Research On Medical Image Database Construction Towards Template Image Learning

Facing the massive medical imaging resources in the era of medical big data,how to carry out effective storage management and further knowledge mining is an urgent problem to be solved.Traditional management methods focus on data storage and retrieval,ignoring the mining of rich anatomical and physiological information.On the other hand,in the field of computer-aided analysis of medical images,the construction of template images for specific populations requires a large amount of training data,and it is urgent to obtain sample resources from a large database of medical images.The template image is a representative image of the crowd obtained by machine learning on a large number of sample images,which contains the average image characteristics of the sample population and the anatomical morphology and pixel grayscale differences between different individuals.Template images can be used for registration of individual patient images of disease diagnosis,human body simulation and other fields.This research is based on a large number of medical images collected by the research group of cooperation with national hospitals throughout the year,and a database is constructed to realize the storage and retrieval of images,and further research on algorithms to realize the automatic mining and learning of massive images,and the construction of different body parts and different collection model representation Template image.The research content of this article includes the following three aspects:(1)Construct a medical imaging database.Based on the 396,435 medical image data collected by the research group,a medical image database framework was designed to realize data retrieval based on image acquisition information,imaging site and patient information.Use MySQL to build a database back-end to provide data storage functions,use Python programming to achieve preprocessing and automatic storage of medical image data,and build a Web-based data retrieval platform through Django to provide online query,retrieval,modification and export of medical image data.(2)A template images generation method based on an active apparent model is used.The deformation field of the image area where the organ is located is used to represent the anatomical morphological change.The principal component analysis method is used to solve the principal components of the shape and grayscale changes.Further,the linear programming method Correlating these principal components with the physiological parameters of the human body,a template images that can adjust the shape of the organ and the pixel gray scale according to the physiological parameters is finally constructed.This method successfully constructed template images for CT(Computed Tomography)and heart PET(Positron Emission Computed Tomography)images of the chest and lumbar spine,and generated targeted template images for people with different physiological characteristics.(3)A template images generation method based on deep learning algorithms is implemented,and an unsupervised VoxelMorph images registration network is used to generate template images that vary from human physiological parameters.The experimental results show that this method can successfully generate brain MRI template images that change the structure of the brain with age.The study further compares the performance of the deep learning method with the previous active apparent model method,and verifies and analyzes the advantages and disadvantages of the two template image generation methods.The active apparent model method has better algorithm interpretability,and the deep learning method The model generation speed and accuracy are higher.This study realized the effective use of medical imaging database resources and laid a preliminary technical foundation for generating representative medical images of the Chinese population.