Feature Extraction and Classification of Automatically Segmented Lung Lesion Using Improved Toboggan Algorithm

The accurate detection of lung lesions from computed tomography (CT) scans is essential for clinical diagnosis. It provides valuable information for treatment of lung cancer. However, the process is exigent to achieve a fully automatic lesion detection. Here, a novel segmentation algorithm is proposed, it’s an improved toboggan algorithm with a three-step framework, which includes automatic seed point selection, multi-constraints lesion extraction and the lesion refinement. Then, the features like local binary pattern (LBP), wavelet, contourlet, grey level co-occurence matrix (GLCM) are applied to each region of interest of the segmented lung lesion image to extract the texture features such as contrast, homogeneity, energy, entropy and statistical extraction like mean, variance, standard deviation, convolution of modulated and normal frequencies. Finally, support vector machine (SVM) and K-nearest neighbour (KNN) classifiers are applied to classify the abnormal region based on the performance of the extracted features and their performance is been compared. The accuracy of 97.8% is been obtained by using SVM classifier when compared to KNN classifier. This approach does not require any human interaction for lesion detection. Thus, the improved toboggan algorithm can achieve precise lung lesion segmentation in CT images. The features extracted also helps to classify the lesion region of lungs efficiently.