Current situation of computer interpretation of electrocardiogram is reviewed. Techniques of filtering or smoothing of digitized curve, point recognition, time normalization and pattern recognition are included in the scope. The most parts of the data processing system of electrocardiogram has been organized with success by several groups, but the stage of pattern recognition is still an important subject for argument.
The trial to divide the normals and the abnormals by means of a single parameter such as polar vector, ventricular gradient and so forth has not yet shown satisfactory results. Statistical approach using probability density function for various parameters characterizing electrocardiographic patterns such as amplitudes, durations and intervals has been proved feasible for only a part of various abnormalities. A discrimination function derived from Fourier analysis distinguished one kind of abnormality from normals but, to be regarded generally applicable, it may need further experience with patterns of various types.
Use of multiple adaptive matched filters resulted in a success in classification of QRS complex of various patterns. The method includes capability of self-learning which is performed by initiating and storing memory filters, being modified frequently as the recognition of various patterns proceeds. So it simulates, at least partly, human pattern recognition and an extensive study on this operation may help understanding the mechanism of pattern recognition by human brain.
It was pointed out that, for engineers, electrocardirgram is important material for bio-medical data processing or pattern recognition because of easiness in approach and handling. For physicians, the first introduction of computer into practice will likely occur in the automatic interpretation of electrocardiogram and it is expected computer diagnosis of electrocardiogram is followed by renovation and reorganization of clinical diagnostics in various fields of medicine through use of computer.
View full abstract