|
HIGHLIGHT OF THE DMD |
||
|---|---|---|
|
|
A complete DMA contains 848 columns and 600 rows of mirrors and measures 10.2 mm x 13.6 mm. Here, a DMA is shown with its glass cover removed. 
|
|
Hyper-Spectral Imaging and Pathology
Hyper-spectral microscopic discrimination between normal and cancerous colon biopsies. Franco Woolfe, Mauro Maggioni, Gustave Davis, Frederick Warner, Ronald Coifman, and Steven Zucker, submitted, 2006.
Algorithms
from Signal and Data Processing Applied to Hyperspectral Analysis:
Discriminating Normal and Malignant Microarray Colon Tissue Sections
Using a Novel Digital Mirror Device System,. M.Maggioni, G.L.
Davis, F. J. Warner, F. B. Geshwind, A.C. Coppi, R.A.DeVerse,
R.R.Coifman, Tech Report, Dept. Comp. Science, Yale University,.
November 2004.
Hyper-spectral
Analysis of normal and malignant colon tissue microarray sections
using a novel DMD system, G. L. Davis, M. Maggioni, F. J.
Warner, F. B. Geshwind, A. C. Coppi, R. A. DeVerse, R. Coifman,
poster session at Optical Imaging NIH workshop, Sep.
2004.
Hyper-Spectral
analysis of normal and malignant microarray tissue sections using a
novel micro-optoelectricalmechanical system,
with G.L. Davis, M.
Maggioni, F. J. Warner, F. B. Geshwind, A.C. Coppi, R.A.
DeVerse, R.R.Coifman.
Spatial-Spectral
Analysis of Colon Carcinoma, with G.L. Davis,
R.R.Coifman, R.Levinson.
Introduction
With light sources of increasingly broader ranges, spectral analysis of tissue sections has evolved from 2 wavelength image subtraction techniques to hyperspectral mapping. A variety of proprietary spectral splitting devices6,9,13, including prisms and mirror, interferometer, variable interference filter-based monochromometer & tuned liquid crystals, mounted on microscopes in combination with CCD cameras and computers, have been used to discriminate among cell types2,7,8,9,13 & endogenous & exogenous pigments1,13 .
Goals
We use a prototype unique tuned light source, a digital mirror array device (Plain Sight Systems) based on micro-optoelectromechanical systems5, in combination with analytic algorithms developed in the Yale Program in Applied Mathematics2,3,11, to evaluate the diagnostic efficiency of hyperspectral microscopic analysis of normal & neoplastic colon biopsies prepared as microarray tissue sections14. We compare the results to our previous spectral analysis of colon tissues10 and to other spectral studies of tissues and cells.
Experimental Details
Platform: The prototype tuned light digital mirror array device5 (Figure 1) trans illuminates H & E stained micro-array tissue sections with any combination of light frequencies, range 440 nm – 700 nm, through a Nikon Biophot microscope. Hyper-spectral tissue images, multiplexed with a CCD camera (Sensovation), are captured & analyzed mathematically with a PC.
|
HIGHLIGHT OF THE DMD |
||
|---|---|---|
|
|
A complete DMA contains 848 columns and 600 rows of mirrors and measures 10.2 mm x 13.6 mm. Here, a DMA is shown with its glass cover removed.
|
|
Image
Source: 147 (76 normal & 71 malignant) hyperspectral gray
scale 400X images are derived from 68 normal and 62 malignant colon
biopsies selected from @ 200 normal and @600 malignant H & E
stained biopsies arrayed respectively on two different slides12
(Figure 2).
Cube: Each hyperspectral image is a 3-D data cube (Figure 3) with spatial coordinates x - 491 pixels, y - 653 pixels & spectral coordinates z - 128 pixels, a total of 41 million transmitted spectra.
|
|
|
|---|---|
|
Figure 2: Microarray biopsies |
Figure 3: Hyperspectral data cube (image from DataFusion Corp) |
|
|
|
|
|---|---|---|
|
Average nucleus spectrum with standard deviation bars. |
A spectral slice of a normal gland |
A spectral slice of a cancerous gland |
|
TISSUE CLASSIFICATION |
|
|---|---|
|
|
|
|
|
|
|
Tissue classification algorithm on a sample |
Tissue classification algorithm on a sample |
|
ALGORITHM FOR NORMAL/ABNORMAL DISCRIMINATION |
|
|---|---|
|
|
|
|
Normal:
|
Abnormal:
|
|
Table 1: Performance of classifier on nuclei patches, cross-validated . |
||
|
Patches/Nuclei (8688) |
True Positive (4860) |
True Negative (3828) |
|---|---|---|
|
Predicted Positive ( malignant) |
94.0% (4568) |
7.3% (280) |
|
Predicted Negative normal) |
6.0% (292) |
92.7% (3548) |
|
CLASSIFICATION OF A WHOLE SLIDE |
|
|---|---|
|
The classification of a whole slide is obtained by selecting 40 random nuclei patches from the slide, and averaging the corresponding classifications. The classification of the slides has no mistakes, since the few errors of classification on the nuclei are averaged out on the whole slide. |
|
|
A SPATIAL-SPECTRAL CLASSIFIER |
|
|---|---|
|
|
|
Number of visitors since September 10th,
2004:
