An Eye Video Oriented and rPPG-Based Intraocular Pressure Detection Method.

Area of Science:

• Ophthalmology
• Biomedical Engineering
• Computer Vision

Background:

• Current intraocular pressure (IOP) measurement relies on inconvenient contact methods.
• There is a need for non-contact, accessible IOP monitoring solutions.
• Ophthalmological diagnostics can benefit from advanced imaging techniques.

Purpose of the Study:

• To develop and validate a novel video-based method for detecting and classifying intraocular pressure (IOP).
• To utilize remote photoplethysmography (rPPG) for non-contact IOP assessment.
• To create an AI model capable of distinguishing between normal and high IOP from eye videos.

Main Methods:

• Developed the IOP-Net model, a convolutional neural network (CNN) with four layers.
• Extracted blood volume pulse (BVP) signals from key regions of interest (ROIs) in eye videos: pupil, iris, and sclera.
• Trained the CNN model using extracted BVP signals for IOP detection and classification.

Main Results:

• The IOP-Net model achieved root mean square errors (RMSE) of 3.14 mmHg on the EVIP-1 dataset and 4.19 mmHg on the EVIP-2 dataset.
• Demonstrated an accuracy of 80.25% in classifying high IOP when the ground truth IOP exceeded 30 mmHg.
• Successfully detected and classified IOP using video-based rPPG signals.

Conclusions:

• The proposed video-based rPPG method offers a promising and potential approach for non-contact IOP detection and classification.
• IOP-Net demonstrates feasibility for widespread application in remote IOP monitoring.
• This innovative technique could enhance the convenience and accessibility of IOP assessment.