基于残差注意力卷积神经网络的糖尿病无创检测研究  

Non-invasive detection of diabetes based on residual attention convolutional neural network

作　　者：张冰[1] 孙蓓[1] 李明珍[1] 杨艳辉 郭立川[1] 周冰 宋振强[1] Zhang Bing;Sun Bei;Li Mingzhen;Yang Yanhui;Guo Lichuan;Zhou Bing;Song Zhenqiang(Zhu Xianyi Memorial Hospital of Tianjin Medical University,Tianjin Institute of Endocrinology,Key Laboratory of Hormones and Development of National Health Commission,Tianjin Key Laboratory of Metabolic Diseases,Tianjin 300134,China)

机构地区：[1]天津医科大学朱宪彝纪念医院,天津市内分泌研究所,国家卫健委激素与发育重点实验室,天津市代谢性疾病重点实验室,天津300134

出　　处：《中国数字医学》2022年第6期87-92,共6页China Digital Medicine

基　　金：天津市卫生信息学会科研项目(TJHIA-2020-002);天津医科大学医院管理创新研究项目(2020YG20)。

摘　　要：目的:降低糖尿病大范围筛查的医疗成本,减轻血糖检测对患者心身的伤害,同时为高风险糖尿病人群提供一种无创、准确、高效、经济的糖尿病检测方法。方法:采用残差注意力卷积神经网络对糖尿病受试者面部图像进行有监督的机器学习,预测受试者未来糖尿病的发病风险;为评估该方法的应用效果,本实验招募了384例糖尿病受试者和137例血糖正常的健康志愿者,比较残差注意力网络与其他卷积神经网络的糖尿病无创检测性能。结果:采用56层残差注意力网络构建的糖尿病无创检测模型在实验中表现出的预测能力最强,准确率高达94.28%,特异性为92.94%,F1值达95.88%。结论:该预测模型检测方法耗时短、成本低且支持大范围筛查及远程诊疗,具有较强的糖尿病检测能力。Objective To reduce the medical cost of large-scale diabetes screening,reduce the physical and mental harm of blood glucose testing to patients,and provide a non-invasive,accurate,efficient and economic diabetes detection method for high-risk diabetes patients.Methods The convolutional neural network of residual attention was used for supervised machine learning of face images of the subjects with diabetes to predict the risk of diabetes in the future.To evaluate the application effect of the proposed method,384 subjects with diabetes and 137 healthy volunteers with normal blood glucose were recruited in this experiment to compare the performance of residual attention network and other convolutional neural networks for nonivasive detection of diadetes.Results The non-invasive diabetes detection model constructed by the 56-layer residual attention network showed the strongest prediction ability in the experiment,with an accuracy of 94.28%,specificity of 92.94%and F1-score of 95.88%.Conclusion This predictive model detection method is time-consuming,low-cost,and supports large-scale screening and remote diagnosis and treatment,which has a strong ability to detect diabetes.

关 键 词：糖尿病无创检测 卷积神经网络 AI辅助诊疗 

分 类 号：R319[医药卫生—基础医学]