| The search of ECG automated diagnosis has been focus in the last years. If the search can be put into practice and its performance is reliable, it will be a great fruit in both medical field and artificial intelligence field. The content of the research includes preprocessing, waveform detection and diagnose technologies. The research involves much theories and technologies and includes a lot of work.A research model based on others fruit is proposed in this paper and it can integrate experience of ECG specialists and the technologies of engineering. However, the model involved too much work, so a research frame which can be extended dynamic is used in practice. The work preformed at present focused on the utilization of the experience of specialists, and some data analyses method in engineering are also utilized in this work.The following work has been performed in this research:1. A researching platform has been constructed. The platform includes MIT-BIH database and its resample database, a 12 channel synchronous ECG database and a virtual ECG generator. The sample rate of the resample database is 250Hz which is the multiple of power line frequency in China and this rate will simplified the research. The 12 channel synchronous ECG database is used to store the sample for the search of ECG auto-diagnose, and also used as an examples database for diagnose knowledge refine in the further research. The virtual ECG generator can generate the desired ECG signal for research, and it will promote the research greatly. 2. Based on the idea of expert system, a deducing and interpreting system for ECG is made. The expert knowledge is managed through database manage engine. The method makes the analyses of expert knowledge more easily. In addition, the system also contains the construct of "script database", "and ruler database" and "interpreting system". These parts make the system modifying and debugging more easily. 3. The ECG preprocessing and characteristic point detection are studied. The ECG preprocessing includes the elimination of high frequency noise, baseline wander and power line interference. The characteristic point detection includes the P, QRS and T wave detecting and the onset and offset point determination.4. Some work on symbol interpretation and diagnosis rulers defines are performed. This part includes the symbol interpretation method research, such as the period, amplitude and the type of waveform of P, QRS and T. On the base of this work, some diagnosis rulers are defined.5. Some classifying methods based on mathematic index are also researched. The determination of VT/VF and classifying ECG waveform based on support vector machine (SVM) are studied. In the determination of VT/VF, the filter method, probability density method and sequential hypothesis method are discussed. In the ECG waveform classifying, the SVM is generalized and used to multi-category ECG classify. The method is tested with the sample extracted from MIT-BIH, and the result is compared with the ANN method. Although the generalization of SVM is better than the ANN, it still has some limitation in great sample test set. The ECG automatic diagnosis cannot depend on only one method. The innovations of this study are as follows:1. Based on the idea of expert system (ES) and combined with mathematic method, a search frame of ECG automated diagnose is proposed. The system, which integrates expert's knowledge and engineering technology, has been put into practice in this work firstly. Through symbol interpretations, the ECG data is mapped as the basic symbols which are used in the description of expert knowledge. With these basic symbols, ECG deducing and interpreting system can get diagnose result.2. A searching platform is constructed. The platform includes MIT-BIH database and its resample database, a 12 channel synchronous ECG database and a virtual ECG generator. In this platform, virtual ECG generator is firstly introduced to generate ECG signal which is not easy sampled in practice. It makes the research mo... |
PDF Full Text Request