Cell proliferation detection based on deep learning

Cell proliferation level is important in clinical diagnosis. Analysis of cell proliferation level can help to judge the development trend of patient's condition, which is helpful for diagnosis. In order to evaluate the level of cell proliferation, it is necessary to calculate the number and proportion of proliferating nuclei. However, when the number of nuclei is large, it will bring a great pressure on doctors, and the accuracy rate will also decline. To solve these problems, this paper proposes a scheme for automatic detection and counting of cell nuclei, which is large in number and densely distributed. Our dataset is obtained from mouse liver cells, a total of 136 samples, each containing 100-300 nuclei and the proliferation of nuclei are stained by immunohistochemical staining, of which 120 are trained using convolution neural network and 16 samples are tested to evaluate the effects of the models. Three models were trained by RetinaNet with backbone networks of Vgg6, ResNet50 and ResNet101 respectively, and compared with Image Pro Plus 6.0. Experiments show that our models achieve higher accuracy compared with 67.2% obtained by manual using Image-Pro Plus 6.0 for detection and counting. It can be seen that in terms of detection accuracy, our models are better than the analysis software widely used in hospitals, effectively solving the problems of long manual detection time and low accuracy. It can effectively help doctors to evaluate the level of cell proliferation, and then quickly make a corresponding diagnosis.