Researchers have developed a novel deep learning-based algorithm for detecting strabismus by analysing subtle eye gaze patterns – a promising advance that could one day transform early screening and referral pathways.

The new approach, detailed in Cureus,1 leverages deep neural networks to estimate eye gaze direction from simple camera images. By training the model to output precise eye gaze angles, the system can detect discordances between the eyes indicative of strabismus.

A total of 12 subjects were included in the study: two case subjects – the first with no ophthalmologic history and the second with diagnosed exotropia – as well as 10 control subjects without ophthalmologic abnormalities. In Case 1 (no ophthalmologic history), the estimated gaze deviation was 4.3° in the right eye and -0.5° in the left. In Case 2 (diagnosed exotropia), the estimated deviation was 0.7° in the right and -10.1° in the left, closely reflecting the clinical diagnosis.

Among the 10 control subjects, a strong correlation was observed between the gaze angles of both eyes.

The authors concluded that the algorithm “demonstrated potential for quantifying strabismus angles through video-based gaze estimation”.

“Since the input images are captured on video, it is considered useful for cases with torticollis or infants who have difficulty staying in a fixed position.

“Although further validation is required, it is also believed that the video input format will provide a certain level of accuracy, even in cases where the input image is rotated in the camera,” study authors wrote.

Reference available at mivision.com.au.