Automatic Pulmonary Lobe Segmentation Using Deep Learning

• DOI: 10.48550/arxiv.1903.09879
• Publication Date: 2019-01-01
• Authors: Hao Tang; Chupeng Zhang; Xiaohui Xie
• Keywords: Robustness; Ground truth

Pulmonary lobe segmentation is an important task for pulmonary disease related Computer Aided Diagnosis systems (CADs). Classical methods rely on successful detection of fissures and other anatomical information such as the location blood vessels airways. With success deep learning in recent years, Deep Convolutional Neural Network (DCNN) has been widely applied to analyze medical images like Computed Tomography (CT) Magnetic Resonance Imaging (MRI), which, however, requires a large number ground truth annotations. In this work, we release our manually labeled 50 CT scans which are randomly chosen from LUNA16 dataset explore use task. We propose pre-processing image by cropping region that covered convex hull lungs order mitigate influence noise outside lungs. Moreover, design hybrid loss function with dice tackle extreme class imbalance issue focal force model focus voxels hard be discriminated. To validate robustness performance proposed framework trained small training examples, further tested independent dataset. Experimental results show approach, consistently improves across different datasets maximum $5.87\%$ compared baseline model.