Diffuse reflectance and machine learning techniques to differentiate colorectal cancer ex vivo

Area of Science:

• Biomedical Optics
• Machine Learning Applications
• Cancer Diagnostics

Background:

• Accurate optical property reconstruction is crucial for non-invasive tissue analysis.
• Distinguishing between normal and pathological colorectal mucosa requires sensitive detection methods.

Purpose of the Study:

• To develop and validate machine learning models for reconstructing the wavelength-dependent absorption coefficient of colorectal mucosa.
• To assess the ability of these models to identify biochemical differences, such as lipofuscin and hemoglobin content, between healthy and cancerous tissues.

Main Methods:

• Utilized diffuse reflectance spectra from ex vivo human colorectal mucosa as input for machine learning algorithms.
• Compared multilayer perceptron regression and random forest regressor models for spectral reconstruction.
• Analyzed reconstructed absorption spectra to quantify lipofuscin accumulation and hemoglobin ratios.

Main Results:

• Multilayer perceptron regression achieved a good match with invasive spectral measurements.
• Both methods identified differentiated lipofuscin accumulation in pathological tissues.
• Calculated hemoglobin ratios indicated higher blood content in pathological samples, consistent with invasive measurements.

Conclusions:

• Machine learning effectively reconstructs optical properties of colorectal mucosa from diffuse reflectance.
• The method shows sensitivity to biochemical differences, enabling lipofuscin and hemoglobin quantification.
• This approach holds promise for developing minimally invasive spectroscopy for colorectal cancer detection and monitoring.