Research Of Image Processing Algorithm For Near-infrared Meibomian Image

Dry eye,a eye disease,caused by tear film breakdown due to the abnormal tear film,induce a uncomforted feeling in the eyes.Clinical diagnosis found that more than a half of the dry eye were caused by meibomian gland dysfunction.Now meibomian gland dysfunction is diagnosis by slit-lamp in clinical.Slit-lamps are used to observe eyelid,meibomian glands and their secretions directly.Because the meibomian glands are covered by the conjunctiva and epithelial cells,the morphology changes of the meibomian glands cannot be observed by slit lamps.The diagnostic accuracy and sensitivity depends largely on the experiences of clinicians.Recent research results show that the near-infrared light have greater penetration depth on human tissues and cells and meibomian gland near-infrared imaging technology has become an important method for the diagnosis of meibomian gland dysfunction.However,after the acquisition of meibomian gland image,physicians shall use image software,for example ImageJ,manually define the meibomian plate region and the meibomian glands region in the meibomian gland near-infrared image.The manual drawing process is relatively cumbersome,time consuming,less accurate,and it is easy to introduce the errors.To solve the above problem,the image processing method was used to automatically analyze the meibomian gland near-infrared light image in this theisis.The image pretreatment algorithms,image enhancement algorithms,meibomian glands region segmentation algorithms,meibomian glands miss region detection algorithms wear separately used to calculate the area loss rate of the meibomian glandand define the grade of the meibomian gland dysfunction.So the automatic diagnosis of meibomian gland dysfunction was realized,avoiding the tedious manual analysis process and improving the objectivity of diagnosis results.The main contents of this paper includes the following sections:(1)Design of meibomian gland near-infrared imaging device.The device for capturing meibomian gland near infrared imaging was designed based on the imaging characteristics of the meibomian glands.The meibomian glands near infrared imaging device prototype was developed according to the design.(2)Research on the preprocessing algorithm for the meibomian glands near infrared image.Firstly the median filtering algorithms and the Wiener filtering algorithm were used separately to reduce image noise.Then using the histogram specification algorithm adjusted the Meibomian gland near-infrared image contrast.Lastly Wallis sharpening algorithm was used to enhance image quality.(3)Research on the automatic segmentation algorithm for the meibomian glands near infrared image.Firstly,OSTU algorithm was used to segment the image,so that the meibomian tarsal region and the eyelid region were identified as a target area.Followed by etching algorithm,the largest connected domain extraction algorithm,and the expansion algorithm processing,meibomian tarsal region was segmented.Finally,convex hull algorithm was used to repair defect edge curve of the meibomian tarsal region.(4)Research on the recognition algorithm for the meibomian glands near infrared image to calculate the missing gland area.Firstly,the automatic threshold iterations algorithm,the largest connected region extraction algorithm,the outer edge extraction algorithm and the corrosion algorithm were used to recognize the gland within the meibomian tarsal region,and calculate the area loss rate of meibomian gland.Then the image difference algorithm was used between meibomian gland region image and the gland region image to get the region of meibomian tarsal region.Then the area loss rate of the meibomian gland was calculated.The grade of the meibomian gland dysfunction was defined according to the area loss rate of the meibomian gland.Finally,the meibomian gland processing algorithm developed in this thesis was test.Test result shows that the absolute value of the deviation between the algorithm and using ImageJ software manually define the area is <5%.So the algorithm meets the needs of meibomian gland dysfunction diagnostic.