Logo of the Physikalisch-Technische Bundesanstalt
symbolic picture: "magazines"

ECG Interpretation by Means of Pattern Recognition

What has been successfully accomplished in the case of fingerprints, namely the direct comparison of the patterns, is now also possible with heart signals. A new method allows signal patterns of a patient’s electrocardiogram (ECG) to be compared with evaluated patterns of other patients’ ECGs stored in a data base. The results will help the doctor to ascertain the patient’s state of health.

Example of ECG evaluation by pattern recognition
The bar graph is the result of the evaluation of those ECGs stored in a database, whose signal patterns are most similar to those of the unknown ECG. While 85,9% of them (blue bar) point to a block in the heart’s conduction system, indications of other heart diseases (remaining bars) are distinctly less pronounced.

The computer programs so far available for the automatic interpretation of ECGs propose a diagnosis on the basis of complex decision rules or with the aid of neuronal networks. Even though many of these methods are rather sophisticated, they can detect, and take into account, the large number of biological signals to a limited extent only. To include rarely found special cases or new medical findings requires software adaptations which are often very expensive.

A method developed by PTB scientists has none of these disadvantages. It starts with the assumption that the same diagnosis can be assigned with high probability to ECGs the signal patterns of which are in good agreement. With the aid of modified correlation methods, reference cases are selected from an ECG database which agree best with the signal patterns of the unknown ECG. When similar ECGs have been found, the results of the corresponding cardiac examinations or the diagnoses can be used to interpret the unknown ECG. The reliability of the method depends decisively on the size and completeness of the database used for this purpose. It has to reflect the spectrum of the most significant cardiac diseases.

The method was tested using 10000 ECGs. The percentage of correct diagnoses ranged between 75 and 95 depending on the respective categories of heart conditions.

Contact at PTB:

D. Kreiseler,
phone: +49 (0) 30 34 81-0