Enda Molloy, Electronic Eng.

1
Automated Image Analysis Techniques for Screening
of Mammography Images

• Enda Molloy, Electronic Eng.
• Progress Presentation, 22/01/09.

2
Outline

• Project Overview
• Current Progress
• Future Plans

3
Project Overview

• The project aims to investigate analysis
  techniques for the screening of mammography
  images, which may be used in automated screening
  of a large set of images.
• This will be achieved by developing a system
  comprising of feature extraction and a
  classification architecture.
• Provide functionality for remote access to the
  data via a web browser.

4
Contrast Enhancement

• Contrast Limited Adaptive Histogram Equalisation

5
Image Segmentation

• Global Thresholding

6
Image de-noising

• Often Mammograms can be affected by Gaussian
  noise. Although the images in the MIAS database
  are not affected, noise is added to the images to
  simulate the effect.
• Wavelet Analysis is used to remove the noise
• Wavelet type and number of levels for
  decomposition are selected, then the FWT of noisy
  image is computed.
• A threshold is applied to the detail
  coefficients.
• Wavelet reconstruction is performed to produce
  the de-noised image.

7
Neural Networks

• An Artificial Neural Network is being used as a
  classification architecture for screening regions
  of interest.
• A Multilayer Perceptron is currently being
  implemented using ROI textural statistics as
  inputs to the input layer.
• The output signal will indicate the appropriate
  class for the input data i.e. Benign, malignant,
  normal.

8
MLP Overview
Mean
Standard Deviation
Benign
Third Moment
Malignant
Uniformity
Normal
Entropy
Kurtosis
Input Layer
Hidden Layer
Output Layer
9
Online Database

• MySQL database, with a table holding usernames
  and passwords of registered users and a second
  table holding image information. The images are
  uploaded directly to the server with the filename
  stored in the database.

10
Future Plans

• Jan 22th Jan 30th
• Continuing with work on MLP i.e. Training and
  testing it.
• Jan 30th Feb 6th
• Complete the online database.
• Feb 6th Feb 27th
• Research and implement a second classification
  architecture.

11
Questions