Research On Dry Eye Detection Based On Image Processing

In recent years,the incidence of dry eye in China has increased year by year,and the traditional invasive detection methods have the problems of slow speed and secondary injury.The research on non-invasive medical image dry eye detection based on image processing has high practical value.Therefore,this thesis studies the dry eye diagnosis of the tear film break time and the dry eye diagnosis method of the meibomian gland.The main work is as follows:(1)Analyzed the dry eye test standard formulated by the Chinese Medical Association,which indicated that the tear film was diagnosed as dry eye when the break time was less than 5s,the diagnosis was normal when it was greater than 10 s,and the meibomian gland test was performed between 5s and 10 s.Therefore,this paper combined with Liaoning Heshi Eye Industry Group to collect eye image data to establish tear film break time image library and meibomian gland image library to provide data support for dry eye detection.(2)The dry eye detection method based on Placido ring for tear film rupture time is studied.The detection process includes applying adaptive filtering and closing operation to remove the eyelash image,and using the elliptical scanning method to mark the contour of the ring for Placido ring recognition.Take the first frame image after opening eyes as the template,then match each subsequent frame to the first frame separately,and finally the matched image and the first frame image are subtracted to find the tear film rupture position,and the break time of tear film is calculated by using the number of frames of the broken image,and based on this time to diagnose dry eye.(3)A dry eye diagnosis method based on the ratio of the meibomian gland area to the eyelid area is studied.When the tear film break time cannot be completely diagnosed as dry eye,the ratio of the meibomian gland area to the eyelid area is calculated.In order to segment the meibomian gland region from the image,this paper firstly applied the morphology algorithm,Wallis algorithm and improved Mask algorithm to preprocess,and found that the improved Mask algorithm is the best.Secondly,the eyelid region is segmented by the minimum cross entropy and the eyelid portion is obtained through the maximum connected domain;the aspect ratio of the slab eccentric rectangle of the meibomian gland image was used again to extract the glandular images;finally,the ratio of the glandular area to the eyelid area was calculated.The experimental results show that the dry eye can be diagnosed when the proportion of the meibomian gland is less than 23%,and the diagnosis is normal when the ratio is greater than 23%.Then the image data is re-acquired to form a verification set to verify the correctness and accuracy of the proposed method.The experimental results show that the accuracy based on the tear film break time detection method is 80.80%,the accuracy of the method is 78.48% which based on the ratio of the meibomian gland area to the eyelid area ratio dry eye detection.Finally,the algorithm is evaluated by using indicators such as accuracy,sensitive,specificity,and precision,the above indicators verify that the proposed method can correctly diagnose dry eye.