Cell Lineage Prediction Based On Deep Learning

At present,the mechanism of cell reprogramming is deeply studied by exploring different cell lineages.The cell lineages studied in this paper are small molecules inducing fibroblasts to differentiate into multifunctional stem cells.In the process of culture,the cell morphology and displacement of the cells vary greatly.It is difficult to predict the success of the cell differentiation by identifying the single cell and predicting the success of the cell differentiation.A new method of cell lineage prediction is proposed in this paper.The whole process of cell differentiation is recorded by real time imaging,and the morphological differences of different cells are learned by deep learning method,so as to predict the success of cell differentiation.This method is more accurate than traditional manual discrimination.The main work of this paper is divided into three aspects as follows:(1)identification and cutting of cell lineage images.In this paper,the optimized edge detection algorithm is used to identify the single cell object in the image,and the single cell is automatically cut into a cell image block by using the identified edge coordinates.(2)anomaly detection and differential feature extraction of cell lineage images.In this paper,in the task of cell image classification,the original data is detected first,and the method of data processing and data enhancement is used according to the result of abnormal detection.At the same time,this paper uses the method of deconvolution network visualization and feature screening to find out 5 differences between the cells of different populations,which are DNA intensity,texture entropy,texture correlation,DNA quality and image contrast.This is helpful for biological scientists to better distinguish different groups of cells in cell culture experiments.(3)prediction of cell lineage based on deep learning.In this paper,Batch Normalization layer is added to AlexNet convolution neural network to form AlexNet-BN.Compared with the original AlexNet network,it reduces the effect of the absolute difference between the pixels of the image and strengthens the relative difference between the features of the pixels,so it is more suitable for the classification task of the cell lineage in this paper.At the same time,the number of negative cells is small,which is not enough to support convolution neural network training.This paper transplants the deep network from the ImageNet database and further optimizes it on this basis to make it adaptable to the difference between the source domain and the target domain.Experimental results show that this method achieves better classification results on a small number of cell image training sets.At the same time,it also has better classification performance in public eye diseases dataset and pneumonia X optical image dataset.