Using Multispectral Imaging for Classifying Nuclei in Histopathology
Author Information
Author(s): Boucheron Laura E, Bi Zhiqiang, Harvey Neal R, Manjunath BS, Rimm David L
Primary Institution: University of California, Santa Barbara
Hypothesis
The additional spectral bands in multispectral imagery contain more information useful for classification compared to the 3 standard bands of RGB imagery.
Conclusion
Multispectral imagery provides minimal additional spectral information for nuclear classification compared to standard RGB imagery.
Supporting Evidence
- The best performance was achieved with either multispectral or RGB imagery, with only a 0.79% increase in performance.
- The single best multispectral band provided a 0.57% performance increase over the best RGB band.
- Principal components analysis indicated only two significant image bands were informative for the classification task.
Takeaway
This study looked at how well different types of images can help identify cells in tissue samples, and found that using more colors didn't really help much more than just using three colors.
Methodology
The study used multispectral imaging with 29 spectral bands and compared it to standard RGB imagery for pixel-level classification of nuclei in histopathology images.
Limitations
The study only focused on a single pixel-level classification task and may not generalize to other tasks.
Participant Demographics
The dataset included 58 H&E stained histopathology images of breast tissue, with 26 malignant and 32 benign images.
Statistical Information
P-Value
0.0000
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website