用于肺结节检测和分类的两阶段深度学习方法

Two-Stage Deep Learning Method for Detection and Classification of Pulmonary Nodules

针对肺部CT数据具有空间信息的特点,提出一种基于深度学习的两阶段方法,即使用两个3D卷积网络有效学习结节特征,对CT图像中的肺结节进行检测和分类。该方法的检测器部分采用基于U-Net的编码器-解码器结构的3D语义分割模型,以预测结节的位置、大小和语义掩码;分类器部分采用3D双路径网络,用于特征的汇总和收缩,并给出分类结果。为充分利用原始数据中的特征信息,将检测器的结果应用于对原始数据进行采样和掩码操作,并通过空间金字塔池化层获得一致的输入尺度。在公开数据集上的实验结果表明,该深度学习方法对CT图像肺结节的检测和分类具有良好的性能。

Aiming at the characteristics of lung CT data with spatial information,we proposed a two-stage method based on deep learning for the detection and classification of pulmonary nodules in CT images.Two 3D convolution networks were used to effectively learn nodule features.For the detector part of the method,a 3D semantic segmentation model was designed for nodule detection with a U-Net-like encoder-decoder structure to predict the location,size and semantic mask of the nodules.For the classifier part,a 3D dual path network was used to summarize and contract features,and the classification results were given.In order to fully utilize the feature in the original data,the results of the detector were used to sample and mask the original data,and a spatial pyramid pooling layer was added before the classifier to obtain a consistent input scale.The experimental results on public datasets show that the proposed deep learning approach has good performance in the detection and classification of pulmonary nodules in CT images.