Deep 3D multi-scale dual path network for automatic lung nodule classification

Lung cancer is the cancer with the highest mortality rate in the USA. Computed tomography (CT) scans for early diagnosis of pulmonary nodules can detect lung cancer in time. To overcome the limitations of the segmentation and handcrafted features required by traditional methods, we take deep neural network to diagnose lung cancer. In this work, we propose a deep end-to-end 3D multi-scale network based on dual path architecture (3D MS-DPN) for lung nodule classification. The 3D MS-DPN model incorporates the dual path architecture to reduce the complexity and improve the accuracy of the model fully considering the 3D nature of CT scan while performing 3D convolution. Meanwhile, the multi-scale feature fusion is used to eliminate the effects which the size of lung nodules varied widely and nodules occupying few regions and slices in CT scan. Our model achieves competitive performance on the LIDC-IDRI dataset compared to the recent related works.