Related Experiment Video
Updated: Oct 17, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Segmentation of Overlapping Cervical Cells with Mask Region Convolutional Neural Network
1Zhongshan Hospital Xiamen University, Xiamen, Fujian 361004, China.
Computational and Mathematical Methods in Medicine
|October 14, 2021
Summary
This study introduces a two-stage Mask RCNN framework for accurate cytoplasm segmentation in cervical cytology images, effectively handling overlapping cells. The deep learning approach significantly improves segmentation performance in cytological analysis.
Area of Science:
- Medical Imaging
- Computational Biology
- Artificial Intelligence
Background:
- Cytoplasm segmentation in cervical cytology is challenging due to fuzzy and overlapping cells.
- Deep learning methods show promise for complex medical image segmentation.
Purpose of the Study:
- To develop an automated method for segmenting overlapping cells in cytology images.
- To improve the accuracy of cytoplasm segmentation in cytological analysis.
Main Methods:
- A two-stage framework utilizing Mask RCNN was developed.
- Stage one proposed candidate cytoplasm bounding boxes.
- Stage two refined boundaries using pixel-to-pixel alignment and category classification.
Main Results:
- The method was evaluated on ISBI 2014 and 2015 datasets.
- Achieved a Dice Similarity Coefficient (DSC) of 0.92 and False Positive Rate per polygon (FPRp) of 0.0008 at a DSC threshold of 0.8.
- Outperformed existing state-of-the-art approaches.
Conclusions:
- The Mask RCNN-based segmentation method is effective for cytological analysis.
- The proposed framework successfully addresses the challenge of overlapping cells.
- This technology can enhance automated diagnostic capabilities in cervical cancer screening.
