Lung Nodule Classificatin And Segmentation

Abstract

One of the top causes of death worldwide, both in developing and underdevel oped nations, is lung cancer, and in particular, lung nodule. The early detection of lung nodules allows patients to have proper treatment before further complications, which improve the survival rate. in computed tomography (CT) images are essential for the diagnosis of lung cancer. However, robust nodule detection has proven to be a difficult issue because of the heterogeneity of the lung nodules and the complex ity of the surrounding area. In this thesis, we examine the impact of using some machine learning approaches for the classification and segmentation of lung nodules from CT images. In the first contribution, we propose a convolution neural network architecture for lung nodule classification. The CNN architecture consist in two convolution, two max-pooling and one soft max layers. This not too deep architec ture grants rapidity in response. For purposes of comparison, we experiment other pretrained CNN models like VGG and RESNET for the same task of classification. For the task of segmentation, the pretrained UNET model has been used for the purpose of image semantic segmentation. The proposed classification and segmen tation methods have been experimentally verified on the (IQ-OTH/NCCD) lung cancer dataset and the Decathlon dataset, respectively. Experimental results show that employing such techniques could immensely help medicine experts in taking decision. This study examines the use of computer-aided diagnosis in the medical field. We recognize that while machine learning (ML) is already being utilized, it is still in its early stages with ongoing developments of more beneficial upgrades.