Multi-Layer Perceptron Based Lung Tumor Classification

Automated tumor delineation is complex and an interesting area of research in medical imaging. The tumor delineation in oncology needs to be precise for treatment planning and delivery. Early detection of the lung cancer is possible using Computed Tomography (CT) Images. In this approach, the CT scan images process through lung volume segmentation, lung nodule segmentation, feature extraction and classification. The lung volumes are segmented by thresholding method while lung nodules were segmented using k-means clustering. The features were extracted using Gray Level Co-occurrence Matrix (GLCM). The texture features like contrast, dissimilarity, Homogeneity, Angular Second Moment (ASM), Energy, Mean and Standard deviation were extracted to represent a non-tumorous and tumorous image. Finally, the features are trained and tested by using K-Nearest Neighbor (KNN) and Multi-Layer Perceptron (MLP) algorithm. The proposed approach achieved the classification accuracy of 98.30% using KNN and 98.31 % using MLP. The experiments were performed using the NSCLC, LIDC datasets.