Correlation Study Of Lower Lid Margin Linearity To Differentiate Diagnosis Meibomian Gland Dysfunction Blepharitis Using A I Image Detection

Posterior blepharitis also known as Meibomian Gland Dysfunction (MGD) is characterized by inflammation of the posterior eyelid margin and associated with changes in eyelid structure and its linearity. It is a common eye condition that affects both children and adults and is often associated with itchiness, redness, flaking, and crusting of the eyelids. Without treatment, it will likely aggravate cornea ocular surface disease and produce evaporative dry eye.  This study aims to analyze the smoothness or linearity of the patient's lower eyelid margin based on artificial intelligence (AI) models and image recognition, to explore its correlation with blepharitis and MGD, so as to provide a basis for clinical diagnosis and early screening.

A retrospective chart review from 2020~2024 of patients for MGD LipiFlow ® and Lumenis M22 Intense Pulse Light (IPL) with dry eye disease were recruited from a Tertiary teaching hospital in northern part of Taiwan. MGD blepharitis were diagnosed base on slit lamp examination by an experience ocular surface ophthalmologist plus a LipiView® exam with meibography showing MG gland atrophy. Normal patients were those for cataract or other ophthalmological disease in the clinic.

External slit lamp images of MGD patients were collected during clinical examination before intervention such as IPL. A total of 65 MGD patients and 54 normal patients were recruited. Eyelid images were processed using deep learning technology to establish an AI analysis model.

For the detection and annotation of lid margin, Canny Edge Detection in OpenCV- Python were used. The optimal function in Canny's detector is described by the sum of four exponential terms and approximated by the first derivative of a Gaussian. The smoothness or linearity of the lid margin of both group was represented by Total variation (TV), and is calculated between the nasal and temporal area, with a smaller total variation (TV) suggests a smoother curve. 

The total variation (TV) is the integral of the absolute value of the derivative over the curve's domain, formulation as below: TV(f′(x))=∫ab∣f′(x)∣ dxTV(f'(x)) = \int_{a}^{b} |f'(x)| \, dxTV(f′(x))=∫ab​∣f′(x)∣dx, base on this, we do the Artificial intelligence model training resulting with Binarary Classification using deep learning networks. 

Explainable AI were use to analyze whether changes in eyelid arc smoothness can be classified. Confusion matrix and AUC show an Accuracy of > 0.8.

Preliminary results showed that the smoothness of the lower eyelid arc was significantly reduced in patient group with blepharitis and MGD(p < 0.05), but the correlation with grade of atrophy of the MGD has not been established.

This study confirmed a significant correlation between the smoothness and linearity of the lower eyelid margin and MGD blepharitis. With our AI ​​analysis model, it can effectively assist in screening and diagnosis of MGD blepharitis. In future, more clinical data can be further combined to improve the accuracy and practicality of AI diagnosis, providing a new direction for the intelligent diagnosis of ophthalmic diseases, with early detection of MGD and less atrophic progress.