Font Size: a A A

Automated Corneal Boundary Segmentation And Wound Analysis Based On AS-OCT Images

Posted on:2019-12-13Degree:MasterType:Thesis
Country:ChinaCandidate:F WangFull Text:PDF
GTID:2404330545471731Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
Cornea is the most thin and transparent tissue structure in the outer layer of the human eye.It is vulnerable to various kinds of trauma.After corneal surgery,the healing of scar and sear elimination play a key role in restoration of normal vision.Therefore,the analysis of wound involution and recovery after corneal surgery has positive clinical significance for the diagnosis and treatment.At present,the research on cornea is mainly focused on corneal layers and corneal histocyte,but the quantitative analysis of corneal wound is still lacking.For the above purpose,we propose a method of corneal boundary segmentation and wound analysis based on anterior segment optical coherence tomography(AS-OCT)images.The method mainly includes four steps.Firstly,preprocessing the corneal image collected on AS-OCT by gray level transformation and morphological operation.Secondly,using improved Mini-Unet convolution neural network to segment cornea and detect the corneal boundary.Thirdly,62 features of grayscale,thickness and texture of cornea are extracted from corneal image patch to train random forest classifier,and detect the wound location on cornea.The positions of some wrong wound region are corrected by K-Means clustering method.Lastly,the changes of corneal wound at multiple time points are analyzed by optical density,thickness difference and wound area indicators.The method of corneal boundary segmentation based on Mini-Unet is tested on 39 corneal images.The similarity between the result of region segmentation and the ground truth is up to 95%,and the unsigned distance error of upper and lower boundaries is 1.29 ± 1.16 pixels.The random forest classifier is test on 21773 samples.The precision,recall,mean false error and mean squared false error are 90.91%,99.86%,1.06%and 13.69%,respectively.The corneal wound analysis algorithm proposed in this paper is helpful for doctors to judge the success rate of surgery,to find out the bad factors affecting corneal healing in time and to adjust the treatment prescription,which is of great significance to the recovery of cornea after operation.
Keywords/Search Tags:anterior segment optical coherence tomography(AS-OCT), corneal imaging, boundary segmentation, wound detection, wound analysis
PDF Full Text Request
Related items