An Algorithm for Real-Time Pulse Waveform Segmentation and Artifact Detection in Photoplethysmograms

• Published in: IEEE journal of biomedical and health informatics Vol. 21; no. 2; pp. 372 - 381
• Main Authors: Fischer, Christoph, Domer, Benno, Wibmer, Thomas, Penzel, Thomas
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.03.2017; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Aged; Algorithm design and analysis; Algorithms; Annotations; Autonomic nervous system; Cardiac output; Classification algorithms; Compression; contour analysis; Data compression; decision list; embedded; Female; Filtering algorithms; Humans; Laboratories; Longitudinal waves; Male; Measuring instruments; Medical devices; Medical electronics; Medical equipment; Microcontrollers; Middle Aged; Oxygen content; Pattern recognition; Pattern Recognition, Automated - methods; photoplethysmogram; Photoplethysmography - methods; Pulse Wave Analysis - methods; Real time; Real-time systems; Segmentation; Signal processing algorithms; Signal Processing, Computer-Assisted; Signal quality; Sleep; Sleep - physiology; Time compression; Time domain analysis; Vascular diseases
• ISSN: 2168-2194; 2168-2208; 2168-2208
• DOI: 10.1109/JBHI.2016.2518202

Photoplethysmography has been used in a wide range of medical devices for measuring oxygen saturation, cardiac output, assessing autonomic function, and detecting peripheral vascular disease. Artifacts can render the photoplethysmogram (PPG) useless. Thus, algorithms capable of identifying artifacts are critically important. However, the published PPG algorithms are limited in algorithm and study design. Therefore, the authors developed a novel embedded algorithm for real-time pulse waveform (PWF) segmentation and artifact detection based on a contour analysis in the time domain. This paper provides an overview about PWF and artifact classifications, presents the developed PWF analysis, and demonstrates the implementation on a 32-bit ARM core microcontroller. The PWF analysis was validated with data records from 63 subjects acquired in a sleep laboratory, ergometry laboratory, and intensive care unit in equal parts. The output of the algorithm was compared with harmonized experts' annotations of the PPG with a total duration of 31.5 h. The algorithm achieved a beat-to-beat comparison sensitivity of 99.6%, specificity of 90.5%, precision of 98.5%, and accuracy of 98.3%. The interrater agreement expressed as Cohen's kappa coefficient was 0.927 and as F-measure was 0.990. In conclusion, the PWF analysis seems to be a suitable method for PPG signal quality determination, real-time annotation, data compression, and calculation of additional pulse wave metrics such as amplitude, duration, and rise time.