Abstract
Cervical cancer is the most common type of cancer worldwide, affecting thousands of women each year. The Pap test is a screening test used for the diagnosis of cervical cancer. In this test cytologist or an expert examines the cervical cells under a microscope. However, fatigue and interobserver variability hinder this manual process. Also, in less developed countries with limited resources, access to an expert for manual analysis is not always guaranteed. This makes the development of an automated system very beneficial.
Often, one of the critical steps in the automated analysis of Pap smear images is the accurate detection of cell nuclei. Many researchers have already proposed algorithms to automatically detect nuclei in cervical cytology images. These proposed methods are based on Machine learning approaches, and the state-of-the-art methods could achieve relatively good results. Nonetheless, developing even more accurate techniques is essential and is being researched.
We propose a deep learning-based automatic method for the detection and segmentation of nuclei in cervical cytology images. The method is composed of convolutional neural networks based on U-Net and EfficientNet architectures. The network combines the EfficientNetB4 as the encoder for feature extraction with the U-Net as decoder for segmentation. The proposed method produces reasonable results for nuclei segmentation and detection.