Cervical Cell Detection Benchmark with Effective Feature Representation

As deep convolutional neural networks have shown promising performance in medical image analysis, a number of deep learning based cervical cytology diagnosis methods were developed in recent years. Most studies have achieved available performance in cell classification or segmentation, however, there still exists some challenges for effective screening. Cervical cell detection is a more significant task in cytology diagnosis for cancers. In this paper, we propose a detection framework with effective feature representation for automatic cervical cytology analysis. We employ elastic transformation and a channel and spacial attention module to obtain a more powerful feature extractor. The experimental results demonstrate the efficiency and accuracy improved by our effective feature representation.