Indian Journal of Science and Technology

Abstract

Objectives: To develop an optimum hybrid approach for lung cancer detection by multiple feature subset extraction and selection based on SVM-weights and Genetic Algorithm (GA-NN) in order to improve the performance measures such as accuracy, sensitivity and specificity. Methods: Initially in preprocessing phase, Computed Tomography (CT) lung images are de-noised using median filter and enhanced using contrast stretching. In the next phase, candidate patch extraction is formed and Gray Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP) features are extracted. This is followed by feature selection using Genetic Algorithm-Neural Network (GA-NN) with SVM weights. Finally, images are classified as cancerous and non-cancerous using multiple classifiers (SVM and KNN). For this research work, CT lung images are collected form LIDC dataset. Around 500 images are used out of which 70% is used for training and 30% is used for testing. Findings: From simulation results and comparative analysis, it is observed that GANN with SVM weights result in better predictive performance metrics with notable improvements. The suggested feature subset reduction outperforms current techniques for detection of lung cancer in CT images. The proposed method has resulted in improved accuracy, specificity and sensitivity by 95.8%, 91.3% and 93.5% respectively which is higher than the existing approaches. Novelty: This work presents a novel approach to detect the lung cancer by multiple feature subset extraction and selection based on SVM-Weights and Genetic Algorithm - Neural Network (GA-NN) with improved accuracy, sensitivity and specificity.

Keywords: Gray Level Cooccurrence Matrix (GLCM); Genetic AlgorithmNeural Network (GANN); KNearest Neighbor (KNN); Local Binary Pattern (LBP); Support Vector Machine (SVM)