Malignant Lung Nodule Detection using Deep Learning

Abstract

Lung Carcinoma, commonly known as Lung Cancer is an infectious lung tumour caused by uncontrollable tissue growth in the lungs. Presented here is an approach to detect malignant pulmonary nodules from CT scans using Deep Learning. A preprocessing pipeline was used to mask out the lung regions from the scans. The features were then extracted using a 3D CNN model based on the C3D network architecture. The LIDC-IDRI is the primary dataset used along with a few resources from the LUNA16 grand challenge for the reduction of false-positives. The end product is a model that predicts the coordinates of malignant pulmonary nodules and demarcates the corresponding areas from the CT scans. The final model achieved a sensitivity of 86 percent for detecting malignant Lung Nodules and predicting its malignancy scores.