Post on 04-Jul-2015
Automatic Cell Counting
Ege EnginMiddle East Technical University
Supervisor: Tomas LukesCzech Technical University
http:/blog.metu.edu.tr/e174088
Agenda
• Introduction
• Different Approaches• Method 1• Method 2• Method 3
• Comparison & Results & Comments
• Nine-Point Circle for Cell Segmentation• Nine-Point Circle Rule• How to use?• Results• Implementation
• Conclusion and further work
Background
• Learn image processing basics
• Fundamentals
• Intensity Transforms
• Spatial Filtering
• Frequency Domain Processing
• Image Segmentation
• Implement different methods in MATLAB and compare them
• Based on the results, improve the algorithm
1st and 2nd Method
1st Method
• Thresholding:• im2bw with graythresh
• Filling with holes• imfill with holes
• Counting:• Numobjects of
bwconncomp
2nd Method
• Thresholding:• im2bw with graythresh
• Filling with holes• imfill with holes
• Counting:• Lenght of bwboundaries
3rd Method
3rd Method• Image: Cell 1
Comparison & Results1st Method
2nd Method
3rd Method
3rd Method( with watershed)
Exact Number(includes incomplete objects)
Coins 10 37 10 12 10
Eight 1 263 4 3 4
Rice 151 158 93 67 101
Cell Image 1 36 518 53 42 50
Cell Image 2 270 2369 50 44 38
Cell Image 3 287 1139 25 75 64
• Cell Image 2• Cell Image 1 • Cell Image 3
Comments• First two approaches uses nearly the same algorithm except their
counting methods.
• The first algorithm underestimate the number of objects.
• The second algorithm overestimate the number of objects.
• 3rd approach :
• Without watershed segmentation has achieved promising results when the cells in the image are not connected.
• With watershed segmentation over-segments or mis-segments the figure, so overestimate the number of objects.
• Challenges: connected cells, incomplete cells on the borders
• Another solution technique can be useful for the solution of connected cells problem.
Nine-Point Circle • Also known as the Feuerbach Circle
• In every triangle, lie on a circle:
• The three midpoints of the sides
• The three base points(feet) of the altitudes
• The midpoints of the three segments from the orthocenter to the vertices
Reference: Dorrie, H. and Woltermann, M, '100 Great Problems of Elementary Mathematics', reworked in 2010
How to use?• Arbitrarily select three points from cell
• Example: A,B,C
• Find 9 points which are• 3 midpoints
• Example: D,E,F
• 3 base points(feet) of the altitudes• Example: G,H,I
• Midpoints of 3 segments from the orthocenter to the vertices
• Example: K,L,M• Please note that: J is orthocenter.
• Count the points inside the cell:• D,E,G,K,L,M => 6 inside / 9 total
• Repeat the procedure until the average is an appropriate result
Results• For Results:
• For Implementation:
Conclusion• When useful?
• When cells have circular shapes
• Why?
• The more circular the shape, higher the average algorithm givesso,When the cells are not connected, the average will be higher.
• Can help to distinguish the cells whether connected or not.
• Why important?
• For cell segmentation, not tried before ( based on subjective(my) research)
• Challanges?
• Defining the standard to understand whether cells are connected
• Clear bordering is necessary to run correctly
• Hard to distinguish if cells are connected circularly
Future Study
• For nine-point rule cell segmentation:
• Divide the given image in appropriate divisions
• Try on cell images
• Documentation related to all studies
Questions
Reference: Velasquez J., Dhuru P. & Vaghani M. , ‘Matlab Digital Image Processing For Microscopy Screening’
Thank you!
Reference: A blood test and examination, 1941 – 1945 retrieved byhttp://research.archives.gov/description/513715 on 20.08.2013
3rd Method
3rd Method
3rd Method