Denoising multiplexed microscopy images in n-dimensional spectral space.

Area of Science:

• Microscopy
• Image Analysis
• Spectroscopy

Background:

• Hyperspectral fluorescence microscopy generates complex images with sparse spectral features.
• Noise and spatial variations in these images challenge accurate spectral analysis.
• Existing methods often struggle to denoise effectively without losing spatial information.

Purpose of the Study:

• To introduce a novel spectral vector denoising algorithm for hyperspectral microscopy images.
• To enhance the accuracy of spectral decomposition analysis by reducing noise.
• To preserve spatial information while improving spectral data quality.

Main Methods:

• The algorithm utilizes n-dimensional Chebyshev or Fourier transforms to cluster pixels by spectral similarity.
• A denoising convolution filter is applied in the identified spectral space.
• The method operates independently of pixel intensity or location.

Main Results:

• Denoising in 3D to 5D spectral spaces significantly reduces unmixing error by up to 70%.
• The algorithm effectively filters noise without sacrificing spatial resolution.
• Tests on simulated and empirical data confirm the method's efficacy.

Conclusions:

• The spectral vector denoising algorithm offers a robust solution for hyperspectral image analysis.
• Improved spectral decomposition accuracy leads to more reliable biological specimen analysis.
• This technique enhances the utility of hyperspectral fluorescence microscopy in scientific research.