Detection, segmentation and classification for cervical cytology image
Changchang Liu, Junwei Deng, Yiqing Shen, Jing Ke · 2019
We design a new framework containing of two neural networks, as one for cell detection and segmentation by pixel-wise annotation and the other for subtype classification by image-level labeling. This model significantly saves the manual annotation effort while preserving the same accuracy in cell boundary location, nuclei location, and subtype classification. Moreover, as the classification result is crucial to computer-aided diagnosis, and the accuracy of neural networks can be reliably achieved at the expense of more training data, we require only image-level labeling to improve this system in the future work.