Hyperspectral Imaging Applied to Medical Diagnoses and Food Safety

1
Hyperspectral Imaging Applied toMedical
Diagnoses and Food Safety

• Presented by
• Richard Gomez
• Oscar Carrasco
• SPIE AeroSense 2003, Orlando, Florida
• April 24, 2003

2
Outline

• Hyperspectral imaging (HSI)
• Medical Diagnoses
• Early cancer detection
• Cervical cancer
• Breast cancer
• Retinal imaging
• Tissue characterization
• Food safety illness prevention

3
Hyperspectral Imaging

• Hundreds of spectral channels, each channel
  covering a narrow and contiguous portion of light
  spectrum
• Allows analyst to perform
• Reflectance redirection
• Emittance objects radiance
• Fluorescence ? excitation
• spectroscopy of each pixel of image scene

Image post-mortem histology however, majority
of histology is conducted by pathologist for
disease determination purposes in support of
diagnosis.
4
Medical Imaging Market

• Estimated at just over 4 billion in 1999 and is
  projected to reach 5.4 billion in 2005
  www.galileo-gp.com
• Research for new optical methods for spectral
  discrimination combined with powerful software
  approach to obtain more information than
  color-based imaging approaches. Includes
• Algorithm to produce spectral information in a
  specified frequency band with a specified
  resolution
• Signal enhancement algorithms to improve
  Signal-to-Noise ratio of interferograms prior to
  transformation

5
Medical Imaging

• Conventional diagnostics allow doctors to see
  inside the body using high frequency energy
• Nuclear medicine X-rays and d-rays
• Ultrasound (sound waves)
• Magnetic resonance imaging (MRI) radio waves
• Computed tomography (CT CAT scans, X-rays)
• Multispectral hyperspectral offer diagnostic
  testing for outside the body or other surfaces
• Hyperspectral imaging microscopy permits the
  capture and identification of different spectral
  signatures present in an optical field during a
  single-pass evaluation, including molecules with
  overlapping, but distinct emission spectra.
  Cytometry, March 15, 2001

6
Hyperspectral Medical Imaging

• Optical imaging specifically, reflective portion
  of electromagnetic spectrum
• V/NIR range (0.38 0.725 1.35 µm)
• Wavelength used is a function of molecular
  excitation
• e.g., skin found to be 0.5250.645 µm

7
Hyperspectral Medical Imaging

• Focused on energy interaction with material with
  respect to
• fluorescence illumination with one ? (usually
  UV 0.3-0.38 µm) interacts with target material
  and emits radiation at a different ?.
• Molecules (fluorochromes or fluorophores e.g.,
  tryptophan, collagen, some proteins) have the
  ability to absorb light of a shorter, higher
  energy wavelength and re-emit it as a longer, low
  energy wavelength Stokes shift where the
  illuminated fluorochromes become excited into a
  higher, unstable energy state, which is then
  relieved by the subsequent production emission
  of photons of lower frequency.
• Certain cellular components fluoresce briefly
  when excited by specific ?
• reflectance energy measured in very precise ?,
  very small increments

8
Fluorescence Spectroscopy Process

• Illumination source
• quartz-tungsten-halogen light
• Hg burner Ar-Ion Laser
• Filter reflected light spectrally discriminated
  and imaged onto a silicon CCD detector.
• Measured reflectance spectra are quantified in
  terms of apparent absorbance and formatted as a
  hyperspectral image cube.

9
Hyperspectral Platforms

• Imaging Spectrometer
• Interferometric device coupled with a scene
  camera for relaying optics (illuminate and
  collect reflected light)
• HSI Microscope research/testing phase
• Collects complete fluorescent spectrum from a
  region of a microscope slide
• Hardware microscope, camera, band-sequential
  filters, light source)
• Software control extraction of data from entire
  spectra available for each pixel includes
  curve fitting
• For both techniques, the key is to determine
  abnormal cell changes (suspicious areas) in
  relation to normal cell development.

10
Microscope HSI systemSource http//innovation.s
wmed.edu/Instrumentation/HIC_images.htm
11
Early Cancer Detection

• Displasia/neoplasia presents with distinct
  molecular characteristics, including
• Nuclear content and size
• Epithelial thickness
• Increased vascularity/neovascularization blood
  to the area (angiogenesis)
• Increased metabolism and greater nutrition demand
• Precancerous areas radiate slightly more heat at
  specific sites of neoplastic cells
• Studies focus on cervical and breast cancer

12
Cervical Intraepithelial Neoplasia CIN

• Suspected causal factors carcinogens, multiple
  cell mutations, virus, more likely from
  multiple causal factors
• Precursors hemoglobin from increased
  vascularity in subsurface vessels of the cervix
• High absorption coefficient compared to other
  tissue constituents
• Abnormality in the scattering of abnormal cells
  in the epithelium
• Most commonly is Human Papilloma Virus (HPV)
• Traditional diagnostic and treatment
• Pap smear
• Colposcopy (examination with powerful microscope)
• Biopsy to remove abnormal cells

13
Cervical Cancer Detection (contd)

• Imager placed about 8 in. from cervix
• Accurate image showing pre-cancerous cells
• Abnormal areas gt Spectroscopy-directed biopsy
• UV Illumination (graphic results next slide)
• Range 470-480 nm, fluorescence emission curve
  generated by averaging pixels at pre-cancerous
  sites.
• White light reflected light in the visible
  range
• Reflectance spectra image centered 546 nm
• HSI cannot distinguish pathological grade (CIN-2)

14
Cervical Cancer Detection (contd)
Fluorescence no displasia
Fluorescence displasia
Source http//www.spectrx.com/Techdata/Cancer/Eu
roginPres2000.pdf
15
Cervical Cancer Detection (contd)
Source Data presented at the Pacific Medical
Technical Symposium, Honolulu, HI. August 17-20,
1998
16
Breast Cancer Detection

• Traditional screening test
• Mammography
• X-ray to visualize the internal structure of the
  breast
• Thermal scanning
• Limited to single IR band with one camera
• HSI sensor and spectral data vector analysis
• MS-IR image, apply unsupervised classification
  algorithm based on multiple spectral data per
  pixel, classify IR heat radiated from abnormally
  reproducing breast cancer cells
• Non-intrusive screening without radiation hazard
• Detects in-situ carcinomas before traditional
  tests

17
Breast Cancer images Source http//www.onr.navy.
mil/media/release_display.asp?ID117
Single IR camera. Left Healthy woman in cooled
room Right same woman 10 minutes later, showing
most heat dissipated.
Single IR camera. Left Breast Cancer patient in
cooled room. Right same woman 10 minutes later,
showing active cancer cells surrounded by tissue
where most heat dissipated.
18
Breast Cancer images (contd) Source
http//www.onr.navy.mil/media/release_display.asp?
ID117
2-camera MS-IR breast image. Left medium ? (3-5
m) less penetration Right long ? (8-12m)
greater penetration Transcribes thermal diffusion
process into two images, filtered for shared
signals while disagreement noise is minimized.
Unsupervised classification of right breast,
indicating abnormal heat distribution near breast
nipple. A doctors diagnosis was Ductal Carcinom
In Situ (DCIS) cancer state zero.
19
Skin Care

• Precancerous areas reveal
• Increased metabolism
• Increased vascularity
• angiogenesis either new vessels forming or
  enlarged due to nutritional needs of abnormal
  cells
• Healthy blood flow to an area will give a
  specific spectral signature
• Angiogenesis can then be noted examined for
  dermal carcinomas

20
Dermal lesions/carcinoma detection

• Application of HSI
• Preventive care analyze abnormal skin lesions
• Dependent on skin color (for reflectivity
  comparisons)

21
Note Early Cancer Detection

• Although HSI technology may aid in cancer
  detection, it is important to note that, at this
  time, this technology should be used in
  conjunction with traditional diagnostic methods.
• Before the use of HSI is widely accepted, there
  needs to be in-vitro studies to determine optimal
  wavelengths for determining spectral imaging
  accuracy.
• For now, HSI will better serve to determine the
  spread of cancerous cells and guide doctors to
  spectroscopy-directed biopsies.

22
Retinal Disease Detection

• Optical imaging to understand pathological
  changes in morphology, abnormal patterns, and
  colors of the retina.
• Normal retina
• reddish appearance due to presence of
  chromophores hemoglobin and melanin that absorb
  more strongly at shorter wavelengths
• Abnormal retina
• change in physiology results in altering the
  reflectivity spectrum
• Diabetic Retinopathy (hemorrhage)
• Age-Related Macular Degeneration (abnormal
  protein/lipid concentrations)
• Abnormal intensity response signifies imbalance
  of melanin and hemoglobin concentrations, and
  surface lesions

23
Tissue Characterization

• Assess blood flow through tissue
• Perfusion/reperfusion versus ischemia
• Traditionally, oximetry test
• Monitors the V/NIR spectral properties of blood
  by recording variations in the percentages of
  oxygen saturation of hemoglobin

24
Tissue Characterization

• Newer tests to determine overall (not at one
  point) perfusion
• Use visible reflectance of body tissue
• Tested using palmar surface of hand (low melanin
  content)
• Results HSI generates reliable images which
  provide spatially resolved chemical information
  in either a static or time-resolved mode.
• This enables doctors to image anatomical features
  with sub-millimeter spatial resolution while
  visualizing chemical changes at the molecular
  level.

25
Potential Medical applications

• Surgery
• Closing after a surgery although closing is by
  performed 1 layer at a time, it could yield
  information on tissue O2 levels
• Diagnostics
• ABG Arterial Blood Gases key test
• Non-invasive, in-vivo O2 determination
• Fluid analysis (function of turbidity/scattering)
  looking for specific agent, not overall
  analysis
• Blood
• Urine
• Semen
• Use of HSI is to be based on efficacy, as seen
  by clinical trials

26
HSI Medical Benefits

• Less invasive
• Accurate diagnostic preventive tests
• Provide for targeted biopsies
• Fewer false positive/negative results
• Decrease probability for on-site testing errors
• Potential to support clinical judgment
• Real-time results
• Cost-effective reducing health care costs
• Useful in effective decision-making by medical
  professionals during diagnosing, follow-up (more
  specific) testing, therapy selection,
  monitoring.

27
Food Safety

• HSI inspection to support safe handling,
  manufacturing, processing, and monitoring of the
  food supply
• Raw poultry products
• Produce fruits/vegetables
• Grains
• Nutritional supplements
• Organic foods laid out (in buffet-style)
• Aid in detecting, identifying, and separating
  contaminated foods

28
Food Safety

• Detect spectral signatures of
• Dirt
• Fly specs
• Fungi
• Fecal matter
• Ingesta (partially digested food from the
  ruptured corps of chicken carcasses)
• Microbial pathogens
• Salmonella
• Escherichia Coli O157H7 (E. Coli)

29
HSI Food Safety Benefits

• Shorter detection time
• Acquisition of unique spectra for bacteria
• Permits accurate results
• Monitoring large quantity of foods
• Identification, removal, cleaning of carcasses
  will further reduce cross contamination during
  production

30
Spectral Library Issue

• Health care industry needs to have a spectral
  library in which a range of spectral signatures
  need to be readily available.
• This range needs to exist for disease detection
  because although skin color may affect the
  signature, it still may be accurate.
• Library issues
• Standards
• Data handling and interpretation
• Accessibility and dissemination

31
Conclusions

• An effective HSI system for medical imaging is
  still in its infant stages.
• Clinical tests to determine optimal wavelengths
  for diagnosis will provide evidence for future
  applications in this area.
• Need to establish a spectral library of
  contaminants to ensure proper HSI monitoring of
  foods.
• Regardless of which field (medical imaging or
  food safety) sees the application of HSI
  technology first, data gathered, analyzed, and
  interpreted needs to be scientifically accurate
  to be of any value during decisions concerning
  illness potential and disease detection.

32

• Questions/Comments