Choroid Neovascularization Growth Prediction Based On Optical Coherence Tomography Images

Choroid neovascularization(CNV)is caused by new blood vessels growing in the choroid and penetrating the bruch membrane and is one of the major causes of blindness.Currently,the most effective medical treatment for CNV is intravitreal injections of anti-VEGF medicine.This treatment can inhibit the growth of CNV,but it requires frequent injections,which may result in impairment of sight or complete loss of vision.Optical coherence tomography(OCT)has the advantages of high resolution and non-invasion and is widely used in clinical treatment evaluation and postoperative follow up of choroidal neovascular disease.Thus,the accurate prediction of the growth of choroidal neovascular disease is of great significance to reduce the number of injection and it can help to make personalized treatment plan.In this thesis,we mainly study the prediction of CNV growth based on OCT images,and the main work and contributions are summarized as following:We proposed a method for predicting CNV growth based on OCT images in this thesis.The method mainly includes three parts: preprocessing,modeling and prediction.Firstly,The OCT images are preprocessed.It includes OCT image denoising,the registration between the images collected at different time points,and retinal layer segmentation based on the graph search algorithm.This method can accurately segment the CNV region,inner retinal layer,outer retinal layer and choroidal,respectively.Secondly,it is modeling.The reaction-diffusion model is applied to simulate the growth/shrinkage of CNV volumes and is solved by using finite element method(FEM).Finally,a multi-time point prediction method is proposed.The optimal model parameters are calculated by the genetic algorithm and fitted to get the predicted parameter.Then the predicted parameter is applied to the predicted image,which will be compared with the ground truth image to evaluate accuracy of the prediction.The proposed method is tested on a dataset with 91 longitudinal OCT images and these images are collected from 7 patients who suffering with choroidal neovascular macular degeneration.The mean true positive volume fraction(TPVF),false positive volume fraction(FPVF)and dice similarity coefficient(DSC)are75.00%,2.19% and 76.40%,respectively.The linear regression analysis of the predicted results and the manually segmented ground truth shows that they have a strong correlation.