Automatic identification of late potentials.

The purpose of the present study was to develop an algorithm for the automated identification of ventricular late potentials (LP) that can be recorded non-invasively by means of the signal averaging technique. This new algorithm was designed to determine the possible presence and the onset and duration of a given LP by analyzing the end of the QRS complex. As there is no objective standard for identifying these late potentials, the new algorithm was developed by continuous comparison and adaptation to visual analysis in 65 patients (algorithm definition phase). In the subsequent trial phase, visual and automated analysis were compared in a second cohort of 50 patients (40 patients with and 10 patients without late potentials). In the latter 50 patients, the results obtained with the new algorithm corresponded to the analysis made by at least two of three independent observers in 92% of cases. In only four patients--one without and three with late potentials--there was no agreement between the observers and the new program. In conclusion, the new algorithm can be used reliably for the evaluation of late potentials. The results are more objective. They are reproducible, which is of great advantage when data from different groups are to be compared or when less experienced investigators are using non-invasive methods for registration of late potentials.