Exploring the Spectrum of Non-Invasive Blood Glucose Monitoring: Wavelength and Physical Factors
Disciplines
Artificial Intelligence and Robotics | Bioimaging and Biomedical Optics | Biomedical Devices and Instrumentation | Dietetics and Clinical Nutrition | Hardware Systems
Abstract (300 words maximum)
Diabetes and metabolic diseases are among the most pressing health concerns of our time. Currently, monitoring blood glucose, a pivotal indicator of these conditions, involves the cumbersome process of frequent blood drawing or subcutaneous needle injections. Fortunately, the drive for precise, non-invasive glucose monitoring methods has intensified, leveraging technological advancements in diverse wavelength multispectral imaging. These devices adopt lasers with photodiodes or cameras to correlate blood glucose with imaging data, building safer, cheaper, and more reusable procedures for measuring blood glucose.
Nonetheless, numerous facets of this process remain unexplored. Light wavelength exerts a substantial influence on the accuracy of estimations due to the diverse interactions between wavelengths and skin. Additionally, there are several patient-related physical factors, including skin color, temperature, humidity, and skin texture, that further impact the light’s transmittance and absorption.
To examine these variables, we engineered a device that harnesses a range of light wavelengths, from 650nm to 980nm, to gather data from human subjects. Concurrently, we measured physical factors, including skin temperature, humidity, skin color, and skin texture. Our methodology starts with the collection of physical factors, followed blood glucose measurement with traditional glucometers, ending with the capture of image data from all wavelengths.
Then, we construct estimation models for each wavelength. Performance is compared through three distinct methodologies and metrics: statistical accuracy with Mean Absolute Error, clinical accuracy through correlation with Clarke Error Grids, and clinical accuracy through agreement with Bland-Altman Plots. Subsequently, we extend this evaluation across all physical factors, exploring their impact on overall model performance across our chosen metrics. After reviewing recent literature, we anticipate a weak, positive relationship between wavelength and the accuracy and precision of models. Moreover, we anticipate skin color and humidity to influence performance the most with a negative relationship between wavelength and the influence of factors.
Academic department under which the project should be listed
CCSE - Information Technology
Primary Investigator (PI) Name
Maria Valero
Exploring the Spectrum of Non-Invasive Blood Glucose Monitoring: Wavelength and Physical Factors
Diabetes and metabolic diseases are among the most pressing health concerns of our time. Currently, monitoring blood glucose, a pivotal indicator of these conditions, involves the cumbersome process of frequent blood drawing or subcutaneous needle injections. Fortunately, the drive for precise, non-invasive glucose monitoring methods has intensified, leveraging technological advancements in diverse wavelength multispectral imaging. These devices adopt lasers with photodiodes or cameras to correlate blood glucose with imaging data, building safer, cheaper, and more reusable procedures for measuring blood glucose.
Nonetheless, numerous facets of this process remain unexplored. Light wavelength exerts a substantial influence on the accuracy of estimations due to the diverse interactions between wavelengths and skin. Additionally, there are several patient-related physical factors, including skin color, temperature, humidity, and skin texture, that further impact the light’s transmittance and absorption.
To examine these variables, we engineered a device that harnesses a range of light wavelengths, from 650nm to 980nm, to gather data from human subjects. Concurrently, we measured physical factors, including skin temperature, humidity, skin color, and skin texture. Our methodology starts with the collection of physical factors, followed blood glucose measurement with traditional glucometers, ending with the capture of image data from all wavelengths.
Then, we construct estimation models for each wavelength. Performance is compared through three distinct methodologies and metrics: statistical accuracy with Mean Absolute Error, clinical accuracy through correlation with Clarke Error Grids, and clinical accuracy through agreement with Bland-Altman Plots. Subsequently, we extend this evaluation across all physical factors, exploring their impact on overall model performance across our chosen metrics. After reviewing recent literature, we anticipate a weak, positive relationship between wavelength and the accuracy and precision of models. Moreover, we anticipate skin color and humidity to influence performance the most with a negative relationship between wavelength and the influence of factors.