Advanced EEG Signal Preprocessing for Precise Non-Invasive Blood Glucose Estimation

Non-invasive blood glucose monitoring through EEG signals presents a comfortable and continuous alternative to traditional, invasive methods like finger pricking, which can be inconvenient for diabetes patients. EEG data captures subtle brain activity changes influenced by glucose fluctuations, making it a promising source for glucose estimation. By applying advanced preprocessing techniques such as Independent Component Analysis (ICA), band-pass filtering, and normalization, key EEG features Alpha, Beta, and Theta power bands were isolated and analyzed. Machine learning algorithms, particularly Support Vector Machines (SVM), were then employed to predict glucose levels, achieving an accuracy of 87%. Notably, Alpha power decreased and Beta power increased as glucose levels rose, aligning with known physiological responses in hyperglycemia. This approach allows for needle-free, continuous glucose tracking, which could help patients maintain stable glucose levels and reduce health complications. Future research will aim to improve model accuracy, incorporate larger datasets, and explore additional biosignals to enhance practical applicability in clinical environments.