Fig. 1

Pipeline of our nodule classification method. Three dimensional patches of nodules and non-nodules are extracted and pre-processing is conducted to balance the ratio of nodules to non-nodules. A three dimensional deep convolutional neural network (3D DCNN) with shortcut layer connections and a 3D DCNN with dense layer connections are trained on the prepared dataset for nodule classification. Finally, the checkpoint ensemble method is applied to boost performance of our nodule classification method