Performance Analysis of KNN and SVM in DiscriminatingPulmonary Nodules

Abstract

Identification of malignant pulmonary nodules plays a crucial role in the diagnosis of lung cancer in the early stage which in turn improves the chances of survival of an individual. Timely detection and treatment are essential to eradicate cancerous lung nodules from the body. Computer-Aided Detection (CAD) system is used to detect the nodules from the thoracic region in the CT image. The lung CT images are taken from LIDC-IDRI database. The input image is enhanced using contrast enhancement. Adaptive thresholding and active contour model without edges are done for segmenting lung parenchyma. Morphological operations are performed using a disk-shaped structured element. Masking process is done for detecting the candidate nodules. Textural features are taken from all the detected candidate nodules. To analyze whether it is a benign nodule or malignant nodule, K-Nearest Neighbor (KNN)and Support Vector Machine (SVM) classifiers are employed. The performance metrics of these two classifiers are analyzed in the classification of benign and malignant pulmonary nodules.  In this study, we analyzed different types of KNN and different types of SVM in pulmonary nodules classification. From the experimental work, we interpret that quadratic SVM outperforms medium KNN with better AUC of 0.82 in the pulmonary nodule classification task.