RESUMEN
Objective In order to make up for the deficiency in the existing photoplethysmography feature point recognition algorithms which need manually setting the selecting threshold and have poor adaptability to complex waveforms, an automatic reognition algorithm for feature points based on monotonic increase in geometrical characteristics of pulse wave ascending branch was proposed. Methods A ‘reference point’ was determined in each pulse period by zero crossing detection after two Hilbert transformation. The nearest concave and convex inflection points that searched around the ‘reference points’ were the notchs and systolic peaks. Results By using the 18 sets of data in the MIT-BIH standard database for verification, the average sensitivity, precision and detection accuracy reached 99.94%, 99.72% and 99.68%, respectively. Compared with the existing four algorithms, there was a significant improvement in the precision. Feature points could still be accurately identified for complex waveforms. Conclusions The proposed algorithm achieved a higher detection accuracy in the process of searching and determining the position of the pulse wave notchs and systolic peaks, and exhibited a stronger adaptability to the waveform change. The research results provide a good foundation for physiological and pathological analysis through pulse wave features extraction in clinic.
RESUMEN
Objective In order to make up for the deficiency in the existing photoplethysmography feature point recognition algorithms which need manually setting the selecting threshold and have poor adaptability to complex waveforms, an automatic reognition algorithm for feature points based on monotonic increase in geometrical characteristics of pulse wave ascending branch was proposed. Methods A ‘reference point’ was determined in each pulse period by zero crossing detection after two Hilbert transformation. The nearest concave and convex inflection points that searched around the ‘reference points’ were the notchs and systolic peaks. Results By using the 18 sets of data in the MIT-BIH standard database for verification, the average sensitivity, precision and detection accuracy reached 99.94%, 99.72% and 99.68%, respectively. Compared with the existing four algorithms, there was a significant improvement in the precision. Feature points could still be accurately identified for complex waveforms. Conclusions The proposed algorithm achieved a higher detection accuracy in the process of searching and determining the position of the pulse wave notchs and systolic peaks, and exhibited a stronger adaptability to the waveform change. The research results provide a good foundation for physiological and pathological analysis through pulse wave features extraction in clinic.