Objective Image Quality Assessment For Optical Coherence Tomography Retinal Images

Optical coherence tomography(OCT)imaging technology is widely used in fundus image acquisition.However,the imaging quality of OCT images is difficult to control,and OCT images of poor quality would affect the doctor’s diagnosis of the disease.Therefore,an objective OCT image quality evaluation algorithm is needed to help doctors perform image screening.Since it has not been long from the invention of OCT technology,the performances of current quality evaluation methods still need to be improved.In this thesis,the application of deep neural network model in OCT image quality assessment and training methods are studied for the characteristics of fundus OCT images.The main research contents include:(1)An objective quality assessment algorithm for OCT images based on the hybrid model of deep residual network and SVR is proposed.The residual image perception features related to the imaging quality of OCT images are extracted by the residual network,and the mapping relationship between perception features and subjective labels is established by SVR model.The most concerned part of OCT images for doctors is retinal structure,which locates at the high-frequency region.So,a high-pass Gaussian filtering method is adopted.The experimental results show that the correlation between the algorithm and the doctor’s subjective evaluation reaches 0.965,which has strong consistency.(2)An objective quality assessment algorithm for OCT images based on learning to rank and shallow residual network is proposed.based on the idea of learning to rank,rank constraints are introduced in the training process of neural networks.At the same time,considering that the OCT images contain less information than natural images,a simplified network is proposed.Finally,we prove the effectiveness of the algorithm in the experiment,and the correlation with the doctor’s subjective evaluation reached 0.98.(3)The quality assessment algorithm in this thesis is applied to the OCT image retina structure layering algorithm.The contrast is adjusted with reference to the local imaging quality of the OCT image,the results of the stratification is optimized.In this thesis,it is proved by experiments that the OCT image retinal structure layering algorithm with quality parameter control has better performance when the imaging quality of OCT image is poor.Our algorithm not only can achieve high consistency with the doctor’s subjective evaluation but can be applied in other OCT image research fields by introducing quality parameter control of OCT images,avoiding the disadvantages caused by poor image quality and improving robustness.