Intelligent Remote Monitoring and Management system for Type1 Diabetes

The work presented in this thesis focuses on developing a telemedicine system for better management of type1 diabetes in children and teenagers. The research and development of the system is motivated by the inadequate communication in the current system of management of the disease, which results in non-compliance of patients following the regimen. This non-compliance generally results in uncontrolled blood glucose levels, which can result in hypoglycaemia, hyperglycaemia and later life health complications. This further results in an increase in health care costs. In this context, the thesis presents a novel end-to-end, low cost telemedicine system, WithCare+, developed in close collaboration between the University of Sheffield (Electronics & Electrical Engineering) and Sheffield Children’s Hospital. The system was developed to address the challenges of implementing modern telemedicine in type 1 diabetic care with particular relevance to National Health Service children’s clinics in the United Kingdom, by adopting a holistic care driven approach (involving all stakeholders) based on specific key enabler technologies such as low cost and reconfigurable design. However, one of the major issues with current telemedicine system is non-compliance of the patients due to invasive procedure of the glucose measurement which could be clearly addressed by non-invasive method of glucose measurement. Hence, the thesis also makes a contribution towards non-invasive glucose measurement using Near Infrared spectroscopy in terms of addressing the calibration challenge; two methods are proposed to improve the calibration of the Near Infrared instrument. The first method combines locally weighted regression and partial least square regression and the second method combines digital band pass filtering with support vector regression. The efficacy of the proposed methods is validated in experiments carried out in a non-controlled environment and the results obtained demonstrate that the proposed methods improved the performance of the calibration model in comparison to traditional calibration techniques such as Principal Component Regression and Partial Least Squares regression.