagnosing infections, trauma, and anemias.
• Current methods (blood sampling, pulse oximetry) have limitations like time, cost, and physical contact.
• Non-invasive, contactless methods are needed for remote or emergency settings.

Purpose of the Study:

• To develop and validate a non-contact, automated method for estimating total hemoglobin levels.
• To utilize facial video data and 3D convolutional regression models for hemoglobin estimation.
• To provide a practical alternative to conventional hemoglobin measurement techniques.

Main Methods:

• A dataset of 279 volunteers with synchronized facial video and pulse oximeter hemoglobin data was collected.
• Three-dimensional (3D) convolutional neural network (CNN) regression models (3D CNN, attention-enhanced 3D CNN, residual 3D CNN) were trained.
• The performance of the models was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Pearson correlation coefficient.

Main Results:

• The residual 3D CNN model demonstrated the best performance on the test set.
• Achieved an RMSE of 1.06, MAE of 0.85, and a Pearson correlation coefficient of 0.73.
• The developed model was implemented in a user-friendly graphical interface.

Conclusions:

• Contactless hemoglobin estimation from facial video is feasible using 3D CNN regression.
• This approach offers a promising, non-invasive tool for hemoglobin level assessment.
• Potential applications include remote patient monitoring and rapid diagnostic screening.