Most Relevant Spectral Bands Identification for Brain Cancer Detection Using Hyperspectral Imaging

Area of Science:

• Medical Imaging
• Biomedical Optics
• Computational Biology

Background:

• Hyperspectral imaging (HSI) offers rich data beyond RGB imaging for medical diagnostics.
• HSI data often contains redundancy, necessitating relevant wavelength identification for improved accuracy and efficiency.
• Noise in specific wavelengths can hinder classification, making band selection critical.

Purpose of the Study:

• To identify optimal spectral ranges in the visual-and-near-infrared (VNIR) region for brain cancer detection using in vivo HSI.
• To develop a methodology for selecting the most informative wavelengths to enhance classification accuracy and reduce computational load.
• To enable the development of real-time HSI acquisition sensors.

Main Methods:

• A methodology employing optimization algorithms, specifically a genetic algorithm, was used to identify relevant spectral bands.
• Support vector machines (SVM) were utilized as the supervised classifier to evaluate the accuracy of selected wavelengths.
• The study focused on in vivo hyperspectral images of brain tissue.

Main Results:

• The genetic algorithm optimization identified key spectral ranges, improving tumor identification accuracy by approximately 5% compared to using all 128 bands.
• The optimized approach utilized only 48 spectral bands, significantly reducing data volume.
• Identified relevant spectral ranges include 440.5-465.96 nm, 498.71-509.62 nm, 556.91-575.1 nm, 593.29-615.12 nm, 636.94-666.05 nm, 698.79-731.53 nm, and 884.32-902.51 nm.

Conclusions:

• The proposed optimization methodology effectively identifies crucial spectral bands for brain cancer detection using HSI.
• This approach enhances classification accuracy and reduces the number of required spectral bands, paving the way for real-time HSI systems.
• The identified spectral ranges provide valuable information for developing targeted HSI sensors for neuro-oncology applications.