A New Strategy to Detect Lung Cancer on CT Images

Abstract

Lung cancer has a very low cure rate in the advanced stages, with effective early detection, the survival rate of lung cancer could be highly raised. Detection of lung cancer in the early stages plays a vital role for human health. Computed tomography (CT) images, which provide electronic densities of tissues, are widely applied in radiotherapy planning. The proposed system based on CT technology consists of several steps, such as image acquisition, preprocessing, feature extraction, and classification. In the preprocessing stage, RGB images are converted to grayscale images, the median filter and the Wiener filter are used to uproot noises, Otsu thresholding method is applied to convert CT images free from noise to binary images, and REGIONPROPS function is used to exact body region from binary images. In the feature extraction stage, features, like Contrast, Correlation, Energy, Homogeneity, are extracted through statistic method Gray Level Co-occurrence Matrix (GLCM). In the final stage, extracted features, together with Support Vector Machine (SVM) and Back Propagation NeuralNetwork (BPNN), are used to identify lung cancer from CT images. The performance of the proposed system shows satisfactory results of 96.32% accuracy on SVM and 83.07% accuracy on BPNN respectively.