Lung Segmentation on HRCT and Volumetric CT for Diffuse Interstitial Lung Disease Using Deep Convolutional Neural Networks

We developed a robust lung segmentation method using a deep convolutional neural network (CNN) for high-resolution computed tomography (HRCT) and volumetric CT images of diffuse interstitial lung disease (DILD). Using scans of 617 patients with DILD types such as cryptogenic organizing pneumonia, usual interstitial pneumonia, and nonspecific interstitial pneumonia, we created gold standards with manually corrected segmentation by an expert thoracic radiologist. We then used a two-dimensional U-Net architecture with deep CNN for lung region segmentation in HRCT images. The U-Net-based segmentation method was found to be highly accurate, showing significant improvement over conventional methods (p < 0.001) for both HRCT and volumetric CT. The insights and methods of this study have been comercialized in Coreline's AVIEW platform, thereby promising its effective use for different clinical protocols.