This study was conducted to develop and validate an artificial intelligence (AI) segmentation algorithm for the standardized, objective assessment of eyelid and periorbital soft tissue position.
Study design
A total of 418 photographs were used to train a deep-learning semantic model algorithm to segment brow, iris, and aperture areas. Measurements included in the algorithm were margin reflex distance (MRD) 1 and 2, lateral canthal height (LCH), medial canthal height (MCH), lateral brow height (LBH), medial brow height (MBH), lateral intercanthal distance (LID), and medial intercanthal distance (MID). Three human graders were trained on the same measurements, and 20% of the images were used to help validate the algorithm. Primary outcome measures included mean absolute difference and dice and intraclass correlation coefficients.
Outcomes
There was close agreement between the AI algorithm and all human graders, with a mean absolute difference of 0.5 mm for MRD1, MRD2, LCH, and MCH, mean absolute differences between 1.5 and 2 mm for LBH and MBH, and a mean absolute difference of about 2.4 mm for LID and MID.
Limitations
The human graders included a medical student, a resident, and an oculoplastic surgeon. To ensure better data, all of the graders should have been oculoplastic surgeons. No comparisons were made between the performance of the AI system and standard clinical practice methods.
Clinical significance
The days of using a ruler to obtain eyelid measurements may be coming to an end. The validity of AI to make eyelid measurements will allow more accurate, objective, reliable, and reproducible pre- and postoperative data in the evaluation of eyelid and orbital disease.