Transfer Learning Based Classification of Cervical Cancer Immunohistochemistry Images

Cervical cancer is the fourth leading cause of cancer-related deaths. It is very important to make the precise diagnosis for the early stage of cervical cancer. In recent years, transfer Learning makes a great breakthrough in the field of machine learning, and the use of transfer learning technology in cervical histopathology image classification becomes a new research domain. In this paper, we propose a transfer learning framework of Inception-V3 network to classify well, moderately and poorly differentiated cervical histopathology images, which are stained using immunohistochemistry methods. In this framework, an Inception-V3 based transfer learning structure is first built up. Then, a fine-tuning approach is applied to extract effective deep learning features from the structure. Finally, the extracted features are designed for the final classification. In the experiment, a practical images stained by AQP, HIF and VEGF approaches are applied to test the proposed transfer learning network, and an average accuracy of 77.3% is finally achieved.