Noninvasive blood glucose sensing by secondary speckle pattern artificial intelligence analyses.

Abstract

Diabetes is a prevalent disease worldwide that can cause severe health problems. Accurate blood glucose detection is crucial for diabetes management, and noninvasive methods can be more convenient and less painful than traditional finger-prick methods. We aim to report a noncontact speckle-based blood glucose measurement system that utilizes artificial intelligence (AI) data processing to improve glucose detection accuracy. The study also explores the influence of an alternating current (AC) induced magnetic field on the sensitivity and selectivity of blood glucose detection. The proposed blood glucose sensor consists of a digital camera, an AC-generated magnetic field source, a laser illuminating the subject's finger, and a computer. A magnetic field is applied to the finger, and a camera records the speckle patterns generated by the laser light reflected from the finger. The acquired video data are preprocessed for machine learning (ML) and deep neural networks (DNNs) to classify blood plasma glucose levels. The standard finger-prick method is used as a reference for blood glucose level classification. The study found that the noncontact speckle-based blood glucose measurement system with AI data processing allows for the detection of blood plasma glucose levels with high accuracy. The ML approach gives better results than the tested DNNs as the proposed data preprocessing is highly selective and efficient. The proposed noncontact blood glucose sensing mechanism utilizing AI data processing and a magnetic field can potentially improve glucose detection accuracy, making it more convenient and less painful for patients. The system also allows for inexpensive blood glucose sensing mechanisms and fast blood glucose screening. The results suggest that noninvasive methods can improve blood glucose detection accuracy, which can have significant implications for diabetes management. Investigations involving representative sampling data, including subjects of different ages, gender, race, and health status, could allow for further improvement.