Noninvasive blood glucose level estimation using bioimpedance spectroscopy and machine learning: an integrated...

Area of Science:

• Biomedical Engineering
• Medical Devices
• Machine Learning in Healthcare

Background:

• Traditional blood glucose monitoring requires painful finger pricks.
• Diabetic patients need convenient and less invasive monitoring solutions.
• Existing noninvasive methods often struggle with accuracy, especially at extreme levels.

Purpose of the Study:

• To develop a noninvasive blood glucose detection method using bioimpedance spectroscopy and machine learning.
• To improve accuracy and reliability in estimating blood glucose levels (BGL).
• To create a potential replacement for traditional invasive blood glucose monitoring.

Main Methods:

• Utilized bioimpedance spectroscopy combined with machine learning.
• Implemented an adaptive data generation method for small and extreme BGL samples.
• Employed sparse group LASSO for frequency and feature selection.
• Constructed a regression model using the XGBoost algorithm with hyperparameter optimization via Optuna.
• Validated the model using five-fold cross-validation on data from healthy individuals and type 2 diabetes patients.

Main Results:

• Achieved a mean absolute relative difference of 9.55% in blood glucose level estimation.
• Demonstrated high clinical acceptability with 99.38% of results in Clarke Error Grid zone A+B.
• Showcased 90.63% of results within the clinically critical zone A.
• The developed method shows promise for wearable device integration.

Conclusions:

• The proposed noninvasive method using bioimpedance spectroscopy and machine learning is accurate and reliable.
• This technique significantly reduces patient discomfort compared to invasive methods.
• The technology holds potential for development into wearable devices for continuous, noninvasive blood glucose monitoring.