Lung Cancer Detection using 3D Convolutional Neural Networks

A large number of cancer deaths in the world is due to lung cancer, which is caused due to unbalanced cell growth. In this paper, we used 3D Convolutional Neural Network (CNN) for identification of lung cancer from the Computed Tomography (CT) scans of the patient, since CNN makes it easier to obtain the important information from the images. Here we use the SPIE-AAPM Lung CT Challenge dataset and employ different morphological preprocessing techniques like conversion to Hounsfield Unit, removing the air region and filling the lung area to obtain the lung nodule mask. We utilize our 3D CNN model for lung cancer detection and obtain a very good evaluation of the model. We divide our preprocessed dataset into 60%, 20% and 20% for training, validation and testing respectively, and obtain training accuracy of 83.33%, testing accuracy of 100% and precision, recall, kappa-Score, and F-score of 1.