Christopher Mitchell - PowerPoint PPT Presentation

About This Presentation
Title:

Christopher Mitchell

Description:

The Cooper Union. Fluorescent Microscopy, Eigenobjects, and the Cellular Density Project ... The Cooper Union. Method Evaluation. For scope of project, ... – PowerPoint PPT presentation

Number of Views:49
Avg rating:3.0/5.0
Slides: 25
Provided by: ceme4
Category:

less

Transcript and Presenter's Notes

Title: Christopher Mitchell


1
  • Fluorescent Microscopy, Eigenobjects, and the
    Cellular Density Project
  • Christopher Mitchell
  • The Cooper Union
  • Stevens REU 2007

2
Overview
  • Previous Work Cell Density Fluorescent
    Microscopy
  • Outline of Project Methodology
  • How Eigenobjects Work
  • Applying Eigenobjects to Nuclear Density
  • The Future of the CDP
  • Note to any future presenters most of the
    content is in the accompanying Slide Notes

3
Fluorescent Microscopy
  • Attach marker chemical to protein
  • Take picture of tissue
  • Analyze marker distribution and concentration

4
Fluorescent Microscopy Uses
  • Examine tissue structure
  • Identify malignant cellular growth
  • Bind to nuclear protein to view distribution of
    nuclei in tissue
  • High density indicative of pre-malignant growth

5
Example of Fluorescent Microscopy
Example from webpage analyzed for this
presentation
6
Cellular Attributes
  • Normal Cells
  • Well-organized
  • Moderate density
  • Specific protein distribution
  • Malignant Cells
  • Chaotic arrangement
  • High density
  • Different protein distribution

7
Advantages of Fluorescent Microscopy
  • Less invasive
  • Earlier detection
  • More precise identification

8
The Future of Fluorescent Microscopy
  • More precise tagging of proteins to identify
    structures
  • Ability to tag multiple proteins to gather more
    information about each cell
  • Greater understanding of inter-cell and
    intra-cell structure as a cause and symptom of
    malignant cellular growth

9
Project Methodology
  • Application of some fluorescent microscopy
    methods to photographic microscopy
  • Primarily uses variance in cellular density and
    nuclear proportions between normal and malignant
    cells
  • Mechanical identification of suspect samples for
    further human or other analysis

10
Advantages ofPhotographic Microscopy
  • Easier to mechanically classify
  • Cheaper to acquire images
  • Human-readable with minimal training

11
Disadvantages ofPhotographic Microscopy
  • Less cellular detail
  • Fewer unique indicators of normal or malignant
    nature

12
Nuclear Identification Methods
  • Wavelet method
  • Signal processing-based (mathematical) solution
  • Eigencell method
  • Computer science (algorithmic) solution

13
The Signals Method
  • Increase image contrast
  • Edge detection using wavelets
  • Count nuclei and create density array
  • Apply statistical analysis

14
The Algorithmic Method
  • Creating training set of eigennuclei
  • Apply image space gt eigennucleus space gt image
    space transform, find Mean Squared Errors
  • Identify and count cells
  • Apply statistical analysis

15
Method Evaluation
  • For scope of project, algorithmic method chosen
  • Easier to code, easier to understand without a
    Signals background
  • More precise even though less efficient

16
Using Eigenobjects
  • To create a training array of eigenobjects, need
    to start with several training images.
  • All images must be the same dimensions
  • Example training set
  • Varied sizes and rotations, but all 24x24 pixel
    images

17
Using Eigenobjects 2
  • Next, all images packed from rectangles into rows
  • Eigenvectors created from each row and sorted by
    associated eigenvalues

18
Using Eigenobjects 3
  • In order to streamline the process, the outer
    product of each row is taken and packed
  • Yields square, symmetric matrix
  • Each row multiplied by original image produces
    one eigenobject

19
Using Eigenobjects 4
  • Trained set of eigenobjects is complete and
    packed into a single array for comparison
  • To compare an image to the training set, it must
    be converted to object space and back to image
    space.
  • Examples of eigenfaces, eigenobjects made from
    faces

20
Using Eigenobjects 5
  • Results of image gt object gt image space
    transformations
  • To determine if the image is the same type of
    object as training set, take Mean Squared Error
    (MSE) between input and output

21
Finding Objects in an Image
  • Method can be applied to find objects in a larger
    image
  • All possible subimages of training set dimensions
    taken, MSEs calculated
  • Threshold-filtered to find objects

22
Applications to Nuclear Density and the CDP
  • Using eigennuclei, center of all cells in
    microscope image can be found
  • Image broken into regions, number of cells in
    each region found
  • Statistical analysis to determine cancer presence

23
The Future of the CDP
  • Optimizations
  • Multipass approach
  • Scaling/rotation
  • Further identification metrics

24
References
  • Cytodiagnosis of Cancer Using Acridine Orange
    with Fluorescent Microscopy (http//caonline.amca
    ncersoc.org/cgi/reprint/10/4/118)?
  • New Cell Imaging Method Identifies Aggressive
    Cancers Early (http//www.sciencedaily.com/releas
    es/2006/03/060307085017.htm)?
  • The Cellular Density Project (http//beta.cemetech
    .net/projects/item.php?id1)?
  • Eigenfaces Group Algorithmics(http//www.owlnet
    .rice.edu/elec301/Projects99/faces/algo.html)?
  • Eigenfaces (http//www.cs.princeton.edu/cdecoro/e
    igenfaces/)?
Write a Comment
User Comments (0)
About PowerShow.com