Lung Cancer Detection and Classification Using Deep Learning

In recent years, so many Computer Aided Diagnosis (CAD) systems are designed for diagnosis of several diseases. Lung cancer detection at early stage has become very important and also very easy with image processing and deep learning techniques. In this study lung patient Computer Tomography (CT) scan images are used to detect and classify the lung nodules and to detect the malignancy level of that nodules. The CT scan images are segmented using U-Net architecture. This paper proposes 3D multipath VGG-like network, which is evaluated on 3D cubes, extracted from Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI), LUng Nodule Analysis 2016 (LUNA16) and Kaggle Data Science Bowl 2017 datasets. Prediction from U-Net and 3D multipath VGG-like network are combined for final results. The lung nodules are classified and malignancy level is detected using this architecture with 95.60% of Accuracy and 0.387732 of logloss.