Classification of Photoplethysmogram Signal Using Self Organizing Map

Photoplethysmography (PPG) is a popular non-invasive technique to estimate cardiovascular functions by the use of infrared light at peripheral body parts. This paper describes an approach for clinical feature extraction from finger PPG under resting condition and its binary classification. The peak and foot fiducial points were detected by an algorithm that utilized the concept of threshold comparison between the consecutive samples. The other fiducial points, dicrotic notch and diastolic peak were detected using acceleration PPG (APG). The extracted clinical features were: pulse width, systolic amplitude, Peak to peak time, and ratio of areas before and after dicrotic notch in a complete cycle. These were fed to a Self Organizing Map (SOM) to form a binary classifier. In the study, short duration PPG data from 30 healthy volunteers and 20 patients with pre-assessed cardiovascular diseases were used. The average detection sensitivity for systolic peaks and foots were 100% and for diastolic peaks and dicrotic notches were 95% and 96% respectively. The correct classification of normal and abnormal data was visually estimated from the weight distance plot and Hit plot. The software can be upgraded to form a PC based online PPG assessment tool.