Research And Implementation Of Medical Image Annotation System For Deep Learning

In recent years,deep learning has been applied to a variety of medical image analysis problems,which has brought a huge breakthrough for the intelligent auxiliary diagnosis of medical image.However,in the deep learning medical image analysis,there is still a lack of high-quality annotation data set,most of the research process needs to build annotation data set first.After the preliminary investigation and analysis,there are the following problems in the construction of high-quality annotation dataset:(1)The quality control of this process lacks the standard framework,and the public image annotation tools lack the quality control function of the annotation process,researchers need to conduct offline quality review,which is not conducive to the accuracy of the annotation dataset.(2)At present,most of the public annotation tools only provide manual annotation,which is inefficient and time-consuming.Some tools provide a scheme of algorithm assisted annotation after a large number of data training,but the diversity of medical research determines that the pre training model after a large number of data training is difficult to obtain,and this scheme is not suitable for medical images.(3)At present,the public image annotation tools mainly provide annotation functions,lack of necessary management for personnel,data sets,etc.,lack of support for team cooperation to complete annotation tasks,are not enough to support large-scale medical image annotation work.In view of the above problems,the main research contents of this paper are as follows:(1)Based on the research of the requirements affecting the quality of medical image annotation data,combined with the process design of medical image data set construction,the quality control framework of medical image data set construction is realized.In view of the quality control in the process of marking,the idea of group intelligence is introduced to design and implement the quality control method of multi person cooperative marking.(2)A semi-automatic annotation system based on deep learning is designed and implemented to improve the efficiency of manual annotation.The system adapts to the needs of pre annotation in the medical field.After obtaining a small amount of annotation data,the pre annotation model is trained and can achieve high precision.Based on the subsequent annotation data,the model is updated iteratively to improve the pre annotation effect.(3)A medical image annotation system is built.Based on the realization of the common format and DICOM format medical image annotation function,it provides team collaboration annotation support and comprehensive and efficient management function,provides quality control function in the process of annotation,and provides semi-automatic annotation function of algorithm.This paper studies and implements a deep learning oriented medical image annotation system.The system has been deployed in ophthalmic hospital to support ophthalmologists’ annotation,which provides a strong support for high-quality medical image annotation data acquisition.