The Development of a Neural Network-Based Ambulatory ECG Monitor

Abstract

The number of hospital cardiac referrals and the delay to appropriate treatment could potentially be reduced by the use of new technology which enables the primary care provider to carry out a long term cardiac examination. The technology uses neural computing techniques in a portable battery powered unit to analyse a patient’s electrocardiogram (ECG) in real time. At the end of the examination the unit is connected directly to a printer to provide a detailed report of the findings. The report can be used as the basis for a referral decision. This paper describes the development of the device and studies carried out to evaluate the performance of the technology employed by the unit. The device employs a panel of Kohonen neural networks together with conventional processing and is embedded in a custom 32 bit micro-controller circuit powered by four AA cells. The first study examined 26 minute ECG traces from 67 individuals comprising cardiac in-patients, rehabilitation patients and healthy subjects and compared the results of arrhythmic analysis with a total of five cardiologist’s interpretations. The results show that the technology is at least as good as the cardiologists, averaging 96% accuracy compared to an average of 89.25% for the cardiologists. The second study employed 24 hour ECG monitoring using the device on 121 patients reporting to General Practitioners with possible cardiac symptoms and examined the effect of using the device on subsequent cardiac referrals. The results showed a reduction of 50% in the number of referrals and a 65% reduction in waiting time for those patients still referred.