Objective:

The study assessed the capability to quantify eye movements and lid position in
patients with MG.

Background:

Ptosis and diplopia are nearly uniformly present in patients with myasthenia gravis
(MG). In the telemedicine setting, the unique presence of video capture lends use of emerging implementation of image-processing techniques that could provide precise, standardized assessment of ocular motility abnormalities.

Design/Methods:

We utilized video recordings of the ADAPT teleMG study performed. Six subjects
underwent standardized video examinations on two occasions 2-3 days apart. We performed a hybrid algorithm combining deep learning with computer vision to give quantitative metrics of ptosis and ocular motility.

Results:

All videos and still images could be used for analysis despite limitations of low
resolution dictated by internet bandwidth and cameras of the examiner and subject. The hybrid algorithm detected the distance between pupil and upper eyelid for ptosis and misalignment of the left and right pupil for diplopia assessment. The algorithm automatically detected the pupillary and lid margins with error of 2 pixels in average. Lid, pupil, and iris detection did not differ based on skin or iris color. We were able to detect ocular misalignment in subjects complaining of diplopia. The method worked both on a fixed image and video in real-time allowing capture of the dynamic changes during the examination.

Conclusions:

We were able to use standard Zoom recordings of lid and eye movements and
quantitate them using a hybrid deep learning computer vision algorithm. The system introduces modern image processing technique that are quick and robust to recover quantitative metrics that are independent of the examiner and would work on most video resolutions. The approach has the potential to quantitate ocular motility abnormalities and could be used to monitor patients.