Segmentation of lung cancer using deep learning

Abstract

Lung cancer is one among the deadliest and dangerous widespread diseases that create a major public health problem. The main aim of this paper is to basically segment the image or to identify the nodule present in the image and provide the accuracy of that segmented image. In this concern, proper segmentation of lung tumor present in the X-ray scans or Magnetic Resonance Imaging (MRI) or Computed Tomography (CT scan) is the first stone towards achieving completely automated diagnosis system for lung cancer detection of the patient. With the advanced technology and availability of dataset, the time required for a radiologist can be saved using CAD tools for tumor segmentation. In this work, we use an approach called data driven for lung tumor segmentation from CT scans by using UNet . In our approach we will train the network by using CT image with tumor having the slices of size (512 × 512 × 1). Our model has been trained and tested on the LUNA16 dataset considering 10 patients, provided by or used by Lung image database consortium (LIDC) and the image database resources initiative (IDRI). In this dataset, our proposed technique will achieve an average dice score of 0.8507. This can further be analyzed or used for other medical images to find the nodule or with other applications such as in brain image segmentation and liver image segmentation.