Non-invasive Diabetes Detection System Using Photoplethysmogram Signals

Diabetes Mellitus (DM) is a chronic condition, where the body is unable to control blood sugar levels. In this paper, a non-invasive method to classify diabetic and non-diabetic cases is discussed. The collection of blood samples from patients by puncturing their fingers causes discomfort, pain, and infection, which are serious drawbacks of commercially available invasive blood glucose level monitoring systems. A novel non-invasive device for classifying blood sugar level is developed using a Near-Infrared sensor (NIR). Photoplethysmogram (PPG) signals, which are sensitive to blood glucose levels, are acquired using NIR sensors. ANOVA statistical analysis is used to identify significant features from the PPG signal. Quadratic SVM provided better results for features such as first derivative crest, time of inflection, and age in classifying the diabetic and non-diabetic subjects. Results from real-time PPG signals are compared with features taken from a diabetes dataset from Kaggle. The results showed that proposed PPG signal analysis method performed better.