自我血糖监测是糖尿病管理中非常重要的一环，有助于及时了解血糖变化，必要时调整饮食、运动及治疗方案，使糖尿病患者达到持续良好的血糖控制。无创血糖检测是最理想日常自我血糖监测方法。本文主要针对代谢热整合法无创血糖检测进行研究，发展了代谢热整合法理论，明确胰岛素在代谢热整合法中的作用，建立适用人群广的无创血糖检测模型，通过临床试验评价无创血糖仪的准确性，研究进一步提高无创血糖准确度的方法。具体工作如下：明确胰岛素在代谢热整合法中的作用，建立简化的人体葡萄糖代谢模型。代谢热整合法测量的是人体参与代谢的葡萄糖浓度，其与人体血液中葡萄糖浓度的关系受胰岛素活性的影响，可以通过线性模型来表达。将描述人体葡萄糖和胰岛素动态关系的Bergman最小模型和代谢热整合法模型结合，通过连续血糖变化曲线得到人体葡萄糖代谢率，建立人体生理参数和人体葡萄糖代谢率之间的迭代模型。建立代谢热整合法测得参数和人体代谢率之间的模型，模型预测结果的相关系数达到0.81。通过分类的方式解决胰岛素对代谢热整合法模型的影响，采用聚类分析对糖尿病人的代谢特征进行细分，并在模型中考虑用药和注射胰岛素的影响，提高了无创血糖的测试准确度。采集大量正常人和糖尿病患者的数据建立无创血糖检测模型。针对糖尿病患者的日常血糖管理中对低血糖和极高血糖事件的关注，利用SVM建立了低血糖分类模型，准确率在90%以上。通过临床试验验证无创血糖仪的准确性，预临床试验招募了254位志愿者，无创血糖仪测试结果与静脉血浆血糖结果对比，其在Parkes误差表中的结果为：A区287（58.33%）、B区194（39.43%）、C区11（2.24%）、D区0、E区0，相关系数为0.69，均方根误差为2.67mmol/L。为进一步提高无创血糖的测试准确度，进行了个体标定模型的研究。通过个体长期跟踪试验建立个体标定模型，与通用模型相比，落在Parkes误差表A区的比例、相关系数和均方根误差等指标都有所改善，得到更好的无创血糖测试结果。

Daily blood glucose monitoring in diabetic is a very important part of self-management of diabetes. It helps to assess the degree of glucose metabolism disorder, reflect the effects of hypoglycemic therapy and guide the adjustment of treatment prescribe. It enables the diabetics to achieve consistently good glycemic control. Non-invasive blood glucose testing is the ideal method for daily blood glucose monitoring. This paper focuses on the research of non-invasive blood glucose monitoring by metabolic heat conformation (MHC) method, develops the theory of MHC method, clarifies the role of insulin in MHC method, establishes a non-invasive blood glucose monitoring model suitable for a wide range of people, and evaluates non-invasive blood glucometer through clinical trials. This article also attempts to study ways to further improve the accuracy of non-invasive blood glucose monitoring. The specific work is as follows:To clarify the role of insulin in MHC method, a simplified model of human glucose metabolism was established. The MHC method measures the glucose concentration that participates in human body metabolism, and its relationship with the glucose concentration in human blood is affected by insulin activity. The relation between the two-glucose concentration can be expressed by a linear model. According to Bergman's minimum model, it is concluded that the amount of glucose metabolism in the human body can be obtained through continuous blood glucose curve. By combining the Bergman model and MHC model, the relationship between human physiological parameters and human glucose metabolism can be established. A model between the parameters measured by the MHC method and human metabolism was established. The correlation coefficient of the model prediction results reached 0.81.The effect of insulin on the MHC model was solved by classification. Cluster analysis was used to subdivide the metabolic characteristics of diabetics, and the effects of medication and insulin injection were considered in the model, which improved the accuracy of non-invasive blood glucose monitoring. Data from a large number of normal and diabetic patients were collected to establish a non-invasive blood glucose model. In the daily blood glucose management of diabetes, special attention is paid to hypoglycemia and hyperglycemia events. A hypoglycemic classification model was established by SVM, and the model accuracy was over 90%.The accuracy of the non-invasive blood glucometer was verified by clinical trials. 254 volunteers were recruited in the pre-clinical trial. The results of the non-invasive blood glucometer were compared with the venous plasma blood glucose results. The results in the Parkes error grid were: 287 (58.33%) in Zone A, 194 (39.43%) in Zone B, 11 (2.24%) in Zone C, 0 in Zone D and E. The correlation coefficient is 0.69, and the root mean square error is 2.67mmol / L. In order to further improve the accuracy of non-invasive blood glucose model, an individual calibration model was studied. The individual calibration model was established by individual long-term follow-up test. Compared with the general model, the proportions of Zone A, correlation coefficients and root mean square error were all improved.