Automatic identification of meibomian gland dysfunction with meibography images using deep learning

Abstract

BackgroundArti cial intelligence is developing rapidly, bringing increasing numbers of intelligent products into daily life. However, it has little progress in dry eye, which is a common disease and associated with meibomian gland dysfunction(MGD). Non-invasive infrared meibography, known as an effective diagnostic tool of MGD, allows for objective observation of meibomian glands. Thus, we discuss a deep learning method to measure and assess meibomian glands of meibography. MethodsWe used Mask R-CNN deep learning(DL) framework. A total of 1878 meibography images were collected and manually annotated by two licensed eyelid specialists with two classes: conjunctiva and meibomian glands. The annotated pictures were used to establish a DL model. An independent test dataset contained 58 images was used to compare the accuracy and e ciency of the deep learning model with specialists. ResultsThe DL model calculated the ratio of meibomian gland loss with precise values by achieving high accuracy in the identi cation of conjunctiva (validation loss < 0.35, mAP > 0.976) and meibomian glands (validation loss < 1.0, mAP > 0.92). The comparison between specialists' annotation and the DL model evaluation showed that there is little difference between the gold standard and the model. Each image takes 480ms for the model to evaluate, almost 21 times faster than specialists. ConclusionsThe DL model can improve the accuracy of meibography image evaluation, help specialists to grade the meibomian glands and save their time to some extent.

Anterior segment optical coherence tomography meibography compared with keratograph meibography

“…27 In addition, there is scope within ophthalmic imaging to embed deep learning models, and a recent study had success in developing automatic identification of meibomian glands with this method. 28 For future investigations, the utilisation of swept-source OCT with deep learning models could enable volumetric scanning image quality enhancement, automated image analysis and interpretation. However, it is important to note when seeking a clinically applicable method for OCT meibography, such deep learning models are not currently available in general optometric practice and require specialist knowledge and skills to implement.…”

“…Not only would the 3D image provide an added benefit due to the anatomical details of the gland structures, but it would also show conjunctival thickening in patients with MGD 27 . In addition, there is scope within ophthalmic imaging to embed deep learning models, and a recent study had success in developing automatic identification of meibomian glands with this method 28 . For future investigations, the utilisation of swept‐source OCT with deep learning models could enable volumetric scanning image quality enhancement, automated image analysis and interpretation.…”

Potential applications of artificial intelligence in image analysis in cornea diseases: a review

“…Another important predictive factor of DED is meibomian gland dysfunction (MGD) which results in the disruption of the tear film lipid layer and increases the tear film evaporation rate [ 60 ]. A mask region based convolutional neural network (RCNN) algorithm has been explored in achieving automatic identification of MGD by applying it on analyses of 1878 non-invasive infrared meibography images [ 61 ]. It was able to derive the ratio of meibomian gland loss through precise segmentation and identification of conjunctiva and meibomian glands [ 61 ].…”

Integration of Artificial Intelligence into the Approach for Diagnosis and Monitoring of Dry Eye Disease

“…This algorithm achieved an AUC of 0.96, an average precision and recall of 83% and 81%, respectively, and an F-score of 84% [ 91 ]. A DL framework based on a Mask R-CNN was also developed for the segmentation of conjunctiva and MGs, which attained high accuracy in the segmentation of the conjunctiva (mean average precision [mAP] > 0.976, validation loss < 0.35) and MGs (mAP > 0.92, validation loss < 1.0) [ 92 ]. The evaluation of each image using this AI model took 480 ms, 21 times faster than that of ophthalmology specialists [ 92 ].…”

“…A DL framework based on a Mask R-CNN was also developed for the segmentation of conjunctiva and MGs, which attained high accuracy in the segmentation of the conjunctiva (mean average precision [mAP] > 0.976, validation loss < 0.35) and MGs (mAP > 0.92, validation loss < 1.0) [ 92 ]. The evaluation of each image using this AI model took 480 ms, 21 times faster than that of ophthalmology specialists [ 92 ]. In 2022, Saha et al [ 93 ] introduced a classification-based DL model that could enable the fast, automated, and objective assessment of the morphologic features of MGs, i.e., the segmentation of MGs, quantitative analysis of the area and ratio of MGs, and determination of the meiboscore.…”