Detecting Lung Nodules Based on Deep Learning with Automatic Segmentation

In this paper an efficient solution for detecting nodules in two-dimensional CT images is proposed. The method proposed here is based on the automatic segmentation of the lung parenchyma region, followed by the detection of possible nodules internal to this region. It is performed with the training of a neural network of the U-net type to perform the segmentation. The identification of the nodes is then performed using a convolutional neural network (CNN). The main contribution of this work resides in the node detection step, defined by the use of a CNN aggregated to the dimensionality reduction of data using the Principal Components Analysis method (PCA). Thus, with the objective of reducing the redundancy of the features extracted by CNN, the PCA was applied in the detection of possible nodes.