Photoplethysmogram-based Cerebrovascular Disease Detection: A Machine Learning Approach

Cerebrovascular disease is responsible for hemorrhage stroke, which is one of the significant global causes of morbidity. Detection at an early stage is necessary to avoid further complications. However, the traditional ways of detecting this disease are image-based analysis like computed tomography (CT) scan and magnetic resonance imaging (MRI). These are costly and time-consuming processes that can be replaced by Photoplethysmogram (PPG) signal analysis. It can identify changes in blood volume and hence has the potential to detect cerebrovascular disease, which causes problems with blood flow in the brain region. So far, only a few studies have used PPG signals to detect Cerebrovascular disease from PPG signal. Also, the use of frequency domain features for detecting the disease is yet to be explored. In this study, 84 PPG segments have been selected from a publicly available dataset. Later, several frequency domain and time domain features were extracted and used separately in different machine learning classifier designs. Among the various models tested, KNN performed the best, providing 97.61% accuracy by combining both types of features. Additionally, the study has determined that frequency domain features extracted from PPG signals have lower capability as they have provided the least accuracy. The study can help develop future medical and wearable devices for disease detection.