LVQ acrosome integrity assessment of boar sperm cells

1
LVQ acrosome integrity assessment of boar sperm
cells
University of León
University of Groningen

• Nicolai Petkov1, Enrique Alegre2 Michael Biehl1,
  Lidia Sánchez2
• 1University of Groningen, The Netherlands
• 2University of León, Spain

2
Contents
1. Introduction
2. Vectorization
3. Analysis by LVQ
4. Results
5. Conclusions
3
1. Introduction
4

• Quality assessment of semen, e.g. by measuring
• concentration, motility, morphology,
  intracellular pattern

5
Acrosome
6
Acrosome reaction and fertilization
7
Acrosome state

• Veterinary experts
• High fraction of acrosome-reacted cells
  means low fertilizing capacity

Acrosome intact
Acrosome reacted
8
Approach

• Fertilization potential estimation by
• Automatic image analysis for
• Estimation of the fraction of
• acrosome-intact sperm cells

9
2. Vectorization
10
Image acquisition
11
Cell head segmentation
cropping
thresholding
histogram stretching
Opening closing
12
Gradient computation
13
Gradient magnitude
Acrosome intact
Acrosome reacted
14
Gradient magnitude along head boundary
15
Gradient magnitude along head boundary
Acrosome intact
Acrosome reacted
16
3. Learning Vector Quantization
17
Labeled data
Vectors of gradient magnitudes along the contour
Class membership
Labeled data
P 152
18
LVQ1 training

• Select randomly example from D
• Find nearest prototype vector (winner)
• Update winner according to

moves prototype towards/away from the actual
example
19
4. Results
20
Prototype profiles
n 1
m 1
intact reacted
m 2
n 1
i n t a c t
reacted
21
Errors (8-fold cross validation)

• m and n prototypes of class 1 and 2, resp.

22
5. Conclusions

• Gradient magnitude along the cell head contour is
  a useful feature vector
• LVQ1 with 3 prototypes (2 for class 1) produces
  (training and test) errors of 0.165
• Veterinary experts call this sufficient for semen
  quality control in an artificial insemination
  center