Chest X-ray Image Annotation Based On Deep Learning

In recent years,with the development of medical imaging technology and the widely used of all kinds of medical imaging equipments,medical imaging has become an essential basis for medical diagnosis and greatly enhance the speed and accuracy of the doctor.In these imaging technologies,x-ray imaging has become one of the most common medical imaging equipments because of its fast speed,low cost and high accuracy.However,a large number of medical images are convenient in diagnostic but also increase the pressure of the doctor’s work.At the same time,with the ever-changing of computer science technology,a variety of advanced artificial intelligence algorithms are gradually applied in many fields and play a huge role.Therefore,the rapid annotation of medical images using computer has become an important research direction in computer-aided diagnosis.Deep learning(DL)is one of artificial intelligence algorithms rising in recent years.Compared with the traditional machine learning algorithm,DL can automatically extract low-level to high-level features by training on a large amount of data and solve the problem of over-reliance on manual feature selection.So far deep learning is widely used in computer vision and natural language processing and has achieved outstanding results.Therefore we studied deep convolutional neural networks and deep recurrent neural networks and apply them to automatic recognition of chest X-ray images and automatic generation of medical image reports.The main work of this paper includes the following aspects: firstly,we introduced the development and application of X-ray imaging technology and clarified the importance of automatic annotation of medical images by computer;Then we introduced the development and actuality of deep learning and other commonly used neural networks,and their process details;After that,we achieved a single label convolutional neural network training and a multi-label convolutional neural network fine-tuning on chest X-ray images based on Caffe;Finally,we completed the training of recurrent neural network and generated medical image annotation with input of medical image report and image feature vector extracted from the last fc layer of onvolutional neural network.