A Novel M^2CNN Method for Cervical Cancer Cell Recognition

Zengyong Xu ； Meili Rao

《淡江理工學刊》 24卷2期 (2021/04) Pp. 215-221

https://doi.org/10.6180/jase.202104_24(2).0010

摘要

In the cervical cancer cell image recognition, there are some problems such as insufficient labeled data, single scale feature and poor robustness, the recognition rate is low. Therefore, we propose a multi-scale multi-feature convolutional neural network (M^2CNN) method for cervical cancer cell recognition. Firstly, we conduct data enhancement due to the limited cervical cancer data set to avoid over-fitting. The image is rotated to three differ-ent scales using the Laplacian pyramid method. In order to enhance the feature robustness, the above three scales are used as the inputs of the model. The model also learns the features of the three scales. The weighted sum of multi-scale information in different layers of the network makes the features of different layers have dif-ferent effects on the final result to improve the robustness of the model. Experimental results show that this method can effectively complete the automatic recognition of cervical cancer cells. In terms of the accuracy and recall rate, our proposed method greatly improves the results. In this paper, deep learning technology is intro-duced into the field of cervical cell-assisted screening, which is of great significance for promoting the research of early automatic screening for cervical cancer.

關鍵字

Cervical cancer cell recognition ； M^2CNN ； Laplacian pyramid method ； data enhancement