Related Experiment Video
Updated: Nov 8, 2025
Author Spotlight: Enhancing PSC-to-Functional Cell Differentiation Using ML Models Based on Live-Cell Bright-Field Imaging
Published on: October 4, 2024
[Banded chromosome images recognition based on dense convolutional network with segmental recalibration]
Jianming Li1, Bin Chen2, Xiaofei Sun1
1Chengdu Institute of Computer Applications, Chinese Academy of Sciences, Chengdu 610041, P.R.China;University of Chinese Academy of Sciences, Beijing 100049, P.R.China;Guangzhou Electronic Technology Co. Ltd., Chinese Academy of Sciences, Guangzhou 510070, P.R.China.
This study introduces a new model, SR-DenseNet, for accurate human chromosome recognition, crucial for diagnosing genetic diseases. Model fusion significantly improves accuracy, paving the way for automated karyotyping.
Area of Science:
- Genetics
- Computer Science
- Medical Imaging
Background:
- Human chromosome karyotyping is essential for diagnosing genetic disorders.
- Accurate chromosome image recognition is vital for automating karyotyping.
- Existing methods require improvement in accuracy and efficiency.
Purpose of the Study:
- To propose an effective model for automatic chromosome type recognition.
- To enhance the accuracy of chromosome image identification.
- To explore the potential of deep learning for automated karyotyping.
Main Methods:
- Development of a Segmentally Recalibrated Dense Convolutional Network (SR-DenseNet).
- Utilizing dense connected layers for multi-level feature extraction.
- Incorporating Squeeze-and-Excitation (SE) blocks for feature recalibration.
- Implementing a model fusion method with an expert group of models.
Main Results:
- SR-DenseNet achieved a 1.60% error rate on the Copenhagen dataset (G-bands).
- Model fusion reduced the error rate to 0.99% on the Copenhagen dataset.
- On the Padova dataset (Q-bands), the model achieved a 6.67% error rate, reduced to 5.98% with fusion.
Conclusions:
- The proposed SR-DenseNet model demonstrates high effectiveness in chromosome type recognition.
- Model fusion further enhances recognition accuracy, showing significant potential.
- The method contributes to the automation of chromosome type recognition for genetic disease diagnosis.
