Unsupervised and supervised methodologies for identification of sample pixels in Fourier transform infrared...

Area of Science:

• Biomedical Optics
• Spectroscopy
• Computational Biology

Background:

• Mid-infrared (MIR) spectroscopy offers label-free analysis of molecular alterations in biological tissues.
• Fourier Transform InfraRed (FTIR) microspectroscopy generates chemical maps by acquiring spectral data from multiple spatial points.
• Effective pre-processing of FTIR images requires accurate identification of sample pixels, distinguishing them from background noise.

Purpose of the Study:

• To present and compare three distinct methodologies for sample pixel identification in FTIR images.
• To evaluate the performance of unsupervised and supervised approaches in distinguishing sample from background pixels.
• To validate these methods using experimentally acquired FTIR images of tissue sections.

Main Methods:

• Development and implementation of three algorithms for pixel classification in FTIR images.
• Utilizing both unsupervised and supervised machine learning techniques.
• Experimental validation on FTIR microspectroscopic data from diverse organ tissue sections.

Main Results:

• All presented algorithms demonstrated high accuracy in predicting sample and background pixels.
• The supervised methodology achieved automatic detection of sample pixels.
• Robust performance was observed across FTIR images from multiple organs.

Conclusions:

• The developed methodologies provide effective solutions for sample pixel detection in FTIR imaging.
• Accurate pixel identification is essential for reliable feature extraction and analysis of FTIR data.
• These findings advance FTIR signal processing, supporting future chemical and clinical applications.