Multi-Class SVM Prediction Model for Lung Cancer Diagnosis

Abstract

Detection of lung abnormalities is essential in the field of medical diagnostics in order to reduce mortality. Early and correct diagnosis of these lung abnormalities leads to timely and appropriate treatment thereby reducing the mortality. This area of research always faces a challenging task to differentiate the cancerous tissues from non-cancerous tissues using CT images. This paper presents the SVM prediction model for the characterization of lung tissues namely fibrosis, suspicious of TB and carcinoma. This model is designed with three set of features formed with Gray-Level Co-Occurrence Matrix (GLCM) and Gray-Level Run- Length Matrices (GLRLM) and is evaluated using True Positive Rate, False Negative Rate, Positive Predictive Value, False Discovery Rate. The classifier model shows trustworthy results with classification accuracy ranging between 93.9% and 98.6% respectively.