Related Experiment Video
Updated: May 20, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
3D Swin Transformer With Multi-Scale Dilated Convolution for White Blood Cell Hyperspectral Image Classification
Yushi Yang1,2, Danfei Huang1,2, Yi Xie1,2
1College of Optoelectronic Engineering, Changchun University of Science and Technology, Changchun, China.
Journal of Biophotonics
|May 19, 2026
Summary
This study introduces a novel deep learning method for classifying white blood cells (WBCs) using hyperspectral imaging. The SwinMDC network significantly improves accuracy by utilizing spectral information for enhanced clinical diagnosis.
Area of Science:
- Medical Imaging
- Computational Biology
- Artificial Intelligence
Background:
- Accurate white blood cell (WBC) classification is vital for clinical diagnosis.
- Current methods often neglect spectral information, focusing mainly on spatial structures.
- Hyperspectral imaging offers rich spectral data for enhanced analysis.
Purpose of the Study:
- To develop an advanced deep learning model for WBC hyperspectral image classification.
- To leverage spectral information for improved diagnostic accuracy.
- To introduce the 3D Swin Transformer network (SwinMDC) for this task.
Main Methods:
- Application of hyperspectral imaging technology for WBC analysis.
- Proposal of the 3D Swin Transformer network (SwinMDC) incorporating multi-scale dilated convolutions.
- Utilizing a 3D multi-scale dilated convolution feature extractor for enhanced low-level representations.
- Integration of a 3D window-based attention mechanism for capturing long-range dependencies.
Main Results:
- The SwinMDC network achieved a 99.07% overall classification accuracy on a three-class WBC hyperspectral dataset.
- Demonstrated superior performance compared to methods focusing solely on spatial structures.
- Highlighted the effectiveness of incorporating spectral information through hyperspectral imaging.
Conclusions:
- The SwinMDC network shows significant potential for clinical WBC analysis.
- Hyperspectral imaging combined with advanced deep learning offers a powerful approach for hematological diagnostics.
- This method enhances the utilization of spectral information for more accurate cell classification.
