Acquisition and analysis of real-world sensing data to aid prediction, diagnosis and monitoring of cardiovascular disease

研究由CiraVegna教授开发的软件收集的数据，并将其发布给Chico教授的患者。在接下来的两年里，该系统将被数百名患者使用。该系统使用移动应用程序、一组蓝牙信标和智能手表，在室内和室外全天候监控患者。分析现有系统收集的数据，识别心血管疾病的特征。为医生定义一个数据分析平台，以分析个人和群体层面上的患者数据。数据分析平台将支持心脏病专家根据收集、清理和建模的数据得出诊断。该项目的关键是开发算法，在最大限度提高传感质量的同时最大限度地减少电池功耗，以长期收集高质量的连续数据；数据质量是在长期收集过程中最大化分类准确性的主要要求，以及收集技术的可用性。该方法将基于(I)强化学习来优化传感策略和电池影响，(Ii)在很大程度上无监督和半监督的机器学习分类算法来识别疾病的特征以支持诊断；迁移学习将被用来创建针对个人用户行为而量身定做的策略，同时跨用户学习，以及(Iii)自适应数据和可视化分析方法。

Work on the data collected by software developed by Professor Ciravegna and under release to patients of Prof. Chico. The system will be used by hundreds of patients over the next two years. The system uses a mobile app, a set of bluetooth beacons and a smart watch to monitor patients 24/7 while indoors and outdoors.Analyse data collected by the existing system and identify signatures of cardiovascular diseases. Define a data analytics platform for doctors to analyse patients' data at an individual and population level. The data analytics platform will support cardiologists in reaching a diagnosis based on the collected, cleaned and modelled data.Key to the project is development of algorithms that maximise sensing quality while minimising battery power consumption to collect high quality continuous data over long periods; data quality is a major requirement for maximising classification accuracy over long term collection, as well as usability of the collection technology.The approach will be based on (i) reinforcement learning to optimise sensing strategies and battery impact, (ii) largely unsupervised and semi-supervised machine learning classification algorithms to identify signatures of disease to support diagnosis; transfer learning will be used to create strategies tailored to individual users' behaviour while learning across users and (iii) adaptive data and visualisation analytics methods.