基于卷积神经网络的X线胸片疾病分类研究

利用人工智能技术可以有效的辅助胸片的诊断。通过对中文X 线胸片检查报告文本挖掘,提出一种以基于异常部位的疾病分类标注方法,并整理出一个基于中文X 线胸片检查报告的疾病分类标注数据集。使用AlexNet ,VGGNet ,ResNet 及DenseNet 4 种不同的卷积神经网络,以及直接训练,ImageNet 预训练和ChestX-14 预训练3 种不同预训练方式对胸部疾病分类评估。结果显示更为复杂以及参数更多的卷积神经网络模型,在X 线胸片图像关键信息的获取方面有着更大的优势,同时采用大型X 线胸片数据集ChestX-14 预训练的模型效果要明显优于其他预训练。

Classification of Chest X-ray Disease Based on Convolutional Neural Network

The artificial intelligence technology can effectively assist the chest X-ray diagnosis. On the basis of the analysis of Chinese reports of chest X-rays, a labeling method of the thoracic disease classification for the chest abnormal parts is proposed and a dataset of the thoracic disease classification labels is complied. The thoracic disease classification is evaluated through four kinds of convolutional neural networks, AlexNet, VGGNet, ResNet and DenseNet and through three kinds of training methods, direct training, ImageNet pre-training and Chest X-14 pre-training. The result shows that the more complicated convolutional neural network with the more parameters, the better performance in obtaining the key information from the chest X-ray images can be. The model pre-trained by Chest X-14, a large dataset of chest X-ray, has the better result than the other pre-training methods.