Rough Set Approach for Classification and Retrieval Mammogram

Download Report

Transcript Rough Set Approach for Classification and Retrieval Mammogram

Rough Sets in Hybrid
Intelligent Systems For Breast
Cancer Detection
By
Aboul Ella Hassanien
Cairo University, Faculty of Computer and Information, IT Dept.
email: [email protected]
WRSTA, 13 August, 2006
Outline




Introduction

Digital mammography

Hybrid intelligent systems
Objective
What is Mammogram?
Mammogram Analysis Framework







Hybrid Intelligent System






Pre-processing phase
Segmentation
Feature Extraction phase
Feature Representation phase
Generated Rules phase
Classification phase
Pre-processing Algorithm – Fuzzy Image Processing
Rough Set data analysis
Rough neural Classifier
Evaluation
Results
Conclusion and Future Work
WRSTA, 13 August, 2006
Introduction

According to the National Cancer Institute:





Breast cancer is the leading cause of cancer deaths in women
today and it is the most common type of cancer in women.
Each year about 180,000 women in the United States
develop breast cancer, and
About 48,000 lose their lives to this disease.
It is also reported that a woman's lifetime risk of developing
breast cancer is one in eight.
Currently, digital mammography is one of the most
promising cancer control strategies in earliest
stages.
WRSTA, 13 August, 2006
What is a mammograms?

A mammogram is a
special kind of X-ray
that allows the doctor to
see into the breast
tissue
WRSTA, 13 August, 2006
Introduction

Hybridization of intelligent systems is




A promising research field of modern artificial intelligence concerned with the
development of the next generation of intelligent systems.
A fundamental stimulus to the investigations of Hybrid Intelligent Systems (HIS) is the
awareness in the academic communities that combined and integrated approaches will be
necessary if the remaining tough problems in artificial intelligence are to be solved.
Recently, hybrid intelligent systems are becoming popular due to their capabilities in
handling many real world complex problems, involving imprecision, uncertainty and
vagueness, high-dimensionality.
A hybrid intelligent system is

one that combines at least two intelligent technologies. For example,



Combining a neural network with a fuzzy system results in a hybrid neuro-fuzzy system.
Combining a neural network with a rough system results in a hybrid neuro-rough system. Etc.
The combination of probabilistic reasoning, fuzzy logic, neural networks and
evolutionary computation forms the core of soft computing, an emerging approach
to building hybrid intelligent systems capable of reasoning and learning in an
uncertain and imprecise environment.
WRSTA, 13 August, 2006
The primordial soup
Belief
Networks
Fuzzy Logic
Neural
Networks
Intelligent Systems
Chaos & Fractals
Rough Sets
WRSTA, 13 August, 2006
Evolutionary
Algorithms
Different methods = different roles
Fuzzy Logic : the algorithms for dealing with
imprecision and uncertainty
Neural Networks : the machinery for learning
and function approximation with noise
Evolutionary Algorithms : the algorithms for
reinforced search and optimization
uncertainty
arising
from
the
Rough
RS
granularity
in
the
domain
of
Sets
discourse
WRSTA, 13 August, 2006
Comparison of Expert Systems, Fuzzy Systems,
Neural Networks and Genetic Algorithms
ES
FS
Knowledge representation

Uncertainty tolerance






Imprecision tolerance
Adaptability
Learning ability
Explanation ability
Knowledge discovery and data mining
NN
GA











Maintainability

* The terms used for grading are:
- bad, - rather bad,  - rather good and
 - good
WRSTA, 13 August, 2006

Objective

Introduce a rough neural intelligent approach for:


Rule generation and image classification.
An application of breast cancer imaging has been
chosen and hybridization of intelligent computing
techniques has been applied to see their ability and
accuracy to classify the breast cancer images into two
outcomes:


malignant cancer or benign cancer.
Computer-based to assist radiologists in
mammography classification of breast cancer
images (Computer Aided Diagnosis System)
WRSTA, 13 August, 2006
Mammogram Analysis Framework
WRSTA, 13 August, 2006
Mammogram Analysis Framework

Pre-processing phase – Fuzzy theory




Feature Extraction phase





Statistical features – concurrence Matrix
Rough Sets Data Analysis


Enhancement
Segmentation: Region of Interest (ROI)
Region Boundary Enhancement
Feature representation – Rough information system
Reduct generation
Rule generation
Classification phase
 Rough neural classifier
Evaluation
WRSTA, 13 August, 2006
Pre-Processing – Fuzzy theory

Mammograms are images that are difficult to
interpret; therefore, techniques are needed to:



Enhance the quality of these images for a better
interpretation.
For this purpose, a pre-processing phase of the images is
adopted to improve the quality of the images and to make
the feature extraction phase more reliable.
It contains several processes;

to enhance the contrast of the whole image;


to extract the region of interest;


Fuzzy histogram hyperbolization algorithm (FHH)
Modified Fuzzy c-mean clustering algorithm
to enhance the edges surrounding the region of interest.

Fuzzy histogram hyperbolization algorithm (FHH)
WRSTA, 13 August, 2006
Feature Extraction

Once the pre-processing was completed,
features relevant to region of interest
classification are extracted, normalized and
represented in a database as vector values

Gray level co-occurrence matrix (GLCM)

Energy, entropy, contrast and inverse difference moment.
WRSTA, 13 August, 2006
Rough Sets Data Analysis





Create decision table
Compute some reduct
with minimal number of
attributes.
Significance of
attributes: calculate the
weight of the attributes.
Rule Generation
Rule Evaluation
WRSTA, 13 August, 2006
Rough neural network: rough neuron
WRSTA, 13 August, 2006
Results (Enhancement)
WRSTA, 13 August, 2006
Results (Segmentation)
WRSTA, 13 August, 2006
WRSTA, 13 August, 2006
Average Execution time
WRSTA, 13 August, 2006
Number of generated rules and
classification accuracy
WRSTA, 13 August, 2006
Conclusion

Introducing a hybrid scheme that combines the advantages of different
soft computing techniques for breast cancer detection.

Fuzzy sets is used as a pre-processing techniques to








enhance the contrast of the whole image; to
extracts the region of interest and then to
enhance the edges surrounding the region of interest.
Then, subsequently extract features from the extracted regions
characterizing the underlying texture of the interested regions.
Feature extractions acquired in this work are derived from the gray-level cooccurrence matrix.
A rough set approach to attribute reduction and rule generation has been
used.
Rough neural networks were designed for discrimination for different regions
of interest to test whether they are cancer or nun-cancer.
The results proved that the soft computing techniques are very
successful and has high detection accuracy.
WRSTA, 13 August, 2006