MT-SCnet: A New Network for Microscopic Hyperspectral Image Segmentation
Author Information
Author(s): Cao Xueying, Gao Hongmin, Zhang Haoyan, Fei Shuyu, Xu Peipei, Wang Zhijian
Primary Institution: College of Computer Science and Software Engineering, Hohai University, Nanjing, China
Hypothesis
Can a multi-scale token divided and spatial-channel fusion transformer network improve segmentation of microscopic hyperspectral images?
Conclusion
The MT-SCnet outperforms existing methods in segmenting microscopic hyperspectral images.
Supporting Evidence
- MT-SCnet achieved 94.45% overall accuracy on the IM dataset.
- MT-SCnet outperformed U-Net by 1.26% in overall accuracy.
- MT-SCnet showed improvements in sensitivity and IoU compared to existing methods.
- MT-SCnet has 34.59M parameters and 120.91G FLOPs, balancing accuracy and computational cost.
Takeaway
This study created a new method to help computers better understand images of tiny tissues, which can help doctors find diseases earlier.
Methodology
The study used a multi-scale token divided and spatial-channel fusion transformer network for segmentation tasks on two datasets.
Limitations
The study may have limitations in sensitivity compared to some existing methods.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website