Study Of Early Warning For Cerebrovascular Risk Based On Deep Beliefs Networks

Cerebrovascular disease (CVD), namely stroke, is chartered by high death rate and high disability rate. However, it is possible to reduce the morbidity and death rate of CVD by identifying the risk of getting infected in advance as well as taking proper prevention and control measure. There had been massive risk assessment models established according to various risk elements from different perspectives by academics and medical professions.This paper studies CVD with focus on the risk early warning in primary defence of CVD using literature analysis, statistical analysis, empirical analysis and other research methods. The research is mainly about how to use the risk factors of cerebrovascular risk to establish models for early warning with existing technology aimed at realize the early warning for disease. The primary studies are as follows:(1) Establishment of risk early warning index system. Through document analysis and summary, a relatively comprehensive primary CVD risk early warning system was set up. Then according to information entropy algorithm in Rough Set Theory, the attributes of primary index are deducted, and a highly representative and sensitive indicator was extracted, resulting in an appropriate risk early warning index system for CVD.(2) Establishment of risk early warning model for CVD. The model is based on the application of Restricted Boltzmann Machine (RBM) and Back Propagation Algorithm (BPA). First, DBNs is built by construction and pile up of RBMs. After that, a small parameter adjustment is made at the top layer in accordance with BPA. At the end of this process, a risk early warning model for CVD is constructed.(3) Model simulation and effect analysis. The study of model attributes and parameter adjustment is realised by abstraction of partial patient information in collaborative prevention and control medical cloud platform. Simultaneously, evaluate the effectiveness of the risk early warning model based on the simulation result. Moreover, compare experiment result with the result of pre-waring model for LM-BP neural network to verify the effectiveness of DBNs in early warning.(4) The application of early warning model under collaborative prevention and control cloud platform. An application scheme of early warning model in information sharing system was put forward in accordance with current situation and data storage mechanism of the information sharing system in medical cloud platform. On the other hand, by combination with medical union model, it establishes a system intervention strategy for collaborative prevention and control of CVD based on early warning result.The main contribution of this paper is that it realises the extraction of high quality risk forecast index by attribute deduction of original system based on Rough Set Theory, and applies DBNs in the risk early warning of CVD. The effect of early warning models based on DBNs has been proved more efficient than traditional artificial neural networks. The Research helps to identify risk of cerebrovascular to promote the primary prevention.