Detection of clusters of small features such as microcalcifications

1
Detection of clusters of small features such as
microcalcifications
BREAST EXPERTS

• Baris Ozyer, ANKARA
• Fatih Titrek, ANKARA
• László Csernetics, SZEGED
• Levente Ficsór, BUDAPEST

2
OUTLINE

• The Problem
• Digital Mammography
• Literature
• Steps to solve the problem.
• If we could have more time.

3
The Problem

• Detection of clusters of small features such as
  circles or microcalcification on a noisy
  features background
• Input Given an image like a mammogram with the
  presence of microcalcifications of different size
  and shape which can be introduced by simulation
  for the purpose of this project
• Output Images with detection indicated

4
Digital Mammography

• Definition Digital mammography is a mammography
  system where x-ray film is replaced by
  solid-state detectors that convert x-rays into
  electric signals.
• The electrical signals are used to produce images
  of the breast that can be seen on a computer
  screen or printed on special films to look like
  regular mammograms.

5
The Advantages of the Digital Mammography

• Fewer patient calls back for additional images.
• Less anxiety for patience
• Less time for doctors
• The doctors can electronically manipulate images

6
Used Programmes to improve the results

• Matlab
• Khoros
• C
• ImageJ

7
Some words about the problem

• Some problems to detect the microcalcifications
  on the images
• High variety of microcalcifications
• High variability of background

8
Literature

• Detection of Microcalcifications in Digital
  Mammograms Using Wavelets, Ted C. Wang and
  Nicolaos B. Karayiannis
• tophat algorithm was applied to obtain unique
  markers for
• Opening, Subtraction,Thresholding
• The tophat algorithm is a morphological transform
  that is used to extract either locally bright or
  locally dark objects, with the use of shape
  information and relative brightness.
• The numerical analysis of the detected
  microcalcifications

9
More Literatures

• Characterization of clustered microcalcifications
  in digitized mammograms using neural networks and
  support vector machines. Papadopoulosa, D.I.
  Fotiadisb, A. Likas, 2005
• Breast cancer diagnosis system based on wavelet
  analysis and fuzzy-neural, Rafayah Mousa,
  Qutaishat Munib, Abdallah Moussa, 2004
• Detection of single and clustered
  microcalcifications in mammograms using fractals
  models and neural networks, L. Bocchi, G.
  Coppini, J. Nori, G. Valli, 2003

10
What is our plan ?

• Normalization to enhance contrast
• Smoothing
• Morphological edge detection
• Threshold
• Finding rings
• Measure the detected objects

11
Normalization to enhance constrast

• Typically normalization is attempting to remove
  global effects, that can be seen by examining
  plots that show all the data for a slide or
  slides.
• Normalization does not necessarily have anything
  to do with the normal distribution that plays a
  prominent role in statistics.

12
Smoothing

• Smoothing is the process of taking an image and
  blurring it so that it looks out of focus. To
  find the edge it is better use smoothing
  technique.

13
Morphological edge detection
The simply difference between a dilated and an
eroded image could be define an edge.

• Smoothed Image

Detected Image
14
Tresholding
The segmentation is determined by a single
parameter known as the intensity threshold. In a
single pass, each pixel in the image is compared
with this threshold. If the pixel's intensity is
higher than the threshold, the pixel is set to,
say, white, in the output. If it is less than the
threshold, it is set to black.

• Detected Image

Tresholded Image
15
Tresholding Detecting the circle
16
Measurements

• Some kind of microcalcifications

17
Further improvement in segmentation

• Local normalization of image parts
• Segment by homogeneity and intensity
• Normalization for the current region
• Microcalcification detection (as mentioned before)

18
Further improvement in segmentation

• Segmentation by homogeneity and intensity

19
Further improvement in segmentation

• Extract the current region

20
Further improvement in segmentation

• Normalization for the current region and edge
  detection

21
Thanks for your attentionsQuestions if any are
welcome