Lung segmentation in CT images using a fully convolutional neural network with multi-instance and conditional adversary loss

Lung segmentation is usually the first step of lung CT image analysis and plays an important role in lung disease diagnosis. We propose an efficient end-to-end fully convolutional neural network to segment lungs with different diseases in CT images. We introduce a multi-instance loss and a conditional adversary loss to the neural network in order to solve the segmentation problem for more severe pathological conditions. Our method is capable of solving the lung segmentation problem under normal, moderate and severe pathological conditions, which is validated on 3 public benchmark datasets with different diseases.