Author(s)

Kai Zhang, Jun Zhang, Xin Cai, Yalan Tang, Dan Lin, Mingxing Chen, Lan Li, Yajun Chen, Fan Xu

Corresponding Author

Yajun Chen

ABSTRACT

The detection of rare pathological types plays an important role in the clinical analysis of cervical cancer to ensure efficient treatment and improve recovery rate. Recently, with the rapid development of medical equipment, the diagnoses of pathological types become more accurate. Traditional approach is based on the experience of medical personnel or relevant physicians, some rare types are more likely to be neglected. In this study, we are concerned with the problem of develop a model based on deep learning to assist physicians in detecting pathological types of cervical cancer; Another important point, insufficient training data has always been a common limitation in the field of medical imaging. To solve this challenge, we propose a Resnet-based pre-training model to extract features of pathological images and classification. In particular, the introduction of migration learning not only reduces the scale of training data, but also effectively avoids over-fitting of deep models. We validate our model on a T2 sagittal image dataset of 641 patients with cervical cancer and compare it with the original algorithm. The experimental results show that the proposed model achieves effective performance in terms of cervical cancer pathologic detection.

KEYWORDS

Artificial intelligence, Deep learning, Transfer learning, Computer-aided