Classification And Recognition Of Abnormal Cells Based On DNNA Ploidy Analysis And Deep Learning

Cancer,as a malignant disease that is the most serious damage to the health of the body,has always been concerned by researchers from all walks of life.How to actively and effectively prevent and treat cancer has become a common topic in the medical field.Also for female groups,cervical cancer is the second leading killer of women 's disease mortality.After years of research,it has been confirmed that screening cervical cells and discovering the presence of abnormal cancerous cells can be effective through early prevention and treatment.Reduce the mortality of the disease.In recent years,with the great influence of SVM algorithm in image classification in machine learning,the DNA ploidy analysis system based on this algorithm and DNA content detection has played an advantage in the detection of cancerous abnormal cells.Compared with the traditional manual reading,the efficiency of this system is obviously improved,but because the classification cells produced by this system often lead to inaccurate results and fewer categories,manual screening is also required.Therefore,this topic selects the currently effective deep learning algorithm,combines the output of the DNA ploidy analysis system,and combines traditional machine learning with classic neural network methods to become a new method of computer-aided diagnosis and treatment of diseases.The research content of this article is to classify the abnormal cells in the cervical image,because the garbage cells identified in the DNA ploidy analysis system usually contain aggregated cells,adhesion cells,neutrophils,and real garbage cells and impurities.Cell class,so this paper uses deep learning methods to retrain cell classification models.The experiment in this paper is a neural network built on the Keras framework.The convolutional neural network is used to learn six types of cell characteristics to obtain a cell classification model.The accuracy of the experimental model corresponding to different networks is calculated.Finally,the corresponding recognition results are displayed through the interface program.In the preliminary experiment of this subject,the Faster R-CNN method was used for cell classification.This method is based on the model training proposed by the region.It uses the method of automatically detecting the cell region to obtain the predicted cell coordinates and categories.The model parameters obtained through training predict the cell sample set with an average accuracy of 0.563,which has a better effect on cell identification during image prediction,but the accuracy of the results is lower than the identification method used in Chapter 4.In the later period,the cell position information generated by the DNA ploidy analysis system and the CNN fusion method are mainly used to identify and classify the cervical shed cells.The experiment first used the ZFNet network to find a cell data set with an appropriate image size.The results showed that the recognition accuracy of 64-bit images was higher than that of 32-bit images,reaching 95.21%,but there was a phenomenon that the two cell recognition results could not be clearly distinguished;and then using OpenCV technology to process the data set,and then through two different deep networks of VGG16 and ResNet50 to continue to improve the accuracy of the model,the final prediction accuracy of the cell model trained with the VGG16 network is as high as 98.74% in the final result,which is a good result.The method mentioned in this article is feasible for identifying cell types.The experimental results show that,compared with the Faster R-CNN method in Chapter 3,the method based on DNA ploidy analysis and CNN fusion has better cell classification performance,and this method can be used as a DNA ploidy analysis system.The auxiliary identification method provides a reliable basis for the subsequent accurate diagnosis.