The Design Of Myocardial Infarction Early Aided Diagnosis System Based On GPU

Heart disease is one of the leading threatens to human health nowadays. Myocardialinfarction (MI), a common cardiovascular disease, is one of the common symptoms ofinternal medicine. This disease, which usually onsets in middle age, has a fatal rate of nearly25%, leading it a research hotspot with importance and emergency in medical field.Electrocardiogram, as a diagnostic method of MI, is worth studying.Deterministic learning theory, using concepts and methods of adaptive control andnonlinear dynamical system, aims at studying the acquisition, representation, storage andutilization of knowledge in unknown dynamic environments. This theory, with learningmechanisms based on dynamic environment, is different from learning theories based onstatic function mapping methods.Dynamical pattern recognition is a complex process, with abundant mode and neuralcomputing affecting greatly the efficiency of algorithm implementation, and has a highrequirement towards specific implementation platforms. Therefore, resolving the massive datacomputing problem is a key of realizing rapid dynamic pattern recognition. To solve theproblem, deterministic learning theory presents a new model for information processing,dynamic parallel distributed processing (DPDP), which has high computational efficiency inlarge-scale dynamic pattern recognition, and can be well implemented using parallelcomputing. Date processing rate can be speeded up using parallel computing based onmessage passing interface (MPI) network clusters, however, substantial delay can be causedby communication and data transmission between computers. Multithread programmingunder multi-core CPU, although avoid the deficiency of the former, of which the start, switch,communication and synchronization of thread is time consuming, and is coarse-grainedmultithreading in most cases. Massively parallel computing based on GPU, employing SingleInstruction Multiple Data (SIMD) command structure and streaming multiprocessorframework, has a high performance of process ability and huge memory bandwidth, hence befit for computing of computationally intensive and massive data. Parallel computing based onGPU can realize finer grained multithreaded programming, and can run thousands oflightweight threads simultaneously without spending much time in the operation process, thusimproving greatly by the program operation efficiency.Electrocardiograms auto-analysis and subsequent disease diagnosis is in essence aprocess of dynamic pattern recognition. In this process, normal and abnormal MI electrocardiosignals are regarded as different dynamic patterns, and subjected to modelingand learning separately. A typical pattern library that can realize the pattern recognition of thedata of new patients is constructed, and then automatic test results is presented, therefore canprovide diagnosis reference for doctors. In this study, in order to deal with the low accuracyand sensitivity of traditional electrocardiogram in diagnosis of MI, an early diagnosis methodof MI based on deterministic learning and dynamical pattern recognition is introduced. Andcombined with GPU general compute to solve the massive data computational problem ofdynamic pattern recognition, a MI early aided diagnosis system is designed. This system canbe employed as a effective tool of MI early aided diagnose in clinical medicine.