Title: Diagnostic Potentials of FTIR Microspectrometry in the Examination of Colorectal Adenocarcinomas
1Diagnostic Potentials of FT-IR Microspectrometry
in the Examination of Colorectal Adenocarcinomas
- Peter Lasch
- P25 "Biomedical Spectroscopy", Robert
Koch-Institut, D-13353 Berlin, Germany
2Objective
Development of an IR based, non-subjective and
automated technique for tissue characterization
3Sectioning
Tissue mounting on IR windows
- FT-IR microspectrometry imaging
- Tissue sections
- Image re-assembling on the basis of validated
classifiers
IR microspectrometry imaging
Image reassembling
IR spectra
Spectral diagnosis
4Problems
- Reproducibility
- Reproducibility or repeatability
- Standardization
- Sample preparation
- Measurement mode
- Spectral resolution
- Spatial resolution
- Water vapor
- Data analysis study design
- Study design
- Spectral preprocessing
- Models of classification
5Reproducibility and Repeatability (1)
- Repeatability is the variation in measurements
taken by a single person or instrument on the
same item and under the same conditions. A
measurement may be said to be repeatable when
this variation is smaller than some agreed limit.
According to the Guidelines for Evaluating and
Expressing the Uncertainty of NIST Measurement
Results, repeatability conditions include - the same measurement procedure
- the same observer
- the same measuring instrument, used under
the same conditions - the same location
- repetition over a short period of time.
(From Wikipedia, the free encyclopedia)
6Reproducibility and Repeatability (2)
Reproducibility is one of the main principles of
the scientific method, and refers to the ability
of a test or experiment to be accurately
reproduced, or replicated, by someone else
working independently. The term is very closely
related to the concept of testability and,
depending on the particular field, may require
the test or experiment to be falsifiable. ? Do
we measure repeatability or reproducibility? Futu
re studies should be carried out by a network of
laboratories, ideally in cooperation with
companies. Goal compilation of databases of
tissue spectra IMPORTANT Standardization of
the measurements
(From Wikipedia, the free encyclopedia)
7Standardization
- Important factors that needs to be standardized
in infrared microspectrometry and infrared
spectral imaging of tissues - Sample preparation
- Optical setup
- Type of optical substrates (window material)
- Spectral resolution (nominal resolution,
apodization, zero filling) - Spatial resolution (difficult)
- Other water vapor, carbon dioxide, hydration
status, ...
8Sample preparation and optical setup
- Sample preparation
- Cryosections or paraffin-embedded tissue
specimens? - Optical (instrumental) setup
-
- Transmission or transflection type measurements?
- confocal or non-confocal setup
- sample thickness
- type of optical substrates (window material)
9Window materials for IR tissue microspectrometry
Which type of optical material should be used
No final solution available For transmission
type measurements a number of materials are
suitable CaF2 suitable, but expensive,
restaining by HE possible, cut-off at ca. 950
cm-1 BaF2 like CaF2, water soluble ? restaining
impossible, cut-off at ca. 850 cm-1 ZnSe large
spectral range, visibility after restaining by
HE?, toxical Si relatively cheap (wafer
material), SiO2 absorptions, visibility after
restaining ?? Ge same as Si, but expensive, high
refractive index material for solid immersion
lenses?? PEN thin films of DuPont's
polyethylene naphthalate PET thin films of
DuPont's polyethylene terephtalate (Mylar)
10Polyethylene naphthalate (PEN) films
Tissue sections mounted on thin films of PEN
(2.5 µm) Initial background measured through
the film ? tissue spectrum indicates that
spectral contributions of PEN cannot be fully
compensated PEN subtraction routine similar to
water vapor subtraction routine?
11Spectral resolution
- Spectral resolution (or resolving power) of a
spectrograph is a measure of its power to resolve
features in the electromagnetic spectrum. It is
usually defined by - R ?/??
- where ?? is the smallest difference in
wavelengths that can be distinguished, at a
wavelength of ?. - Biomedical IR spectrometry What is the optimal
spectral resolution? - A number of factors define the spectral
resolution in FT-IR spectrometry - maximum retardation of a scan (nominal
spectral resolution) - post-FT parameters ZFF, apodization function
(From Wikipedia, the free encyclopedia)
12(No Transcript)
13Apodization
14(No Transcript)
15Spatial resolution
- Spatial resolution - most critical measurement
factor in IR imaging - What would be the optimal spatial resolution in
tissue microspectrometry? - Factors with an impact on the spatial resolution
- Geometry single detector mapping experiment
step size, aperture diameter array detector
measurements detector size, geometry of focal
plane array detectors - Optics mirrors, Cassegrain objectives, field
stop, confocal or non-confocal setup - Physics of far field microspectrometry
diffraction limit
Lasch P. Naumann D. 2006 Spatial Resolution in
Infrared Microspectroscopic Imaging of Tissues,
Biochim Biophys Acta (BBA) - Biomembranes 1758.
814829
16Spatial resolution determines the IR patterns
17Other factors water vapor
Purging of instruments required!
18(No Transcript)
19(No Transcript)
20(No Transcript)
21Study design and data analysis
- Study design
- Is the sample representative?
- Sample numbers?
- Classification or quantitative analysis?
- Covariates and confounding factors
- validity of the gold standard?
- Data analysis (classification)
- Quality test
- Spectral preprocessing
- Feature selection
- Strategies of classification concept driven
(supervised) or data driven - (unsupervised)
22How representative is your sample?
23How representative is your sample?
24Class definitions
25Class definitions
26Classification analysis
- Quality tests
- Spectral preprocessing
- Feature selection
- Strategies of classification concept driven
(supervised) or data driven (unsupervised)
27Quality tests
- Goal removal of outliers
- In IR imaging - five independent tests
- 1. Spectral contributions of water vapor
- 2. Sample thickness (integrated intensity)
- 3. Signal-to-noise-ratio (SNR)
- 4. "Test for an additional band" (e.g. issue
embedding medium) - 5. Bad pixel test to eliminate spectra from dead
pixels of focal plane array detector measurements.
28Spectral preprocessing
- Goal Remove the quantitative nature of the
spectral information - ? Increase robustness of classification
- Derivation
- Normalization (min/max, area, vector)
- Cut (spatially / spectrally)
- Subtraction
- Interpolation (spatially / spectrally)
- Smoothing
- Baseline correction
- Water vapor correction
- Dispersion correction
- MSC/EMSC, and many more
29Feature selection
- Feature selection, also known as variable
selection, feature reduction, attribute selection
or variable subset selection, is the technique,
commonly used in machine learning, of selecting a
subset of relevant features for building robust
learning models. - (From Wikipedia, the free encyclopedia)
- Why feature selection?
- Dimensionality reduction
- Enhancing generalization capability
- Speeding up learning process
- Improving model interpretability
30Unsupervised classification
- Unsupervised classification a data driven
technique - Unsupervised learning is a method of machine
learning where a model is fit to observations - Clustering various forms of clustering are
unsupervised - Agglomerative hierarchical clustering
- k-means clustering
- Fuzzy C-means clustering
- Quality threshold (QT) clustering
(From Wikipedia, the free encyclopedia)
Lasch P, Haensch W, Naumann D, Diem M. 2004
Cluster Analysis of Colorectal Adenocarcinoma
Imaging Data A FT-IR Microspectroscopic Study.
Biochimica et Biophysica Acta (BBA) - Molecular
Basis of Disease, Vol. 1688, Issue 2, 176-186
31Comparison of clustering techniques (1)
HE staining
FCM 2 classes
KMC 11 classes
AHC 6 classes
AHC 8 classes
AHC 11 classes
32Comparison of clustering techniques (2)
Hard software 966 MHz Intel Pentium IV
workstation (512 MB RAM), operating system W2k,
CytoSpec Imaging data set 8281 vector
normalized first derivative spectra in the
spectral region of 950 - 1750 cm-1 were clustered
and images of 11 clusters were re-assembled.
Third column a first order approximation of the
processing time dependence on the number of
objects (n) and the number classes (k)
Cluster analysis valuable explorative tool, but
not suitable to analyze spectral databases
33Supervised classification
- Supervised learning is a machine learning
technique for creating a function from training
data. The training data consist of pairs of input
objects (typically vectors), and desired outputs. - Example Artificial neural networks (ANN)
- Support Vector machines (SVM)
- The interplay of unsupervised and supervised
classification techniques in combination with
feature feature selection provides the optimum of
classification accuracy.
(From Wikipedia, the free encyclopedia)
34Classification analysis study design
35Example the colon database
- Spectral data from 28 patient samples, about 1.5
million IR spectra - 10 histopathological grading of G1
- 9 grading of G2
- 9 grading of G3
- IR Measurements
- Spectrum Spotlight, SR (nominal) 6.25 x 6.25 µm2
- Transmission mode, cryosections on CaF2
- 8 scans, 8 cm-1 spectral resolution
- Happ-Genzel apodization function
Lasch P, Diem M, Hänsch W, Naumann D, 2007
Artificial Neural Networks as Supervised
Techniques for FT-IR Microspectroscopic Imaging.
J. Chemometrics (Published Online 28 Mar 2007 )
36The colon database
- Tissue structure of point spectra
- Adenocarcinoma 1505
- Fibrovascular stock 778
- Mucin 335
- Fat 26
- Crypts 153
- Lymphocytes 41
- Lamina muscularis mucosae 160
- Tunica muscularis 126
- Necrosis 169
- Lamina propria mucosae 128
- Submucosa 34
- Vessel (blood, lymph) 193
37Optimization of the classifier I
38Optimization of the classifier II
39(No Transcript)
40Summary Outlook
Objective development of an IR based,
non-subjective and automated technique for tissue
characterization
- Excellent instrumentation for IR
microspectrometric imaging commercially available - Problems of standardization can be solved
- Concepts of study design class definitions
- A large-scale multicentric study to collect
database spectra is required. - Participation of companies important.