Abstract
Cancer is a cause of premature death in humans. Cancer-related deaths are increasing worldwide over the years. Among all cancers, lung cancer contributes to a major proportion of deaths. The survival rate of patients with lung cancer can, however, be increased to a great extent by timely and efficient diagnosis and treatment. The present study aimed to develop a computer-aided diagnosis system with a novel segmentation method and adoption of radiomic features to classify benign and malignant nodules. Initial pre-processing was performed by wavelets for denoising the image. The novel segmentation method based on fuzzy c-means clustering enhanced by genetic algorithm segments of the lung region in the low-dose computed tomography images. The population size was initialized with cluster centres randomly. The fitness value was calculated by minimizing the individual pixel distance from the cluster centres. The best parents were selected according to the fitness value. The process of crossover and mutation was applied to the parent chromosomes. These steps were repeated for a finite number of times to obtain the optimum solution. Nodules were extracted by morphological operations of retrieving the objects based on their size. Radiomic features were extracted from the images, and these feature sets were reduced to 12 features by applying the least absolute shrinkage and selection operator (LASSO) for feature selection. Benign and malignant nodules in the low dose computed tomography (LDCT) images were classified using the ensemble method of AdaBoost. The sensitivity, specificity, and accuracy achieved by this method were 91.66%, 100%, and 98%, respectively.