DenseNet-centercrop:一个用于肺结节分类的卷积网络

为解决由肺部CT图像对肺结节进行良恶性分类的问题,提出了一个新颖的端到端深度学习网络DenseNet-centercrop。通过在原有的DenseNet结构中的稠密块间增加新的分支,引入了中心剪裁操作。该网络结构具有2个优势:(1)不仅最大程度保留了DenseNet的结构,而且将其稠密连接机制扩展到了稠密块水平,大大丰富了肺结节的多尺度特征。(2)参数量较少,是一种轻量化的网络结构。将基于该网络的肺结节良恶性分类方法在LIDC-IDRI数据集上进行评估,实验结果表明,DenseNet-centercrop极大地提高了DenseNet的性能,较现有的其他肺结节良恶性分类方法具有更高的AUC分值和分类精度。

DenseNet-centercrop: A novel convolutional network for lung nodule classification

To solve the problem of benign or malignant diagnosis of pulmonary nodule with original thoracic computed tomography (CT) images, this paper presents a novel end-to-end deep learning architecture named DenseNet-centercrop. DenseNet-centercrop has two compelling advantages: (1) DenseNet-centercrop preserves the structure of DenseNet at utmost, further reinforces the densely connected mechanism to a level of dense blocks and enriches multi-scale features of lung nodules. (2) It is a lightweight structure with small scale of parameters. We evaluate DenseNet-centercrop on LIDC-IDRI benchmark. Experimental results show that DenseNet-centercrop not only largely boosts the performance of DenseNet, it also has higher accuracy and AUC score on the task of lung nodule classification in comparison with state-of-the-art approaches.