The Recognition And Localization Of Induced Pluripotent Stem Cells Based On Deep Learning

In recent years,some diseases caused by irreversible damage of functional cells in the body,such as Parkinson's syndrome,Alzheimer's disease,diabetes,etc.,have gradually become a major problem that endangers people's health.Currently,the most effective treatment for the above diseases is the third generation of treatment based on stem cells and regenerative medicine.And induced pluripotent stem cells cultured artificially provide a solution for the sources of stem cell.However,the artificial culture process of induced pluripotent stem cells is very complicated,and due to the uncertainty of manual operation,the cell quality of different batches is greatly different and there are potential pollution factors.At the same time,the manual operation method is inefficient,the cultivation cost is high,and the cycle is long,which is difficult to meet the large-scale clinical and scientific research needs.Therefore,there is an urgent need for a fully automated stem cell induction culture device.To achieve automatic induction and culture of stem cells,automated identification and localization of cell images is an essential step.The focus of this thesis is to identify and locate the location of cell clusters in the picture based on the deep learning method,so as to obtain the position coordinates of the cell cluster,and then provide position information for the automated picking of the cell cluster.The specific research contents of this thesis are as follows:(1)A method of cell identification based on Faster RCNN is proposed and compared with SSD and YOLO algorithms.At the same time,in order to improve the recognition accuracy and positioning accuracy of Faster RCNN,It is also proposed to add a Alexnet network to the second classification of the cells after the result of the Faster RCNN recognition,which improves the accuracy of cell recognition,and the accuracy of cell recognition reaches 99%,and creates favorable conditions for subsequent cell splicing and fusion.(2)A panoramic image stitching method based on Fourier transform is proposed.Compared with the traditional SIFT image stitching method,both speed and accuracy are improved.Firstly,the Fourier transform is used to solve the translation transformation relationship between adjacent pictures.Then,the coordinate transformation formula between the panoramic images is used to solve the panoramic image transformation.Finally,the panoramic image can be spliced,and the accuracy of image stitching reaches 100%.(3)A method of cell region fusion based on SVM is proposed.A classifier is trained based on the coincidence degree and the center distance as a basic feature to determine whether adjacent cell clusters belong to the same cell area,and the accuracy of the classification is 90%.Thereby,different cell regions are fused,and finally the coordinate positions of all the cell regions relative to the entire cell culture dish region are obtained.The three algorithms are combined to detect the cell area,and the detection accuracy is 98.8%.