Automated processing of continuous glucose monitor (CGM) data to study onset of diabetes

详细信息   

• 作者：Khatri ; Aman
• 学历：Master
• 年：2015
• 关键词：Applied sciences ; Continuous glucose monitoring ; Dat
• 导师：Lingasubramanian,Karthikeyan
• 毕业院校：The University of Alabama
• Department：Engineering
• 专业：Biomedical engineering;Electrical engineering;Computer science
• ISBN：9781321701814
• CBH：1587058
• Country：USA
• 语种：English
• FileSize：4140875
• Pages：75

文摘

Diabetes is a long term condition that causes high levels of blood glucose,and it is necessary to get a complete picture of glucose levels which can lead to better treatment decision and better glucose control. The advent of Continuous Glucose Monitoring (CGM) is helping researches to track the blood glucose levels continuously and to understand the effects of impaired glucose levels on human body which in turn can lead to better treatment of diabetes. Current CGM systems process enormous amounts of data and have limitations in regards to data accuracy,precision,and reliability of raw glucose data. The inaccuracy in data also produces larger relative error in the estimates of glycemic variability than in the estimates of mean glucose and other related multiple measure of variability and multiple clinical end points. One of the objectives of the study is to automate the cleaning process of raw CGM data so as to replace the manual approach which very methodical but time consuming. Additionally,this study presents an automated procedure that predicts meal consumption and provides the intake time and glycemic load. The proposed study is performed on data generated from three categories of participants,normal weight,over weight and obese. The success of the automated model to inspect and clean data is based on comparison of the datasets resulting from automated processing to those resulting from manual inspection and cleaning. These results obtained from automated protocol are found to be in agreement with the results obtained from manual inspection. Additionally,the average percent of correctly detected meals for normal weight participants is 82.777,the average percent of correctly detected meals for overweight participants is 85.933,and the average percent of correctly detected meals for obese participants is 80.589. The overall success rate for determining the meal times is 83.099. We have also shown some success in determining more specific information about the nutrition values of meal like glycemic load. The ability of the proposed model to predict meal intake is essential to understand the onset of diabetes. This proposed study results in a flexible platform which can facilitate important clinical studies on diabetes and possibly on other biological issues related to blood glucose.