Research On Segmentation Of Artificial Skin Equivalents In OCT Images Based On Convolutional Neural Networks

This article mainly focuses on the segmentation technology of OCT images of Artificial Skin Equivalents.In recent years,Artificial Skin Equivalents has been widely used in the field of cosmetic toxicity evaluation.In order to ensure the quality and consistency of the Artificial Skin Equivalents used for testing,it is necessary to develop appropriate means for real-time online monitoring during the production and preparation of Artificial Skin Equivalents.Optical Coherence Tomography(OCT)technology has the advantages of non-invasive,non-destructive,online detection,etc.and is very suitable for non-destructive evaluation of tissue engineering products including Artificial Skin Equivalents.Although many researchers have applied OCT technology to the evaluation of Artificial Skin Equivalents,there are still some deficiencies in current research.Firstly,OCT images of Artificial Skin Equivalents have the characteristics of strong noise,blurred edges,and large amount of data.However,in existing research literature,the image processing methods used are still limited to simple threshold segmentation,edge detection,peak detection,etc.It does not introduce deep learning technology that has outstanding performance in other biomedical image processing fields,and lacks accurate image segmentation processing.Secondly,the existing research mainly deals with the processing of 2D OCT images of Artificial Skin Equivalents,and only the2 D section information of Artificial Skin Equivalents can be observed.Using the 3D OCT data of the Artificial Skin Equivalents,the overall condition of the Artificial Skin Equivalents can be more accurately assessed.In response to the above problems,this paper conducts research on OCT image segmentation of Artificial Skin Equivalents.The main work is as follows:(1)According to the correlation between the quality index and structural features of Artificial Skin Equivalents,it is determined that the target of Artificial Skin Equivalents OCT image segmentation is the whole area of Artificial Skin Equivalents and the horny layer.On this basis,we collected and constructed a corresponding dataset for neural network training and verification.(2)Based on the improvement of the classic U-Net convolutional neural network,a convolutional neural network for Artificial Skin Equivalents OCT image segmentation is constructed.On the one hand,this article adds a batch normalization layer to U-Net,which can speed up the network’s convergence speed during the training process.On the other hand,this article also constructs neural networks of different sizes and dimensions to ensure that a smaller neural network is selected when there is no significant difference in segmentation effects to reduce the amount of calculation and speed up the processing.In order to avoid inaccurate verification results caused by random errors when the data set is divided into training set and test set,this paper uses six-fold cross-validation to train and verify the neural network,which improves the reliability of the verification results.(3)This paper studies the method of using neural network to process 3D OCT data of Artificial Skin Equivalents.By cutting the 3D OCT data of the Artificial Skin Equivalents into a series of 3 consecutive 2D-BScans,and then stitching them into3 D data,the 3D OCT data of the Artificial Skin Equivalents is segmented.This method combines the contextual information of the 3D OCT data,which can effectively avoid the problems of large amount of calculation and slow processing speed caused by the direct use of neural networks to segment 3D data.In addition,this paper uses the orthogonal prediction method,predicting twice along the X and Y directions and averaging the results.The test results show that using this method can significantly improve the image detail quality after segmentation.(4)To cope with the demand for large-scale processing of Artificial Skin Equivalents OCT data,this paper optimizes the data processing flow of Artificial Skin Equivalents OCT image based on the analysis of the hardware occupancy and computing time in different stages of data processing,and greatly improves the processing speed of Artificial Skin Equivalents OCT image.