Research On Parallel Reconstruction And Segmentation Of CT Image Based On Spark

With the advent of the era of big data,the medical industry is constantly generating a large amount of data.Moreover,medical image can provide intuitive insights for doctors and patients to understand the internal information of the human body,and is widely used in clinical examination,diagnosis and treatment.Computed Tomography(CT)technology acquires tomographic images of human anatomical structures in a non-invasive manner,which has the advantages of convenient examination and clear structure.It is currently one of the most important methods used in clinical diagnosis of diseases and evaluation of curative effects.However,CT image reconstruction is a computationally intensive and time-consuming process.Since the development of CT technology,reconstruction acceleration has been a problem that researchers are working on.In addition,image analysis such as segmentation of liver tumors and pulmonary nodules is a common requirement in the medical field.With the development of socially efficient and automated development,automated image analysis methods are urgently needed in clinical applications to meet actual needs.Since 2012,with the rapid development of big data and artificial intelligence technology,it is possible to build a platform with a big data processing framework as the main body and deep learning technology as a branch to accelerate CT image reconstruction and complete medical image analysis tasks.It is of great significance to serve patients,reduce the workload of doctors,and improve medical efficiency.In order to realize the rapid completion of CT image reconstruction process and analysis tasks for the massive medical data in the era of big data,the research goal of this paper is to parallelize the CT image reconstruction algorithm program and realize the CT image liver distributed segmentation by using deep learning technology on the distributed framework Spark platform.The research content mainly includes the following four aspects:First,by installing Centos7 computer operating system,configuring Spark and its related software running environment,and setting up wired LAN,etc.,a Spark cluster with a size of 8 nodes is built,which is the basic platform of CT image distributed reconstruction and liver distributed segmentation;Secondly,for the distributed computing design features of Spark framework,the Filter Backprojection(FBP)algorithm represented by analysis in CT image reconstruction is designed to be parallel The rmodel was tested using FBP to reconstruct the same image size,number of different images,and different image sizes and the same number of images;Thirdly,the Simultaneous Algebraic Reconstruction Technique(SART)program,which is represented by iteration in CT image reconstruction,is designed as a parallel model,and the same image size,different iterations,and the number of iterations are reconstructed when SART is used to reconstruct a single image.Experiments show that the parallel reconstruction method of the two algorithms is proportional to the number of cluster nodes,and the parallel efficiency approaches 1.Fourthly,for the insufficiency of convolutional neural network,such as loss of information in pool operation and poor recognition ability of metamorphosis,capsule network model was used to segment the abdominal CT liver region and use the LiTS2017 abdominal CT dataset to train four fine-tuned capsule networks.The model is used to segment the image liver region,and the model with the least model parameters and the test set segmentation precision up to 93.20%is deployed into the Spark cluster to complete the parallel segmentation task of batch images.Based on the above work,big data and deep learning technology are applied to the medical field,and the processing flow of CT image from reconstruction to segmentation is basically realized.