| Summary: | Diabetes Mellitus (DM) is characterized by the body's impaired ability to process glucose, resulting in elevated blood glucose levels. Traditionally, monitoring Blood Glucose Levels (BGL) has required invasive procedures. Recent advancements in Artificial Intelligence (AI) have shown promise in estimating BGL using non-invasive Wearable Devices (WDs). These devices, typically worn on the body, utilize sensors to capture real-time data, enabling continuous monitoring. They can detect various physiological signals such as heart rate, blood pressure, glucose levels, and body temperature. WDs equipped with machine learning (ML) algorithms can identify specific data patterns known as Digital Biomarkers, aiding in categorizing and assessing the underlying condition. This utilization of biomarkers for tracking glycemic events marks a significant advancement in self-monitoring facilitated by non-invasive WDs. To achieve this, it is essential to investigate the correlations between characteristics obtained from non-invasive methods and indicators of glycemic health, with a focus on ensuring accuracy.
The research design and methodology aimed to evaluate the performance of AI models in predicting BGL among individuals with diabetes using data collected from submissions to the 1994 AI in Medicine Symposium. The experimental framework included Data Collection, Data Preparation, ML Model Selection/Development, and thorough evaluation of metrics.
The results demonstrate the AI models' ability to accurately estimate the relationship between glycemic metrics and features extracted from the datasets, as indicated by Root Mean Square Error (RMSE) values (95.90), indicating robust BGL estimation. This study supports the use of commercially available WDs for BGL estimation in diabetic populations, underscoring the potential of AI-assisted non-invasive WDs in diabetes management.
|
|---|