Transcript Power Point Template - Intelligent Modelling and Analysis

Methods of Interpretation of a Non-stationary Fuzzy System for
the Treatment of Breast Cancer
Xiao-Ying Wang1, Jonathan M. Garibaldi1, Shang-Ming Zhou2, Robert I. John2
1
The University of Nottingham, Nottingham, UK
2 De Montfort University, Leicester, UK
Speaker: Dr. Xiao-Ying Wang (Sally)
Supervisor: Dr. Jon Garibaldi
FUZZ-IEEE 2009
Outline
Conclusions
Introduction
Nonstationary FS
Experiments
Data
Description
Type-1 FS
FUZZ-IEEE 2009
NS FS Output
Processing
Introduction
• Breast Cancer treatment decision making
• Multidisciplinary team
(oncologist, radiologist, surgeon, pathologist)
• Computational intelligence techniques in
breast cancer diagnosis and decision making
• Uncertain and imprecise terms
• Traditional fuzzy methods (Type-1, Type-2)
• Non-stationary fuzzy sets
Nonstationary FS
(2)
An example of a non-stationary fuzzy set with multiple instantiations
mf (x,2,5,8)
mf ( x, p1 (t ), p2 (t ), p3 (t ))
mf (x,2,5,8)
An example of a non-stationary fuzzy set with multiple instantiations
Nonstationary FS
(3)
Perturbfunctions
f1 (t )  f 2 (t )  f 3 (t )  f (t )
Normaldistribution function  0.02
k1  k 2  k3  k  1
n  20
Nonstationary FS
(1)
An outline of a non-stationary fuzzy inference system
?
Data
Description
(1)
• Breast cancer post operative (adjuvant)
treatment decision data
• From City Hospital Nottingham Breast
Institute (multidisciplinary team)

Attributes + Treatment decisions
(1310 real patients cases)
Data
Description
(2)
• Attributes:
 Patients’ age
 Lymph node stage, the number of positive lymph
node found from samples
 Nottingham prognostic index (NPI) value
-an indication of how successful treatment might be
-NPI = (0.2 x tumour diameter in cms) + lymph node stage + tumour grade
 Estrogen receptor (ER) test result
 Vascular invasion test result
Data
Description
(3)
• Treatment Decisions
Hormone therapy
Radiotherapy
Chemotherapy
Further operation
Follow up
Clinical guideline for adjuvant therapy following surgery
Data
Description
(4)
Type-1 FS
(3)
Fuzzy rules derived directly from the clinical guidelines
Type-1 FS
(1)
Type-1 FS
(2)
No
[0, 55]
Maybe (55,56]
Yes
(56, 100)
Type-1 FS
(4)
• Confusion matrix obtained by the original type-1 fuzzy system
Agreement:
(982+2+124)/1310 = 84.6%
NS FS Output
Processing
(1)
• Type-1 fuzzy system (FS)
Non-stationary FS
• Perturbation function – normal distribution
standard deviation
iteration = 30
  0.01, 0.02,... 0.1
• Output processing methods:
– Existing non-stationary FS output approach
Sum-avg
–
method
–
method
Majority
Ns-avg
Ns-avg
NS FS Output
Processing
(2)
NS FS Output
Processing
(3)
Sum-avg
NS FS Output
Processing
(4)
Majority
Experiments
(1)
The number of agreements obtained over a range of variation for the
three output processing methods
1150
No. of Agreement
1140
Majority
1130
Ns-avg
1120
1110
1100
Sum-avg
1000
0
0.01 0.02 0.03
0.04
0.05 0.06 0.07 0.08 0.09 0.1
Experiments
(2)
The best confusion matrices obtained for the three different methods of
Output Interpretation
Ns-avg
Sum-avg
Majority
Advantage on output of NS FS
• Improvement of accuracy
• Best no. of agreement achieved on sd = 0.08
Conclusions
•
•
•
•
Breast cancer follow up (adjuvant) treatment
Type-1, Type-2, non-stationary FS
Non-stationary FS applies to decision making
Proposed two new ways to interpret NS FS
Output processing.
• Majority method improves the accuracy of a
NS FS
Future work
• Represent variation within FIS
• Variation comparison between FIS and real
clinical experts
• Potential other output processing methods in
NS FS
References
•
•
•
•
•
•
B. Kovalerchuk, E. Triantaphyllou, J. F. Ruiz, and J. Clayton, “Fuzzy logic in computer-aided
breast cancer diagnosis: Analysis of lobulation,” Artificial Intelligence in Medicine, vol. 11, no.
1, pp. 75–85, 1997.
C. A. Pena-Reyes and M. Sipper, “A fuzzy-genetic approach to breast cancer diagnosis,”
Artificial Intelligence in Medicine, vol. 17, pp. 131–135, 1999.
H. A. Abbass, “An evolutionary artificial neural networks approach for breast cancer
diagnosis,” Artificial Intelligence in Medicine, vol. 23, no. 3, pp. 265–181, 2002.
X. Xiong, Y. Kim, Y. Baek, D. W. Rhee, and S.-H. Kim, “Analysis of breast cancer using data
mining and statistical techniques,” in Proceedings of 6th Intelligence Conference on Software
Engineering (SNPD/SWQN’05), Maryland, USA, 2005, pp. 82–87.
S.-M. Zhou, R. I. John, X.-Y. Wang, J. M. Garibaldi, and I. O. Ellis, “Compact fuzzy rules
induction and feature extraction using SVM with particle swarms for breast cancer
treatments,” in Proceedings of the IEEE Congress on Evolutionary Computation (CEC 2008),
Hong Kong, China, 2008, pp. 1469–1475.
J. M. Garibaldi, M. Jaroszewski, and S. Musikasuwan, “Non-stationary fuzzy sets,” IEEE
Transations on Fuzzy Systems, vol. 16 (4), pp. 1072–1086, 2008.
FUZZ-IEEE 2009
Literature
• Fuzzy sets to represent the opinions for radiologists in analysing two
important features from the American College of Radiology Breast Imaging
Lexicon [Kovalerchuk et al 1997]
• Fuzzy-genetic method to Wisconsin BC diagnosis data. Genetic algorithm
was used to generate a fuzzy inference system [Pena-Reyes and Sipper 1999]
• Evolutionary arificial neural network for BC diagnosis [Abbass 2002]
• Data mining for decision trees and association rules to discover
unsuspected relationship within BC data [Xiong 2005]
• Particle swarming optimisation within a support vector machine for
recommending treatments in BC [Zhou et al 2008]
How to process the output of NS FS
Average
What’s NS FS ?
A fuzzy system where the variability is
introduced through the random alterations to
the parameters of the membership functions
over time