Clustering based lung lobe segmentation and optimization based lung cancer classification using CT images

Abstract

A serious condition with a high death and morbidity rate worldwide is lung cancer. To increase the likelihood that a person will survive, early diagnosis is urgently required. Various existing approaches are modeled to detect lung cancer, but low visibility of tumor region and the negative rates of images still result a complex task to recognize infected regions. Also, the traditional methods failed to enhance the accuracy of the lung cancer classification. Hence, this research developed a method named Shuffled Social Sky Optimizer-based Multi-Object Rectified Attention Network (SSSO-based MORAN) to effectively classify lung cancer disease. The proposed SSSO algorithm is the integration of the Shuffled Shepherd Optimization Algorithm (SSOA) and social ski-driver (SSD) algorithm, respectively. The input computed tomography (CT) image is supplied to the pre-processing phase, where the Gaussian filtering is employed to pre-process the image, and thereby the Region of Interest (ROI) is acquired from the input image. Then, the lung lobes are segmented using the proposed Deep Renyi entropy fuzzy clustering (DREFC). With the segmented lung lobes, the nodule region is identified from the lung image, and the process of cancer classification is done based on features. The features considered for the lung cancer classification are Local Gabor XOR Pattern (LGXP), Gray-Level Co-occurrence Matrix (GLCM) features, Global Binary Pattern (GBP), Tetrolet transform, and statistical features. The proposed algorithm effectively showed higher performance of accuracy, Mean Absolute Error (MAE), sensitivity, and specificity of 0.896, 0.104, 0.8969, and 0.845, respectively.