Development of an Artificial Intelligence Algorithm to Estimate Tear Meniscus Height Using Anterior Segment Optical Coherence Tomography and a Smartphone-Based Slit-Lamp Microscope

To enable tear meniscus height measurement independently of the clinical setting, we developed an artificial intelligence-based estimation model using a smartphone-mounted slit-lamp microscope (Smart Eye Camera) and anterior segment optical coherence tomography data. In this prospective study, we analyzed data from 533 eyes of 298 patients who visited Keio University Hospital between March 2021 and January 2025. On the same day, both anterior segment optical coherence tomography and Smart Eye Camera video recordings were obtained. Machine learning was performed using the Light Gradient Boosting Machine to estimate tear meniscus height based on these datasets. After appropriate preprocessing, the artificial intelligence model estimated tear meniscus height with a mean absolute error of 0.113 ± 0.094 mm and a mean squared error of 0.021 ± 0.035 mm, indicating moderate accuracy. We successfully developed an artificial intelligence algorithm capable of estimating tear meniscus height from Smart Eye Camera video and anterior segment optical coherence tomography data by identifying relevant anatomical landmarks. Although the results are promising, the current estimation accuracy remains limited. Future work will focus on improving performance by incorporating additional features such as iris segmentation, increasing the number of training samples, and applying more advanced machine learning techniques. As the present study was conducted in a Japanese cohort, the findings should be interpreted within the context of an Asian population.