FUZZ-IEEE 2009 Methods of Interpretation of a Non-stationary Fuzzy System for the Treatment of Breast Cancer Xiao-Ying Wang 1, Jonathan M. Garibaldi 1, Shang-Ming Zhou 2, Robert I. John 2 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 Introduction Data Description Non- stationary FS Type-1 FS NS FS Output Processing Experiments Conclusions
Breast Cancer treatment decision making Multidisciplinary team (oncologist, radiologist, surgeon, pathologist) Computational intelligence techniques in breast cancer diagnosis and decision making Computational intelligence techniques Uncertain and imprecise terms Traditional fuzzy methods (Type-1, Type-2) Non-stationary fuzzy sets Introduction
Non- stationary FS (2)
Non- stationary FS (3)
Non- stationary FS (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 (1)
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 (2)
Treatment Decisions Hormone therapy Radiotherapy Chemotherapy Further operation Follow up Data Description (3)
Data Description (4)
Type-1 FS (3)
Type-1 FS (1)
No [0, 55] Maybe (55,56] Yes (56, 100) Type-1 FS (2)
Confusion matrix obtained by the original type-1 fuzzy system Type-1 FS (4) Agreement: ( )/1310 = 84.6%
Type-1 fuzzy system (FS) Non-stationary FS Perturbation function – normal distribution standard deviation iteration = 30 Output processing methods: – Existing non-stationary FS output approach – method NS FS Output Processing (1) Majority Sum-avg Ns-avg
NS FS Output Processing (2)
(3) Sum-avg
NS FS Output Processing (4) Majority
Experiments (1) Majority Sum-avg Ns-avg No. of Agreement
Experiments (2) Ns-avg Sum-avg Majority
Improvement of accuracy Best no. of agreement achieved on sd = 0.08
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 Conclusions
Represent variation within FIS Variation comparison between FIS and real clinical experts Potential other output processing methods in NS FS Future work
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, C. A. Pena-Reyes and M. Sipper, “A fuzzy-genetic approach to breast cancer diagnosis,” Artificial Intelligence in Medicine, vol. 17, pp. 131–135, H. A. Abbass, “An evolutionary artificial neural networks approach for breast cancer diagnosis,” Artificial Intelligence in Medicine, vol. 23, no. 3, pp. 265–181, 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
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 ]
Average
A fuzzy system where the variability is introduced through the random alterations to the parameters of the membership functions over time