Research On Parallel Processing Methods For Big Data In Medical&Healthcare

With the continuous development of the information technology,there are more and more information systems used in hospitals.With the popularity of networking,mobile medical,automatic analysis instrument and wearable devices,hospitals,doctors and patients have become the direct creator of data.A large amount of medical data will be stored in database.To fully exploit the potential value of big medical data play an important role in improving the quality of medical service and saving the cost of health care.However,the process of mining the potential value is full of challenges.One problem is the burden of computational.The growth of data dimension and quantity made it is hard to solve complex problems with traditional serial programs.If a doctor want to dig out some useful information from thousands of clinical records,may be a few hours or even several days will cost in the step of model training.Doctors will waste a large number of time and energy on analysis of big medical data.Another problem is that the completeness and timeliness of medical data often leads the machine learning models have poor generalization ability.It is hard to process new patient with personal data.In the background of bif medical data,to solve the two probems mentioned above,this paper studies the current parallel computing methods to deal with the opportunities and challenges of medical big data.The main research contents are as follows:This paper studed the hardware architecture,storage model and programming model of CUDA parallel platform.Combined with the characteristics of ECG data,we designed a parallel heart beat segmentation and feature extraction methods.Combined with the characteristics of generalized regression neural network,we designed a parallel generalized regression neural network to classify the heartbeat.So as to provide supplementary information for the diagnosis of cardiovascular diseases.In the premise of ensuring the precision of the classification,our method obtained a hundred times increase on efficiency.This can greatly reduce the data processing time and save more time for doctors.So that doctors can put more energy into the process of diagnosis and treatment.This paper studed the hardware architecture,storage model and programming model of OpenCL parallel platform.we designed an automatic user-adapted physical activity classifier using a smartphone.The data used in this paper were obtained from acceleration sensors and gyroscopes integrated into smartphones.We used Adaboost-stump to classify five common activities:cycling,running,sitting,standing and walking,and achieved a satisfactory classification accuracy.The program can fully exploit the computing power of smartphones and the users can train their own personalized model conveniently on the smartphone.We designed a clinical decision support system base on parallel computing and machine learning.This system included linear regression,logistic regression,neural network,Adaboost for classification and k-means,FCM for clustering.In the process of running efficiency,the above 6 algorithms were parallel designded in 3 different frameworks.The system shows excellent operating efficiency and classification accuracy.This system can complete the role of providing medical assistance effectively and accurately.