基于健康流数据的健康演进趋势识别与实时状态评测关键技术研究

With increasing advances in modern information and communication technologies, it has become not only technically available but also financially viable to monitor and collect individuals’ health data by using low-cost miniature biosensors in daily living settings. The use of health monitoring techniques is beneficial for tracking changes and detecting anomalies of individuals’ health for older adults and sub-health population. Based on an in-depth understanding of the inherent correlations between changes in health condition and variations in health data, this project is proposed to explore and investigate technologies for recognizing change patterns and evaluating real-time status of individuals’ health from his/her health stream data. Specifically, this project aims at technologies and methods for: 1) detecting progressive changes in health stream data and recognizing patterns of these changes; 2) association relationship between different change patterns from individuals’ different health data streams; and 3) personalized online system for real-time evaluation of individuals’ health condition. With the problems above being handled, a series of research conclusions can be achieved including:. 1) An adaptive detection method for finding progressive changes in health stream data and for recognizing patterns of these changes;. 2) A personalized model that is built on association rules extracted from different change patterns, expert knowledge and personal profits for evaluating real-time status;. 3) An automatic online system that can be designed and implemented by integrating all the technologies and methods.. The proposed technologies, methods and system will play an important role in persistent monitoring of progressive health changes and early detection of health problems for both elderly individuals and sub-health population.

随着现代信息通信技术的进步，应用微型生理传感器在日常生活环境下监测并收集人类健康数据已经成为可能，对老年人及亚健康群体实时掌握健康状态变化及发现潜在的健康问题奠定了实现的技术基础。从健康状态迁移与健康参数渐进性改变之间的固有关联入手，以传感器捕获的实时健康流数据为研究对象，本项目提出并研究基于健康流数据的健康演进模式识别与实时状态评测关键技术。涉及的研究研究包括：1）健康流数据渐变的检测及其演进模式的识别；2）多源健康流数据演进模式之间关联的提取；3）健康演进模式依赖的个性化实时评测模型及检测算法。项目的主要研究目标是：1）提出面向健康流数据渐进性改变的检测技术及演进模式的自适应识别技术；2）提出基于演进模式的个性化健康评测模型及实时检测算法；3）基于关键技术创新成果，设计并开发健康状态实时评测系统。这些技术的研究并应用实现，对老年人及亚健康群体及早发现潜在的健康问题起到积极的支撑作用。

构建面向健康流数据的实时状态识别模型、关联规则提取算法及个性化安全辅助服务，对人口健康、智能辅助及疾病诊治具有重要研究意义和广泛应用价值，以行为健康数据和生理健康数据为研究对象，本项目开展了如下研究内容：. 一是提出了面向异常状态识别的数据流不确定性检测方法和算法，包括GPS轨迹序列、红外传感器移动事件序列、医学诊断文本及SPECT图像。. 二是提出了面向异常行为检测的个性化识别模型，包括基于GPS轨迹的个性化安全数字围栏及异常轨迹检测算法、基于AIR传感器的异常移动序列检测算法、基于深度学习的SPECT核医学图像分类模型。GPS异常轨迹检测的AUC值不低于0.995，SPECT图像分类的最高AUC值超过0.889。. 三是提出了面向模式发现的数据流挖掘方法，包括APP关联规划的挖掘和疾病及其表征关联关系的发现。挖掘结果与真实情况的吻合度超过87%，具有一定的可靠性。. 四是构建了基础研究平台，原型实现的自动检测系统，，积累了一定的研究数据，包括穿戴式生理感知数据、核医学检查记录和移动轨迹数据。特别是核医学图像数据库由超过5000个病例的骨现象和肺灌注图像构成。. 这些关键技术和方法的提出，为进一步研究健康流数据关联模式发现和实时状态识别奠定了基础，也为个性化辅助服务和自动医疗辅助的投入使用提供了基础技术方案。