Segmentation of Overlapping Cervical Cells with Mask Region Convolutional Neural Network.

Jiajia Chen,Baocan Zhang,Tao Huang

doi:10.1155/2021/3890988

Abstract

The task of segmenting cytoplasm in cytology images is one of the most challenging tasks in cervix cytological analysis due to the presence of fuzzy and highly overlapping cells. Deep learning-based diagnostic technology has proven to be effective in segmenting complex medical images. We present a two-stage framework based on Mask RCNN to automatically segment overlapping cells. In stage one, candidate cytoplasm bounding boxes are proposed. In stage two, pixel-to-pixel alignment is used to refine the boundary and category classification is also presented. The performance of the proposed method is evaluated on publicly available datasets from ISBI 2014 and 2015. The experimental results demonstrate that our method outperforms other state-of-the-art approaches with DSC 0.92 and FPRp 0.0008 at the DSC threshold of 0.8. Those results indicate that our Mask RCNN-based segmentation method could be effective in cytological analysis.

Highlights

Cervical cancer is one of the most common types of cancer among women, which causes tens of thousands of deaths every year
The main contributions of this work include the following: (i) A new method using Mask RCNN is proposed for segmentation of overlapping cervical cells, where small amount of annotated images is needed and any a priori knowledge about cells is not required (ii) Our proposed method achieves superior results compared to other state-of-the-art methods in some terms of measures
The segmentation performance is evaluated by four evaluation metrics (i.e., dice similarity coefficient (DSC), FNRo, TPRp, and FNRp), which are the original metrics in the two International Symposium on Biomedical Imaging (ISBI) competitions

Summary

Introduction

Cervical cancer is one of the most common types of cancer among women, which causes tens of thousands of deaths every year. The diagnostic procedure requires a large amount of time and is tedious These issues have motivated the development of automated diagnostic techniques. Those techniques are largely based on the cell images acquired by a digital camera connected to a microscope. Many approaches have been proposed for complete segmentation of overlapping cervical cells [4,5,6]. Two high-quality datasets containing the original cell images and their annotations were made publicly available Those two datasets make the evaluation and comparison of different segmenting methods possible. Phoulady et al [6] proposed a framework to detect nuclei and cytoplasm in cervical cytology extended depth of field (EDF) images.

Evaluation

Method

Datasets and Experimental Design

Results and Comparison

Discussion

Conclusion