结合区域检测和注意力机制的胸片自动定位与识别

朱伟,张帅,辛晓燕,李文飞,王骏,张建,王炜

Automatic thoracic disease localization and recognition by combining region proposal network and attention mechanism

Wei Zhu,Shuai Zhang,Xiaoyan Xin,Wenfei Li,Jun Wang,Jian Zhang,Wei Wang

表1 不同模型在ChestX?ray14数据集上的AUC值对比

Table 1 The AUC scores for different models on Chest X?ray14 dataset

Disease Kumar[13]* Guendel[12] Li[20] Liu[14] Ours Atelectasis 0.762 0.767 0.8 0.79 0.822 Cardiomegaly 0.913 0.883 0.87 0.87 0.903 Effusion 0.864 0.828 0.87 0.88 0.9 Infiltrate 0.692 0.709 0.7 0.69 0.741 Mass 0.75 0.821 0.83 0.81 0.862 Nodule 0.666 0.758 0.75 0.73 0.721 Pneumonia 0.715 0.731 0.67 0.75 0.79 Pneumothorax 0.859 0.846 0.87 0.89 0.865 Consolidation 0.784 0.745 0.8 0.79 0.869 Edema 0.888 0.835 0.88 0.91 0.902 Emphysema 0.898 0.895 0.91 0.93 0.867 Fibrosis 0.756 0.818 0.78 0.8 0.921 PT 0.774 0.761 0.79 0.8 0.788 Hernia 0.802 0.896 0.77 0.92 0.851 mean 0.795 0.807 0.81 0.83 0.843