Segmentation of 3d medical images for detection and classification of lung tumor using content-based features

Lung cancer is one of the most fatal types of lung disease, in which early detection of this cancer can prevent its dangerous consequences. This paper presents a method for the detection and classification of lung tumors based on three-dimensional (3D) images of the TCIA dataset. The proposed algorithm has been formed by 2 main steps, segmentation, and classification. In the first level, Histogram Equalization has been chosen to adjust image intensity, then MSER and SURF are applied to select the 2D slices, containing tumors. Besides, the fuzzy system and k-means algorithm are used to segment the selected 2D slices, resulting in a unique 3D segmented model. Finally, SVM is implemented for tumor classification, using the GLCM-HOG features. The most significant item of the method is that, unlike the 2D methods, this scheme provides the depth of tumors based on the capabilities of 3D space. Furthermore, it has less computational complexity and subsequently takes less time than the deep learning-based method. Experimental results demonstrate the superior performance of the proposed method compared to the new algorithms, with DCS = 0.99 ± 0.008, accuracy = 91.67%, recall = 100%, and precision = 85.71%.