Detection of crystals in Microarray images

1
Detection of crystals in Microarray images

• P.Dilip
• Rishabh Jain

2
Introduction

• Microarray grids are used by the chemical
  researchers who are interested in finding which
  chemical reactions and under what conditions
  produce crystals.
• A Microarray is a collection of chemical droplets
  in wells arranged as a grid

3

• Each microarray image has
• An octagonal boundary
• The central part inside the boundary which is
  highly illuminated. Illumination gradually
  decreases as we move away.
• A droplet inside the boundary.
• Crystals inside the droplet (found very rarely).

4
Microarray Cell
5
Work done so far

• Edge detection using Prewitt
• Applied Hough transform for circles on the edge
  detected image in order to detect the droplet.

6

• Applied Hough transform for straight lines (after
  converting the image into binary and then
  applying edge detection) in order to detect the
  octagonal boundary.
• Removed the part outside the boundary and kept
  only the necessary part for further processing.

7

• Modelled the background of the image obtained
  after removing the part outside the boundary.
• This was done so that we can extract the
  foreground of the image by filtering out the
  image with the background model already created.

8

• The background of many images are almost
  symmetric with respect to its vertical or
  horizontal or diagonal axis passing through its
  mid point.
• Used this property of the images to model the
  background by reflecting half of the image
  against its corresponding axis.

9

• Original Image Inside Boundary
• Background Foreground

10
Drop detection

• In the above foreground image the region in which
  the drop lies has higher illumination
• Applied a threshold on the foreground image to
  get the region with the drop

11
(No Transcript)
12

• The above method is not the best method of
  modeling background.
• Another method we propose to tackle this problem
  is first equalizing the illumination pattern in
  the image and then detect the region containing
  'interesting points'. This region will contain
  the droplet and hence the crystals inside (if
  present).

13

• The illumination pattern in the images follow a
  Gaussian like curve.
• Illumination pattern in image Gaussian curve

14

• Used a scaled version of gaussian to model the
  illumination pattern.
• Probed on different values of the scaling factor
  and standard deviation to get the optimal curve.
• Subtracted this from the image to equalize the
  illumination.

15
Illumination equalization
Before Gaussian After
16

• Detected the droplet using edge detection
  (sobel).
• After the removal of boundaries, the only edges
  left in most of the images are the edges in the
  droplet region.
• We use this property of the binary image obtained
  to detect the droplet region.

17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
Future work

• Try to equalize the illumination of the image
  using a more appropriate function which would be
  much more complicated than a Gaussian.
• Extract the features from the region we get and
  then train it with the help of labeled data.

21
References

• Digital Image Processing by Gonzalez Woods
• Image Analysis of Coccidian Parasite by Sumedha
  Gholba, Alisha Rossi and Bismaya Rath
• http//www.mathworks.com/
• http//homepages.inf.ed.ac.uk/rbf/HIPR2/hough.htm
• http//astronomy.swin.edu.au/pbourke/other/distri
  butions/