基于Transformer网络的气胸图像分类方法

Pneumothorax image classification method based on Transformer network

气胸是一种致命的疾病,借助胸部X光片的自动辅助诊断系统可以减少诊断时间并节省宝贵的治疗时间。提出了一种基于Transformer网络的气胸图像分类方法,针对气胸图像样本少,采用了迁移学习的方法初始化网络权重,为了提高图片的质量,使用直方图均衡化方法增强图像质量,提高网络分类的准确率。结果表明,所提方法的分类准确率达到了90.1%,与卷积网络DenseNet-121相比,准确率提高了3.6%,可以推广到临床的辅助诊断当中。

Pneumothorax is a deadly disease,and automated diagnostic aids based on chest X-rays can reduce diagnosis time and save valuable treatment time.A classification method of pneumothorax images based on Transformer network is proposed.In view of the few samples of pneumothorax images,the transfer learning method is used to initialize the network weights.To improve the quality of the images,a histogram equalization method is used to enhance the image quality and improve the accuracy of the network classification.The results show that the classification accuracy of the proposed method reaches 90.1%,which is increased by 3.6%compared with the convolutional network DenseNet-121.It can be extended to clinical auxiliary diagnosis.