State and parameter estimation in dynamical metabolic models for personalized model-based management of diabetes

在本研究项目中，开发了一种基于模型的糖尿病患者个性化诊断方法，该方法基于葡萄糖-胰岛素代谢模型来估计血糖的进展。应该使用已建立的血糖测量和新的传感器来识别该模型。接着分析了该系统的状态和参数的能控性和能观性。为了获得优于通用模型方法的更好的预测，所需的参数集应适用于每个人。代谢系统的异常变化可能发生在糖尿病中，这些参数的变化可以单独表征这些参数的变化。所获得的方法可以帮助使用者进行诊断和治疗。糖尿病是一种慢性代谢性疾病。它是由于身体不能产生和/或使用胰岛素造成的。无论哪种特定类型的糖尿病，人们都需要终生胰岛素治疗。糖尿病的治疗倾向于通过人工反馈控制来支持血糖、胰岛素稳态的紊乱的生理控制。作为传感器，患者可以测量血液或皮下组织中的葡萄糖水平。作为促进器，可以注射胰岛素。反馈控制是患者在接受临床指导后自行建立的。代谢过程就像技术系统一样，可以用微分方程式来描述。因此，可以假设糖尿病的诊断和治疗得益于基于模型的过程。由于不同人之间代谢行为的巨大差异，以前的系统识别方法只考虑患者的个体情况。此外，好的标准模型只对从集体中收集的平均值描述这些过程。事实上，医生们正试图通过开发“个性化医疗”来解决这些问题。这意味着除了细粒度模型之外，还需要基于模型的方法来设计一个高效的、个性化的模型识别系统，该系统需要较少的测量点。重点放在新型的连续测量传感器上，这种传感器可以第一次捕获波形中的动态。应该可以获得比只考虑严格限制更好的诊断信息。这可以通过根据患者的当前情况调整胰岛素治疗或优化血糖测量的数量来实现。考虑到饮食或体力活动等干扰因素，对未来血糖水平的预测是可能的。

In this research project a model-based procedure for personalized diagnosis of patients with diabetes mellitus is developed which estimates progress of blood glucose based on a model of glucose insulin metabolism. The model should be identified using established blood glucose measurements and novel sensors. Analysis of controllability and observability of states and parameters of that system follows. Required parameter sets shall be fitted to each individual in order to obtain better predictions superior to generic model approaches. Abnormal changes in the metabolic system, which could arise in diabetes mellitus, can be individually characterized by a change of these parameters. The gained methods can assist users in their diagnosis and treatment.Diabetes mellitus is a chronic metabolic disease. It results from the body's inability to produce and/ or use insulin. Regardless of the specific type of diabetes affected people need a lifelong insulin therapy.Therapy of diabetes tends to support the disturbed physiologic control of glucose insulin homeostasis by an artificial feedback control. As sensor a patient can measure glucose levels in blood or in subcutaneous tissue. As actuator insulin can be injected. Feedback control is established by the patient oneself after clinically instruction.Metabolic processes are, like technical systems, describable by the use of differential equations. Therefore, it can be assumed that diagnosis and treatment of diabetes mellitus benefits from a model-based proceeding.Previous approaches to system identification consider individual condition of patients only insufficient, because of huge variabilities in metabolic behavior between different persons. Also good standard models describe these processes only for mean values gathered from collectives. Actually physicians try to tackle these problems by developing 'personalized medicine'. That means for a model-based approach besides a fine-grained model, to design an efficient and personalized model identification system, which needs few measurement points. The focus is on novel, continuously measuring sensors which make it possible to capture the dynamics in the waveforms for the first time.It should be possible to derive diagnostic information that are much better in quality superior to considering only rigid limits.This can be done by adapting the insulin therapy based on the current condition of the patient or to optimize number of glucose measurements. Taking into account disturbances such as meals or physical activity a prediction of future blood glucose levels is possible.