基于人工智能算法的无创血糖检测仪的临床准确性研究

Study on the clinical accuracy of non-invasive blood glucose detecting device-based on AI algorithm

目的:对比分析基于人工智能(AI)算法的NeoGlu01无创血糖仪与有创血糖仪检测末梢血糖的准确性。方法:招募2022年12月至2023年2月北京地区社区的96名糖尿病患者,采用德国罗氏智航有创血糖仪对患者末梢血糖进行检测,同时用NeoGlu01无创血糖仪采集手指生物信号,获得AI计算血糖值。采用Pearson相关性分析两种仪器所测血糖值。结果:有创血糖仪检测96名糖尿病患者末梢血糖值和NeoGlu01无创血糖仪AI计算值与标定的AI计算值均具有很好的一致性(R^(2)=0.978、0.882,P<0.05);有创血糖仪检测糖尿病患者末梢血糖值与NeoGlu01无创血糖仪的AI计算值均在6.2~16.7 mmol/L之间,差异无统计学意义(P>0.05);以末梢血糖值为参照的Parkes误差网格分析的2次结果显示,无创血糖仪的AI计算值总体准确性指标平均绝对相对误差(MARD)值分别为11.05%和11.44%,准确率及可重复性良好。结论:NeoGlu01无创血糖仪预测血糖值结果与有创血糖仪检测末梢血糖值具有良好的一致性和准确性。

Objective:To compare and analyze the accuracy of the invasive glucose detecting device and the non-invasive glucose device(NeoGlu01)based on artificial intelligence(AI)algorithm in detecting peripheral blood glucose.Methods:A total of 96 diabetic voluntary patients were recruited from the community of Beijing district from December 2022 to March 2023.The Roche ACCU-CHECK®Guide invasive glucose device was used to detect the peripheral blood glucose of volunteers,and the NeoGlu 01 type of non-invasive glucose device was used to collect the signal of the finger,so as to obtain the blood glucose value that was calculated by AI algorithm at the same time.Pearson correlation analysis was adopted to analyze the detected blood glucose values of the two kinds of devices.Results:Both the peripheral blood glucose values of invasive glucose detecting device and the calculated AI values of NeoGlu01 type of non-invasive glucose device for 96 diabetic patients had favorable consistency with calibration value(R^(2)=0.978,0.882,P<0.05),respectively.Both the detected values of peripheral blood glucose of invasive glucose detecting device and the calculated AI values of NeoGlu01 type of non-invasive glucose device were between 6.2 and 16.7 mmol/L for 96 diabetic patients,and the difference was no significant(P>0.05).Parkes error grid analysis,that referred to the values of peripheral blood glucose,indicated the two times of the results of the values of mean absolute relative difference(MARD)of total accuracy indicator of calculated AI value of non-invasive glucose device were respectively 11.05%and 11.44%,and both the accuracy rate and repeatability of that were favorable.Conclusion:The predictive blood glucose value of NeoGlu01 non-invasive blood glucose device has favorable consistency and accuracy with the detected value of invasive glucose detecting device in detecting peripheral blood glucose.