Linear Support Vector Machine Model of an IR Spectral Library: Application to Skin Cancer

Area of Science:

• Biomedical Optics
• Spectroscopy
• Machine Learning

Background:

• Infrared (IR) spectroscopy offers potential for non-invasive skin cancer detection.
• Developing rapid, portable devices is crucial for clinical translation.
• Murine models closely mimicking human disease are vital for research.

Purpose of the Study:

• To derive and measure the
• Decision Contribution Spectrum
• using IR spectroscopy and support vector machines (SVM).
• To assess the feasibility of a fast, hand-held spectral probe for human skin cancer detection.
• To evaluate a fiber-loop sensor probe for external cancer detection on live mice.

Main Methods:

• Acquired over 4,000 IR spectra from a murine squamous cell carcinoma model (SKH1 mouse tumor) using Fourier Transform Infrared (FTIR) microscopy.
• Trained and tested a linear SVM model to identify key spectral features and assess data reduction without quality loss.
• Designed and tested a reduced-range, reduced-sampling mid-IR spectral probe and a fiber-loop sensor probe on live mice.

Main Results:

• The
• Decision Contribution Spectrum
• was successfully derived and measured, highlighting contributions to tumor/non-tumor classification.
• A linear SVM model effectively selected features, demonstrating potential for data compression.
• Preliminary tests on live mice showed the fiber-loop sensor could detect tumors externally without adverse effects.

Conclusions:

• The study provides a novel spectroscopic method for analyzing spectral data in cancer detection.
• The developed spectral probe designs and preliminary sensor results support the feasibility of a hand-held device for human skin cancer detection.
• Further research is warranted to justify clinical trials on live human skin.