Abstract
Detection of pulmonary nodules in chest computed tomography scans play an important role in the early diagnosis of lung cancer. A simple yet effective computer-aided detection system is developed to distinguish pulmonary nodules in chest CT scans. The proposed system includes feature extraction, normalization, selection and classification steps. One hundred forty-nine gray level statistical features are extracted from selected regions of interest. A min-max normalization method is used followed by sequential forward feature selection technique with logistic regression model used as criterion function that selected an optimal set of five features for classification. The classification step was done using nearest neighbor and support vector machine (SVM) classifiers with separate training and testing sets. Several measures to evaluate the system performance were used including the area under ROC curve (AUC), sensitivity, specificity, precision, accuracy, F1 score and Cohen-k factor. Excellent performance with high sensitivity and specificity is reported using data from two reference datasets as compared to previous work.