Quantitative Assessment of Strabismus and Selected Vision Related Anomalies

Abstract

Decreased vision, double vision, eye fatigue and strain associated with strabismus often require no-surgical or surgical options which may lead to overcorrection or under correction with a need for a follow up surgery. Therefore, quantitative assessment of the degree of strabismus can serve as very useful tool for deciding on therapeutic options and evaluation of their outcomes. Further, US focused statistics indicate that 4% of the population has strabismus while some global estimates attribute this anomaly only to 0.034% of world population. These contrasting statistics further lead to a necessity of having a uniform quantitative tool for a broader application to determine the scope and the degree of this anomaly in different populations. At this point variety of tests are used including Hirschberg test, Cover test, and Central Corneal Light Reflex Ratio. Therefore, the developed technique allows automatic quantitative detection of a presence of possible strabismus and calculation of linear and vertical deviations of eyes in digital images. In particular, the proposed algorithm was structured in seven stages: (1) face matching (2) face detection and alignment (3) extraction of region of interest (4) locating the iris of both eyes and their center positions (5) selection of reference points in the eyes (6) calculation of linear and vertical deviations (7) making prediction using pre-trained regression model. This methodology has 93% of accuracy, 84% of sensitivity and 30% of specificity as tested on 128 images. In particular, the outcome encompasses a methodology for two graphical user interfaces which have real time as well as local image processing capability; a bounding box approach to make the face of a person aligned; and determination of numerical linear and vertical deviations of the eyes in millimeters. While the deviation of normal eyes is close to zero, the higher numbers indicate pre-strabismus or strabismus conditions respectively.