Dynamic Pacing Ecg Analysis Methods

With the rapid advancements in pacing technology, the -pacemaker function has been enhanced and its indication has been widened. Because of the widespread use of pacemakers and the necessity to monitor the patients with implanted pacemakers, it is important to analyze paced ECGs (PECGs) automatically. The previous technique can only be used in those which are paced by single-chamber pacemakers. Three subjects are studied in the dissertation: building PECG database, studying the method of the single-chamber and dual-chamber PECGs analyzation and developing a software used in PECGs interpretation.The lack of database is one of the main reasons to hinder the automatic PECGs analysis advancing. A PECG database is built by recording PECGs in a hospital. Each sample consists of 4 files, i.e., data file, annotation file, head file and pulse file. Two leads data are compacted stored alternatively in data file. The beat type and its address are annotated in the annotation file. The basic information of the sample is described in the head file, including record number, the number of leads, sampling rate, the data length, etc. Pulse file states the time that pulse delivers. Except pulse file, the formats of the other three files are as the same as those of MIT-BIH ECG arrhythmia database. In order to get abnormal PECGs that are difficult to record in hospitals at present, some PECGs published on magazines and books are converted into samples by digitilizer. Later they will be replaced by recorded samples.It is first used by neural network-based PECGs analyzation. Waveform data (which are called PQRST ) are used as the input of the network, which makeit not necessary to detect P wave, determine PR interval in analyzing dual-chamber and atrial paced ECGs. The PQRST data include the main information in a beat. It is consist of P wave, QRS wave, part of T wave and pacemaker pulse ( if there is ) . They are got by detecting R wave andexpanding them forward and backward. The model of the network is ART2 ( Adaptive Resonance Theory ) . According to the structure of the ART2 network, the algorithm is developed. ART2 can cluster PQRST by unsupervised training. Because the network may not work stably by fast learning algorithm or slow learning algorithm, mediate learning algorithm is proposed. Stable clustering results are archived. Combining the whole information, including pacemakers' mode, programming parameters, pulse, QRS characteristics and cluster result, the beat type is determined finally. The method is evaluated by MIT-BIH ECG arrhythmia database, SDMU ECG database and PECG database which is created by the dissertation.