Research On Deep Learning Nuclei Detection Algorithm Of Histopathology Image

The detection and classification of nuclei in histopathology images is an important step in many pathology diagnoses,and the results provide an important basis for subsequent diagnosis.With the advancement of digital medical technology and the abundance of pathology image data,the acquisition of histopathology images has become easier and more popular.The efficiency and accuracy of manual detection is no longer sufficient to cope with the demand for such large-scale detection,and therefore relevant image processing algorithms are necessary to detect and classify cell nuclei in histopathology images.In this paper,a deep learning algorithm for cell nuclei detection in histopathology images is carried out to address the problems arising in the process of cell nuclei detection and the shortcomings of traditional algorithms,and the specific research work is as follows:(1)To address the problems of poor localization accuracy and low efficiency of current automated cell nuclei detection methods,a lightweight deep learning end-toend cell nuclei detection method Cell-Det is proposed.this method uses fully convolutional networks to learn the features of histopathology images and combines feature fusion methods to obtain the results of cell nuclei detection,which effectively solves the problems of localization accuracy and detection efficiency and avoids the problems of training as well as the shortcomings of previous models based on sliding windows.The method effectively solves the problems of localization accuracy and detection efficiency,and avoids the shortcomings of previous sliding window-based models in the training as well as detection process.The experimental results show that the model has significantly improved the evaluation metrics over some of the current optimal models on the public dataset,with the F1 score improving to 0.851 and the accuracy of the model significantly improving to 0.842 on the metric of detecting unbalanced data.(2)To address the problems of low accuracy and inaccurate localization of existing models in fine-grained cell nuclei classification tasks,a multi-scale fully convolutional network model CFCN based on dilated convolution and feature fusion is proposed,which uses a backbone network designed in cross stage partial to efficiently extract cell nuclei features,and then uses a cascaded dilated convolution module to enhance the perceptual field of the model and extract multi-scale features.The feature fusion method using path aggregation networks also works well to localize the classification results of cell nuclei.Experiments show that the proposed model is able to classify nuclei accurately in two different staining environments,with an F1 score of 0.750 and a significant increase in accuracy compared to the best-performing SFCNOPI,indicating that the model can effectively solve the problems caused by the imbalance between positive and negative samples.