SoC Design of ECG Intelligent Classification

Every year, many people die of cardiovascular disease around the world, even though most of this disease can be effectively prevented at an early stage. Traditional heart rhythm monitoring methods rely on doctors, resulting in poor real-time performance. This paper designs a System-on-Chip (SoC) targeting at the electrocardiography (ECG) intelligent classification. Real-time heart rhythm monitoring brought by artificial intelligence (AI) can effectively help catch the early treatment. A Deep Neural Network (DNN) model is adopted to classify the four common categories of heart rhythms. The test results show that the DNN model used in this paper manages to classify the four common heart rhythm categories with a correct rate of 96.4%. In order to accelerate the operation process, the system contains a specific coprocessor running the complex operation process of DNN. The coprocessor and the central processing unit (CPU) interconnect with each other by Nuclei Instruction Co-unit Extension (NICE) bus. Finally, a 40-time increase in the system operation speed is achieved. Meanwhile, the hardware logic resource consumption is reduced by about 32% through quantizing the model parameters from 32-bit floating-point number into 16-bit fixed-point number.