A Comparison of Methods for Mammogram Registration

1
A Comparison of Methods forMammogram
Registration

• Saskia van Engeland,
• Peter Snoeren,
• Jan Hendriks,
• Nico Karssemeijer

2
Outline

• INTRODUCTION
• METHODS
• RESULTS

3
Introduction

• Mammogram registration is an important technique
• to optimize the display of cases on a
  digital viewing station
• find corresponding regions in temporal pairs of
  mammograms for computer-aided diagnosis
  algorithms

4
METHODS

• Data Set
• Nipple Alignment
• Center of Mass Alignment
• Mutual Information
• Warping

5
Data Set
6
Nipple Alignment

• Manner
• we tested was the alignment of previous and
  current views by registering the nipples
• use the distance to the nipple to find
  corresponding locations of lesions in two
  mammographic views
• Implement
• (1) segmentation into background and breast
  tissue.
• (2) shape of the skin contour, line patterns
  of the glandular tissue, and the geometry of the
  breast were determined and fed into a three-layer
  backpropagation network

7

• Disadvantages
• nipple depends strongly on the correct
  segmentation of the mammogram
• only work when the breast is positioned correctly

8
Center of Mass Alignment

• Manner
• Mammographic views can also be registered by
  aligning the CMs of the breast area
• analogous to the previous method

9

• Disadvantages
• performance of this method will depend strongly
  on segmentation
• MLO views is the varying proportion of the
  pectoral muscle that is visible
• applying the Hough transform on this ROI,
  Finally, the highest peak is selected

10
(No Transcript)
11
Mutual Information

• Manner
• MI gives a measure of the strength of the
  dependence between variables .For two images A
  and B it is defined as
• is their joint entropy
• is the gray value distribution of
  image
• is the amount of information that
  image contains
• about image .
• maximization of (1) by transformation of one of
  the images is equivalent to registering the
  images.

12
(No Transcript)
13

• Implement
• the reference and floating image were segmented
  into background and breast tissue.
• Possible transformations were
• translation
• rotation with respect to image center
• scaling
• vertical shearing.
• This in contrast to the previous two methods,
  which only
• used translation.

14
(No Transcript)
15
Warping

• Manner
• This technique used a set of automatically
  determined control points ( and
  ) located on the breast contour and the
  pectoral muscle
• control points were obtained after segmenting the
  mammograms into breast tissue, pectoral muscle,
  and background
• interpolating between the control points

16
(No Transcript)
17
RESULTS

• Comparing Results From Different Techniques
• MI gtgt CM gtgt nipple alignment gtgtwarping

18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22

• END
• THANKS EVERYONE