Lung Nodule Classification using Shallow CNNs and Deep Transfer Learning CNNs

Lung cancer incidence is more than any other type of cancer. The death rate due to lung cancer can be avoided if it is detected earlier with the help of Computed Tomography (CT) images and classifying them either as benign or malignant by an effective Computer Aided Diagnosis (CAD) System. Deep Learning is gaining traction nowadays in almost all fields of human endeavors. It is beginning to play an inimitable role in detecting lung nodules giving accurate results and thereby reducing the need for human intervention. In this paper, two-layered Convolutional Neural Network (CNN) named as ConvLung was developed to classify lung nodules into benign and malignant types and its performance was compared with the state-of-the-art pretrained CNN architectures. It was observed that deep transfer learning based CNNs such as the Xception network and Inception-ResNet50v2 network can differentiate benign and malignant nodules in a better manner when compared to the ConvLung model.