Leveraging ECG Signals for Automated Diabetic Patient Detection using CNN

Nor Surayahani  Suriani; Norzali  Mohd; Shaharil  Mohd Shah; Syahira Ahmad  Tarmizi; Siti Noorbalqis S  Rosli

Authors

Nor Surayahani Suriani Faculty of Electrical and Electronics Engineering, Department of Electronics Engineering, Universiti Tun Hussein Onn Malaysia, Malaysia
Norzali Mohd Faculty of Electrical and Electronics Engineering, Department of Electronics Engineering, Universiti Tun Hussein Onn Malaysia, Malaysia
Shaharil Mohd Shah Faculty of Electrical and Electronics Engineering, Department of Electronics Engineering, Universiti Tun Hussein Onn Malaysia, Malaysia
Syahira Ahmad Tarmizi College of Computing, Informatics and Mathematics, Universiti Teknologi Mara, Malaysia
Siti Noorbalqis S Rosli TDK-lambda Malaysia, Senai Industrial Area, Kulai Jaya, Johor, Malaysia

Keywords:

Blood Glucose, Machine Learning, Non-invasive monitoring, Signal Filteration

Abstract

Increasing blood glucose (BG) levels can lead to diabetes, affecting millions of adults worldwide. Insulin facilitates glucose absorption into cells for energy, and severe hypoglycemia in insulin-treated diabetics may cause abnormal ECG changes. Therefore, continuous monitoring of BG levels is critical. Traditional monitoring involves invasive finger pricks, whereas non-invasive methods, such as this study's approach, avoid the need for blood samples. This research proposes an IoT-based, non-invasive BG monitoring system that uses near-infrared (NIR) light and ECG signals. The ECG data are preprocessed using a Butterworth filter and analysed with a convolutional neural network (CNN). Several machine learning algorithms were compared to thirty subjects' ECG readings to test their performance and achieved almost 95% accuracy in detecting diabetic (DM) or healthy (non-DM) status.