Research On Detection And Classification Of Histological Cell Image Based On Deep Learning

As an important research direction in the field of medical images,histopathological cell image detection and classification has been widely used in computer-aided diagnosis,biological research and other fields.With the rise of deep learning,neural network is applied to medical image analysis,which can realize the automatic detection and classification of histological cell images.Most of the existing medical cell image detection methods ignore the influence of other factors in the topological domain,such as space,which leads to the inevitable defects in the accuracy and generalization of the algorithm.At the same time,the existing medical cell image classification methods are too simple in the prediction classification label,which limits the accuracy of cell image classification.With the development of deep learning,the more diversified network model architecture provides a theoretical basis for cell image detection and classification.This paper makes an in-depth study on the detection and classification of histological cell images,the main contents of which are as follows:(1)Study on the method of nuclear detection combined with spatial information.In order to solve the problem that the output of the existing neural network is affected by other factors in its topological domain,on the basis of the traditional convolution neural network,combined with the spatial position information,an improved convolution neural network model for histological cell image detection is proposed.Taking the traditional convolution neural network as the carrier,the convolution neural network model based on spatial information is constructed,which makes the model have the ability to fuse spatial information and eigenvector.Histopathological cell images were preprocessed by color deconvolution.Finally,through the experiment to verify the effectiveness of the model for nuclear detection of histological cell images.(2)Study on the method of nuclear classification combined with the prediction mechanism of adjacent sets.In order to solve the problem of single prediction of classification labels by existing neural networks,an adjacent set prediction method is proposed,which weighs all the relevant predictors.Based on Softmax CNN,a histological cell image classification model combined with adjacent set prediction is constructed to predict the classification label of the detected nucleus and further classify the nucleus.Finally,through the experiment to verify the effectiveness of the model for nuclear classification of histological cell images.(3)Comparative experimental study and analysis of nuclear detection model and classification model.Research and analysis of the existing mainstream nuclear detection and classification methods,combined with the model detection results in Chapter 2,based on the accuracy,recall rate and F1 score index,set up nuclear detection comparative experiments;combined with the classification results of the model in Chapter 3,based on the AUC value and F1 score index,set up nuclear classification comparative experiments.Through the comparison and analysis with the comparative literature experimental results,quantitatively evaluate the performance of the nuclear detection model and classification model proposed in this paper.In this paper,the innovations in the study of histological cell image detection and classification are as follows: based on the study of convolution neural network for nuclear detection of histological cell image,a convolution neural network method based on spatial information is proposed,it is used to detect the nucleus of histological cell image.Based on the study of nuclear classification of histological cell images by Softmax CNN,an adjacent set prediction method is proposed,which is connected with convolution neural network to predict the classification labels of detected nuclei and realize the classification of nuclei.