| Retinal disease is the most common disease of ophthalmic diseases,which has seriously affected human’s vision,and it has aroused wide concern because of the growing number of patients suffering from this.The change of the retinal layer tissue structure is one of the most obvious features of the fundus disease,so it is possible to directly observe the change in the thickness of the retina or even the disappearance of some layers of the retina.Optical coherence tomography(OCT)is a new fundus imaging technology with high image resolution,no harm to the eyes,no need to contact the eyes,etc.It can directly display the various cell layers of the retinal layer structure.Doctors can directly observe the subtle changes of the layer to diagnose the disease and now it has become an important routine examination in clinical treatment.While the precise segmentation of the retinal layer and the quantitative analysis of the thickness of the retinal layer have become important prerequisites and guarantees for disease diagnosis.At present,manual segmentation of retinal layers by doctors still occupies a dominant position,which is not only time-consuming and laborious,but also subjective and affects the accuracy of image segmentation.However,the existing computer-aided segmentation requires segmentation of the layer structure according to the characteristics of each layer of the retina.Although computer-aided algorithms have been able to segment the retinal layer automatically,and can reduce the manual segmentation time and improve the segmentation efficiency of the doctor.But the current mainstream algorithms are affected by the image quality.If the image noise is loud and the retinal boundary is not obvious,the segmentation result will be severely affected,resulting in low accuracy of retinal segmentation.At the same time,clinical retinal diseases are very complicated,and there are a variety of symptoms pertaining to this.However,most of the current methods only focus on the segmentation of a certain layer,so it is difficult to analyze complex diseases,which makes the retinal segmentation process still faces great challenges.Therefore,this paper proposes an OCT retinal layer structure segmentation method based on deep learning.By learning the A-scan curve matching relationship,the segmentation result or offset field is obtained.Solved the shortcoming of using a method to split all layers in the traditional method..The main contributions of this research are as follows:1.This paper uses the processing of two or more columns of A-scan convolutions to strengthen the connection of each pixel between the two columns to achieve image segmentation.The segmentation result can be got by matching the gray value curve to get the offset between the curves,and all the layers in the retina can be segmented by a model instead of a certain layer.2.This paper uses FCN and U-Net convolutional neural network to realize the calculation between A-scans.This model can segment the retinal layer without human involvement.Besides,it can not only segment the retinal layer of normal people,but also segment images of patients.3.In this paper,the offset field composed of the offsets at the corresponding positions between the two columns of A-scans is calculated,and the required layers are manually selected through the boundaries of the offset fields to achieve OCT retina segmentation.The experiment proves that the segmentation method of OCT retinal layer structure based on curve group matching is the closest to the doctor’s segmentation results.Detecting changes in the layer structure of the retina is an important diagnostic and therapeutic step,so quantitative analysis of the thickness of the layer structure of the retina is much more valuable.This paper uses SPSS to analyze the relationship between retinal thickness and gender in normal people,and compares the thickness differences in each retinal layer and all layers of normal people by taking pigment epithelial detachment disease as an example. |
PDF Full Text Request