Lung Nodule Segmentation Using UNet
Subham Kumar, P. Malin Bruntha, S Isaac Daniel, J. Ajay Kirubakar, R. Kiruba, Siril Sam, S. Immanuel Alex Pandian
Abstract
Cancer is a major public stumbling block in health worldwide. The highest morbidity and mortality among both men and women were found due to lung cancer. Segmentation of the lung nodule plays a vital role in the treatment of lung cancer. This paper aims to segment such lung nodules using the Computed Tomography (CT) images. U-Net is a modified Convolutional Neural Network made for segmentation of biomedical images. We use U-Net for segmenting lung nodules and along with this, we use various optimizers to find the best result. The dataset is collected from LIDC-IDRI and these CT images are further pre-processed to obtain nodule images and the masks. The pre-processed images are trained using U-Net architecture and the performance of the model is calculated. Using Adam optimizer, the accuracy of 0.9871, the Dice Similarity Coefficient (DSC) of about 0.8205 and the Intersection over Union (IOU) of about 0.7539 are obtained from the proposed computer-aided detection (CAD) system for its segmentation performance. The results of this experiment shows U-Net can be used in lung nodule segmentation.