Premium
Potential of Hyperspectral Imaging for Label‐free Tissue and Pathology Classification
Author(s) -
Deal Joshua Andrew,
Favreau Peter,
Weber David,
Rich Tom,
Leavesley Silas
Publication year - 2016
Publication title -
the faseb journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.709
H-Index - 277
eISSN - 1530-6860
pISSN - 0892-6638
DOI - 10.1096/fasebj.30.1_supplement.51.2
Subject(s) - hyperspectral imaging , spectral signature , principal component analysis , spectral imaging , chemical imaging , sample (material) , signal (programming language) , wavelength , imaging spectroscopy , spectral line , pattern recognition (psychology) , materials science , artificial intelligence , biological system , optics , computer science , chemistry , remote sensing , physics , geology , biology , chromatography , astronomy , programming language
The ultimate goal of this project is to develop a novel imaging system that allows classification of both normal and pathogenic tissues based on fluorescence excitation spectra. Current imaging diagnostic techniques have limited capacity to classify normal tissue or detect pathogenic states. In vivo imaging suffers from poor signal to noise ratios and excessive data acquisition times while most alternative imaging techniques offer little molecular specificity. Our previous research has shown that excitation scanning hyperspectral imaging produces image data with high signal to noise ratios, fast acquisition times, and the potential to determine molecular composition. Here, we present data comparing characteristic spectral signatures for a range of tissue types. These data serve as a basis for establishing normal tissue spectra and characterizing molecular composition per tissue type. Tissues from Sprague Dawley rats were collected as 2 cm cubes and stored in cold PBS. Immediately following harvest, a custom inverted microscope (TE‐2000, Nikon Instruments) with a Xe arc lamp and an array of thin film tunable filters (VersaChrone, Semrock, Inc.) were used to collect hyperspectral image data from each sample. Scans utilized excitation wavelengths from 360nm to 600nm in 5nm increments. Three fields of view were recorded for each sample. Hyperspectral images were analyzed with custom Matlab scripts including linear spectral unmixing (LSU), principal component analysis (PCA), and Gaussian mixture modeling (GMM). Spectra were examined for potential characteristic features such as unique intensity peaks at specific wavelengths or ratios between intensity peaks. The resultant spectra vary among tissues types, which alludes to a unique spectral signature per tissue type. Additionally, excitation spectra appear to be a mixture of pure endmembers with commonalities in multiple tissue types, potentially identifiable through GMM. These results suggest that most tissues have common molecules and that intensity ratios of these molecules may serve as a signature for characterizing tissue composition. In addition, variations in signatures could indicate a change in molecular composition coincident with changes in pathology. Thus, excitation scanning hyperspectral imaging may be a potential diagnostic tool with the ability to distinguish tissues of varying pathology. Support or Funding Information This work is supported by P01HL066299 and the Abraham Mitchell Cancer Research Fund.