ABSTRACT
Implantable cardioverter defibrillators are life-saving devices for people with heart disease. They sense the electrical activity of the heart through leads attached to its tissue. The sensed signals are called intracardiac electrograms and their interpretation is in many instances still a challenging pattern recognition task. This is especially the case because the defibrillators are battery powered, and most conventional recognition techniques are computationally intensive. We present here neural network techniques for electrogram recognition and describe their application to the detection of two rhythms that cannot be recognized by present day defibrillators. The implementation of such networks in micropower very large-scale integration is also described. A method for resolving the problem of morphology changes due to tissue growth is addressed by a method in which the neural network continuously learns using patterns that are automatically labeled.
