Deep learning (DL) with infrared meibography can diagnose dry eye disease, according to a study published in The Ocular Surface.

Researchers sought to develop a deep learning-based automated approach to segment meibomian glands (MG) and eyelids, quantitatively analyze the MG area and MG ratio, estimate the meiboscore, and remove specular reflections from infrared images.

The investigators captured a total of 1600 meibography images. They precisely annotated 1000 images with multiple revisions and the images were graded 6 times by meibomian gland dysfunction (MGD) experts. Researchers trained 2 DL models separately to segment areas of the MG and eyelid. They used those segmentations to estimate MG ratio and meiboscores utilizing a classification-based DL model. They implemented a generative adversarial network to remove specular reflections from original images.

The retrospective study included 572 eyes from 320 patients (mean age: 54.6 years, 213 women, 107 men) who visited Yeouido St. Mary’s Hospital for a dry eye exam. According to the researchers, the mean ratio of MG calculated via investigator annotation and DL segmentation was consistent 26.23% vs 25.12% in the upper eyelids and 32.34% vs 32.29% in the lower eyelids, respectively.

Compared with 53.44% validation accuracy by MGD experts, the DL model achieved 73.01% accuracy for meiboscore classification on validation set and 59.17% accuracy when tested on images from an independent center. The DL-based method successfully removes reflection from the original MG images without influencing meiboscore grading, the researchers note.

Reference

Saha RK, Chowdhury AMM, Na KS, et al. Automated quantification of meibomian gland dropout in infrared meibography using deep learning. Ocul Surf. Published online June 24, 2022. doi:10.1016/j.jtos.2022.06.006

Deep Learning Method Can Diagnose Dry Eye Disease

Reference

Want to read more?