基于深度学习的哮喘患者CT影像黏液栓自动识别

针对肺部计算机断层扫描（CT）影像中人工识别黏液栓效率较低、识别效果不佳等问题，提出一种基于深度神经网络的黏液栓自动识别模型。针对黏液栓不规则的特点，在骨干网络中引入可变形卷积来提取特征，并在检测网络中引入可变形感兴趣区域池化进行特征尺度归一化。针对黏液栓的中小目标特性，提出采用加权特征金字塔网络进行多尺度特征融合。实验结果表明，与传统的更快区域卷积神经网络相比，所提模型的平均精度提升了4%，可为辅助诊断哮喘的严重程度提供参考。

Automatic Recognition of Mucus Impaction in CT Images of Asthmatic Patients Using Deep Learning

In view of the low efficiency of manual identification of mucus impaction in chest computed tomography (CT) and the poor recognition effect, a deep neural network based automatic recognition algorithm for mucus impaction is proposed. In order to deal with the irregular characteristics of mucus impaction, deformable convolution is added to the backbone to extract features, and deformable region of interest pooling is used in the detection network to normalize the feature scale. Besides, feature pyramid network with weight coefficient is used for multi-scale fusion according to the characteristics of small and medium objects. The results show that compared with the traditional faster region convolutional neural network, mean average precision of the proposed algorithm is improved by 4%, which can provide auxiliary reference for the diagnosis of asthma severity.