Studies on Identification of Lung Cancer Cells Using Artificial Neural Networks

Abstract

The computational efficiency of neural network and the resistance of fuzzy sets with newlinetheir proven record in pattern recognition problems have caused a great amount of newlineinterest in the combination of two. The thesis entitled Studies on Identification of newlineLung Cancer Cells using Artificial Neural Networks has resulted in the development newlineof extended fuzzy neural classifier because the above facts have motivated us. The newlinethesis is divided into five chapters. newlineFirst chapter discusses about lung cancer, types and significance of early detection. It newlinealso gives the literature review of existing computer aided diagnosis systems and newlinedetection techniques for pattern classification and recognition. The techniques are newlinemainly based on artificial neural network architectures and fuzzy neural networks. newlineSecond chapter illustrates the existing techniques for feature extraction from the newlinesuspected area of lungs and also focuses on the fuzzy neural approaches. The newlinestatistical texture features and moment invariant features of the suspected area have newlinebeen extracted and fed to the fuzzy neural classifiers. The artificial neural and fuzzy newlineneural are among the approaches commonly used for pattern recognition and newlineclassification in various computational intelligence systems. These are thoroughly newlinediscussed in this chapter. It has also projected on architecture and learning algorithm newlineof Fuzzy Hypersphere Neural Network (FHSNN) classifier. newlineThird chapter has projected on Pruned Fuzzy Hypersphere Neural Network newline(PFHSNN) classifier with its block diagram and architecture. The PFHSNN is an newlineextension of Fuzzy Hypersphere Neural Network (FHSNN) classifier. It utilizes fuzzy newlinesets as pattern classes in which each fuzzy set is a union of fuzzy set hyperspheres. A newlinepruning procedure is incorporated into FHSNN after its leaning phase to reduce the newlinenetwork size. The experimental results show efficiency of PFHSNN classifier over newlineFHSNN classifier in terms of number of hyperspheres created, training time and recall newlinetime. The performance of PFHSNN algorithm is also compared with the FMN and the newlineFHSNN algorithms and it is observed that the PFHSNN yields better recognition rates newlinein comparison with the FMN and the FHSNN algorithms. newlineFourth chapter discusses the general overview and the pruning algorithm of proposed newlinePruned Fuzzy Hypersphere Neural Network (PFHSNN) classifier. During the pruning newlinevi newlinephase of PFHSNN, a confidence factor (CF) of each hypersphere created is calculated by considering its usage index and accuracy index. Then the hyperspheres with low CF are eliminated and only the hyperspheres with high CF are used in testing phase for pattern classification. This process can engage several seconds. So the training time taken by PFHSNN classifier is minutely higher than FHSNN classifier. However, the experimental results explore the different capabilities of PFHSNN over FHSNN classifier. newline