Spectral Similarity Measures for In Vivo Human Tissue Discrimination Based on Hyperspectral Imaging

Area of Science:

• Medical imaging
• Spectroscopy
• Biomedical engineering

Background:

• Spectral similarity measures are crucial for distinguishing human tissues in scientific research.
• Previous studies showed limited success in classifying pathological and non-pathological tissues using these measures.
• Hyperspectral (HS) imaging offers potential for in vivo tissue analysis.

Purpose of the Study:

• To evaluate spectral similarity measures for in vivo human tissue discrimination using HS images.
• To introduce and assess novel hybrid spectral measures: SID-JM-TAN(SAM) and SID-JM-TAN(SCA).
• To demonstrate the application of these measures in supporting HS image annotation and tissue labeling.

Main Methods:

• Analysis of spectral signatures from 13 human tissue types and two materials (gauze, instruments).
• Collection of HS images from 100 patients during surgical procedures.
• Evaluation of classical and newly developed hybrid spectral similarity measures.

Main Results:

• The proposed hybrid measures demonstrated superior tissue discrimination compared to classical methods, achieving up to 6.7 times higher values.
• Successful automatic checking of annotated thyroid and colon tissues in 73% and 60% of spectra, respectively.
• High accuracy (up to 90%) in automatic labeling of wrongly annotated tissues, with hybrid measures showing the best performance.

Conclusions:

• The developed spectral similarity measures reliably discriminate between human tissue types.
• Hybrid measures show significant potential for enhancing HS image analysis in surgical settings.
• Future integration with clinical tools will support physicians in HS image annotation and tissue labeling.