Automated Vision-Based High Intraocular Pressure Detection Using Frontal Eye Images

Area of Science:

• Ophthalmology
• Computer Vision
• Medical Imaging

Background:

• Glaucoma, a leading cause of vision loss, is often linked to elevated intraocular pressure (IOP).
• Current computer vision methods for IOP assessment typically require optic nerve fundus images.
• Early detection of increased IOP is vital for preventing vision impairment.

Purpose of the Study:

• To develop a novel, vision-based framework for initial intraocular pressure (IOP) screening using only frontal eye images.
• To introduce a fully convolutional neural network (FCN) for sclera and iris segmentation from frontal eye images.
• To establish a correlation between frontal eye features and IOP status.

Main Methods:

• Utilized a fully convolutional neural network (FCN) for segmenting sclera and iris from 400 frontal eye images.
• Extracted six features: mean sclera redness, red area percentage, Pupil/Iris diameter ratio, and three novel sclera contour features.
• Applied Support Vector Machine (SVM) and Decision Tree classifiers for IOP status determination (normal vs. high).

Main Results:

• The proposed framework achieved an overall accuracy exceeding 97.75% when using the Decision Tree classifier.
• Successfully segmented sclera and iris using FCN on frontal eye images.
• Introduced novel sclera contour features for IOP status determination.

Conclusions:

• The developed framework demonstrates high accuracy in initial IOP screening using frontal eye images.
• The novel FCN architecture and sclera contour features contribute significantly to IOP assessment from non-invasive imaging.
• This approach offers a promising, accessible method for early glaucoma detection.