Research On Quantification Method Of Fundus Vessel Based On OCT Images

With the development of society,the number of people suffering from eye diseases in people of all ages is increasing rapidly.At present,most eye diseases are manifested in the condition of blood flow in the fundus,and the eyeball,as a precision organ of the human body,requires safe and high-precision equipment for detection.OCT fundus images not only have the advantages of traditional two-dimensional ophthalmology images,but also contain deep cell tissue and blood flow information in the fundus.More importantly,compared to OCTA images,only part of the capillary information is missing.Therefore,ordinary OCT images can be combined with computer image processing technology to segment blood vessels in OCT images and quantify their characteristics to achieve the effect of OCTA images and assist doctors in the diagnosis of patients’ fundus diseases.The work of this paper is mainly based on the quantification method of fundus blood vessels based on OCT images.Perform preprocessing operations on OCT images,denoise enface OCT images,perform blood vessel shadow segmentation on RPE layer enface OCT images,and perform blood vessel segmentation on OCT b-scan images,and automatically calculate the fundus blood flow based on the segmentation results.Quantitative indicators to help doctors more accurately judge the patient’s condition and make an accurate diagnosis.The main contents of this paper are as follows:(1)In the preprocessing stage of the OCT image,the original data of the OCT image is processed,including the acquisition of the original image,data cleaning,registration,layering and projection,and fully understand the characteristics of the OCT image.(2)In the process of enface OCT image noise removal,this paper proposes a twodimensional transform domain Fourier filter method to remove directional noise from enface image,using Fourier transform(FT),wavelet transform(WT)and non-subsampling Contourlet transform(NSCT)to transform or decompose enface image,and two-dimensional Fourier filter to denoise the transformed or decomposed image,and finally reconstruct the angiography image to obtain en with higher contrast and sharpness face OCT image.(3)In the process of segmentation of blood vessel shadow in RPE layer enface OCT image,this paper proposes an improved Unet blood vessel segmentation algorithm,which cuts the original image and increases the number of training samples,so that the overall parameter of the model is small,and the image processing speed is small.It is fast and has excellent performance in segmenting blood vessel shadows compared to traditional algorithms and the original Unet network.(4)In the process of OCT b-scan image vessel segmentation,this paper proposes a knowledge-injected fully convolutional neural network to segment retinal vessels from OCT b-scan images.The results prove that the deep network is combined with an appropriate prior The combination of knowledge can have a significant improvement effect in specific medical image processing tasks.(5)In the calculation process of fundus vessel,this article mainly uses the existing image processing technology to automatically extract the quantitative indicators of fundus vessels to facilitate the early diagnosis of the disease,including the location of the end points,continuous points,bifurcation points and intersections of vessels,vessel length measurement,vessel direction detection and vessel diameter measurement.