Age Prediction From Optical Coherence Tomographic Imagesbased On Deep Learning Model

Optical coherence tomography(OCT)is a non-invasive,highresolution imaging technique of retina tissue based on low coherence light interference.It can clearly distinguish the structures of various layers of the retina.When people get older,the retina tissue changes,and the corresponding OCT images represent these changes in various ways.The prediction of retina age based on OCT images plays an important role in detecting the aging degree of the eye,evaluating human health status,and managing age-related chronic diseases.Therefore,this work uses deep learning methods to predict the retina age from OCT images,in order to find early prevention and treatment methods regarding various potential eye diseases.The main work of this project is as follows:(1)Obtain the experimental dataset and preprocess it.Firstly,we select suitable OCT image data,label it with age,organize and summarize it to obtain a dataset for predicting the age of OCT images.Then,the OCT image is preprocessed,and the image processing method of histogram equalization is used to improve the contrast of the OCT image.Finally,a 3D block matching algorithm is used to denoise the image,reducing the noise in the image background.By extracting regions of interest,irrelevant information in the image background is reduced,and the speed of carrying out the deep learning model is improved.(2)Propose a deep learning model based on an improved ResNet network and use it for OCT image age prediction.Firstly,a deep residual neural network ResNet model is used to conduct age group binary classification experiments based on OCT images to determine the feasibility of age prediction in OCT images.Subsequently,in order to further achieve accurate prediction of retinal age,ResNet is optimized based on the hierarchical structural characteristics of ophthalmic OCT images.An improved model based on ResNet network is proposed,and a Convolutional Block Attention Module(CBAM)is added to achieve the allocation of attention weights for different layers of OCT images.Finally,in order to verify the performance of ResNet+CBAM network model,the average absolute error and determination coefficient are used as evaluation indicators,and OCT image age prediction experiments are conducted regarding the persons with different ages.The experimental results show that the proposed ResNet+CBAM network model achieves the lowest average absolute error compared with ResNet,VGG and DenseNet.At the same time,by visualizing the attention mechanism,the attention heatmaps are obtained,improving the interpretability of deep learning networks and allowing the visual observation of the impact of different regions for the age prediction.(3)Explore the impact of different retina structures for age prediction.Based on the OCT image age prediction experiment,we propose an ablation experimental scheme based on specific regions of the OCT image for age prediction.It preliminarily determines the amount of age-related information contained in different structures of the fundus and quantifies it,providing certain reference value for studying and judging the aging degree of different retina tissues.