Non-invasive Hyperglycemia Detection via Electrocardiogram Using Discrete Wavelet Transform and Machine Learning

Abstract

INTRODUCTION AND AIM: Diabetes mellitus (DM) is a chronic metabolic disorder characterized by elevated blood glucose levels (BGLs), traditionally monitored through invasive methods. Continuous monitoring of BGL is essential to prevent severe complications. This study proposes a non-invasive approach to detect hyperglycemia by analyzing electrocardiogram (ECG) signals using the discrete wavelet transform (DWT) and machine learning techniques. The relationship between diabetes and electrocardiographic (ECG) signals is highly relevant in biomedicine, as DM is a significant risk factor for cardiovascular diseases. MATERIALS AND METHODS: A total of 210 individuals (healthy and diabetic, aged 18-70 years) were analyzed, with ECG signals processed using DWT to extract heart rate (HR) in beats per minute (bpm) and heart rate variability (HRV) parameters. These features were used as inputs for support vector machine (SVM) and k-nearest neighbors (KNN) classifiers to distinguish between normal and hyperglycemic states (>150 mg/dL). RESULTS: The DWT-based automatic feature detection achieved an accuracy of 99.8% for both HR and HRV. Pearson correlation analysis revealed moderate correlations between glucose levels and HR (0.2985) and HRV (-0.373), with a combined correlation index of 0.6428. CONCLUSION: Classification results showed an accuracy of 97% for normal glucose levels and 93% for hyperglycemia detection using both SVM and KNN. These findings indicate that ECG signal characteristics can serve as an adjunct for non-invasive hyperglycemia detection.