Differentiability of cell types enhanced by detrending a non-homogeneous pattern in a line-illumination Raman...

Area of Science:

• Biomedical Optics
• Chemical Imaging
• Machine Learning in Spectroscopy

Background:

• Line illumination Raman microscopy offers rapid spatial and spectral data acquisition for biological samples.
• Non-uniform laser intensity in line illumination can introduce artifacts, compromising data accuracy and machine learning model performance.
• Standard spectral preprocessing methods, optimized for raster scanning, may not be suitable for line illumination, potentially introducing artifacts.

Purpose of the Study:

• To investigate artifacts introduced by standard spectral preprocessing in line illumination Raman microscopy.
• To develop and validate a new detrending scheme to mitigate artifacts caused by non-uniform laser illumination.
• To enhance the differentiation accuracy between cancerous and normal human thyroid follicular epithelial cells using the proposed method.

Main Methods:

• Utilized cancerous (FTC-133) and normal (Nthy-ori 3-1) human thyroid follicular epithelial cell lines.
• Developed a detrending scheme combining random forest regression with position-dependent wavenumber calibration.
• Compared the proposed detrending scheme against standard preprocessing methods for spectral analysis.

Main Results:

• Standard preprocessing methods introduced artifacts in line illumination Raman microscopy data.
• The proposed detrending scheme effectively minimized artifactual biases stemming from non-uniform laser sources.
• The new method significantly improved the differentiability between cancerous and normal thyroid epithelial cell states.

Conclusions:

• A novel detrending scheme based on random forest regression is effective for line illumination Raman microscopy.
• This approach overcomes limitations of standard preprocessing, reducing artifacts and improving spectral data quality.
• The enhanced spectral analysis facilitates more accurate classification of biological samples, such as cancerous versus normal cells.