Related Experiment Video
Updated: Dec 7, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
3.0K
Semantic segmentation of HeLa cells: An objective comparison between one traditional algorithm and four deep-learning
Cefa Karabağ1, Martin L Jones2, Christopher J Peddie2
1Research Centre for Biomedical Engineering School of Mathematics, Computer Science and Engineering, Department of Electrical & Electronic Engineering, City, University of London, London, United Kingdom.
Plos One
|October 2, 2020
Summary
A traditional image-processing algorithm achieved superior accuracy (99%) and Jaccard index (93%) for segmenting HeLa cell nuclear envelopes compared to four deep learning models, highlighting its effectiveness in cell morphology studies.
Area of Science:
- Biomedical Imaging
- Cell Biology
- Computational Pathology
Background:
- Accurate cell morphology quantification is crucial for understanding health and disease states.
- Cell structure segmentation is a foundational step in quantitative morphology.
- HeLa cells, a common cervical cancer cell line, are frequently studied for cellular changes.
Purpose of the Study:
- To compare the performance of a traditional image-processing algorithm against four deep learning architectures for semantic segmentation of the nuclear envelope in HeLa cells.
- To evaluate segmentation accuracy and Jaccard index for each method against a ground truth.
Main Methods:
- Five algorithms were evaluated: one traditional image-processing pipeline and four deep learning models (VGG16, ResNet18, Inception-ResNet-v2, U-Net).
- Serial Block Face Scanning Electron Microscopy images of HeLa cells (300 slices, 2000x2000 pixels) were used.
- Deep learning models were either pre-trained with ImageNet and fine-tuned or trained from scratch.
Main Results:
- The traditional algorithm achieved the highest segmentation accuracy (99%) and Jaccard index (93%).
- Deep learning models showed varying performance: VGG16 (93%, 90%), ResNet18 (94%, 88%), Inception-ResNet-v2 (94%, 89%), and U-Net (92%, 56%).
- The traditional method outperformed all tested deep learning architectures in this specific segmentation task.
Conclusions:
- Traditional image-processing techniques can be highly effective and even superior to deep learning for specific cell segmentation tasks, such as nuclear envelope segmentation in HeLa cells.
- Careful algorithm selection and evaluation are essential for accurate quantitative cell morphology studies.
- Further research may explore hybrid approaches or optimized deep learning architectures for improved performance.
