부산대학교 도서관

Dicrotic wave fiducial point can represent a useful key metric for vasomotor activity evaluation and its morphology during a complete cardiac cycle gives insight about hemodynamic activity, especially for changes in blood pressure values. As an important fiducial point, despite of human evaluation, its behavior is rarely interpreted by an automated algorithm. The last approach is suitable for individual monitoring, but a drawback from scalability point of view. This paper proposes an algorithm for dicrotic wave detection in various blood pressure signal morphologies, extracted from photoplethysmography waveforms. Four major classes with an easy nomenclature to be understand are defined, based on different signals morphologies. A total of high quality 400 periods where visual inspected and selected to build a reference database for algorithm performance testing. An overall performance of 95.25% is obtained, but this detection rate is pulled down especially by depicted classes where dicrotic wave is almost absent and affected by the artifacts. The tested dataset has discarded unrealistic waveform suffering from artifact effect. The main source of error is due to multiple weak inflection points during one cardiac cycle, comparable in morphology with the investigated fiducial points.