A new method based on deep learning and image processing for detection of strabismus with the Hirschberg test

Area of Science:

• Ophthalmology
• Computer Science
• Medical Imaging

Background:

• Strabismus, affecting 4% of the global population, is a condition where eyes misalign.
• Traditional strabismus detection methods include the Hirschberg, Cover, and Krimsky tests.
• The Hirschberg test measures the corneal light reflex's position relative to the pupil to assess eye alignment.

Purpose of the Study:

• To develop and evaluate a deep learning and image processing algorithm for automated strabismus screening.
• To measure the degree of strabismus using an automated Hirschberg test on digital facial images.
• To assess the accuracy and applicability of the algorithm across diverse patient demographics.

Main Methods:

• Utilized deep learning and image processing to detect key eye features: pupil, iris, and corneal reflection.
• Implemented an automated Hirschberg test to calculate horizontal and vertical eye shifts from facial images.
• Validated the algorithm on 88 strabismic patients of various ages and sexes.

Main Results:

• The algorithm achieved high accuracy, correctly measuring the left eye in 91% (80/88) and the right eye in 90% (79/88) of patients.
• Measurements showed a maximum error of ±2°, attributed to the 2D representation of a 3D eye.
• The method demonstrated robustness, unaffected by patient age or facial morphology.

Conclusions:

• The developed software successfully automates strabismus screening via the Hirschberg test.
• This AI-driven approach provides accurate and reliable measurements for strabismus detection.
• The algorithm shows potential for widespread use in clinical settings for strabismus screening across all age groups.