Significance of haralick features in bone tumor classification using support vector machine

Accurate classification of bone lesions into benign and malignant tumors plays a key role in determining the treatment course (surgical intervention or radiation), an essential part of radiologists work. In this work, we investigate the significance of Haralick textural feature components in an application to computer-aided classification of the bone tumor into benign and malignant. The Haralick features from Computed Tomography (CT) images are extracted to form the dataset. Support Vector Machine (SVM) with medium Gaussian kernel function is trained and tested with the dataset consisting of various malignant and benign tumors. The dataset prepared by extracting the Haralick features are subjected to Correlation-Based Feature Subset (CFS) Selection. The accuracy of the classifier is measured in each case. The study reveals an increased accuracy post-feature selection.