An ECG Ischemic Detection System Based on Self-Organizing Maps and a Sigmoid Function Pre-processing Stage

Abstract

The detection of ischemia from the electrocardiogram is a time-consuming visualization task requiring the full attention of the physician. This task may become inviable when many patients have to be monitored, for example in an Intensive Coronary Care Unit. In order to automate the detection process and minimize the number of misclassified episodes we propose the use of an Artificial Neural Net (the

Self-Organizing Map - SOM) along with the use of extra parameters (not only ST segment and T wave deviation) measured from the ECG record. The SOM is a widely used Neural Network which has the ability to handle a large number of attributes per case and to represent these cases in clusters defined by possession of similar characteristics.

In this work we propose a three-block ischemic detection system. It consists of a pre-processing block, a SOM block and a timing block. For the pre-processing block we use the sigmoid function as a smoothing stage in order to eliminate the intrinsic oscillation of the signals. The SOM block is the ischemic detector and the timing block is used to decide if the SOM output meets the ischemic duration criteria.

With this strategy, 83:33% of the ischemic episodes (over 7 records) tested were successfully detected, and the timing block proved to be robust to noisy signals, providing reliability in the detection of an ischemic episode. The system could be useful in an Intensive Coronary Care Unit because it will allow a large number of patients to be monitored simultaneously detecting episodes when they actually occur. It will also permit visualisation of the evolution of the patient's response to therapy.