A Neural Network System for Detection of Life-Threatening Arrhythmias, Based on Kohonen Networks

摘要

A system using Kohonen networks is designed for automatic recognition of malignant cardiac rhythms and their differentiation from benign rhythms. The data were contained in digital files. Initially the QRS complex was detected and a set of features are extracted. The features passed to a Kohonen network for classification of the beat as normal or aberrant. This information together with the RR interval is passed to a buffer containing information on 20 consecutive beats. Statistical parameters of the 19 RR intervals together with the beat class were passed to a second Kohonen network for rhythm classification. Multiple ECG files, each containing approximately 3 minutes of data, were used for testing of the system. The system differentiated correctly between malignant and benign rhythms, but had difficulty in identifying correctly the various types of rapid benign rhythms.