Semi-supervised Cell Classification Based on Deep Learning

Abstract

Pathological examination is an important diagnostic means for cancer, including clinical cytological examination and histopathological examination. In pathological examination, it is often necessary to judge the type of cells. According to identifying the cells type or the number of different cell, doctors can determine whether to have cancer or the stage of cancer. However, pathological images contain a large number of different types of cells, which often need to be labeled with the professional knowledge of pathologists. In order to reduce the burden of pathologists, there are more and more methods to use computer aided cell classification. In recent years, with the rise of deep learning, it has become common to apply it to the segmentation and classification of pathological images. And the semi-supervised learning method can make good use of the image information of a large number of unlabeled samples to improve the performance of the model. But the existing methods of semi-supervised cell classification are not simple enough, and the sample selection mechanism cannot make full use of the characteristics of semi-supervised learning. Therefore, we propose a semi-supervised cell classification framework based on reliable sample selection mechanism, which can flexibly train different classifiers according to different data sets. The framework makes full use of semi-supervised learning, which makes the classification accuracy of model improved steadily.