A Ranking-based Weakly Supervised Learning model for telemonitoring of Parkinson’s disease
A Ranking-based Weakly Supervised Learning model for telemonitoring of Parkinson’s disease
复制标题
DOI：
10.1080/24725579.2022.2091065
复制
发表时间：
2022-06
影响因子：
--
通讯作者：
Dhari F. Alenezi;Hang Shi;Jing Li
中科院分区：
文献类型：
--
作者：
Dhari F. Alenezi;Hang Shi;Jing Li
摘要远程监护是利用电子设备对病人进行远程监护。需要一个模型来将患者的移动终端收集的数据转换为疾病严重程度评估的预测评分。标记样本很少，这使得训练监督学习模型变得困难。另一方面，有大量的样本没有精确的标签，但其相对排名可以从领域知识。我们提出了一个基于排名的弱监督学习（RWSL）模型来整合这两种类型的数据。我们应用RWSL预测帕金森病的严重程度的基础上移动收集的敲击活动数据的患者。RWSL实现了高预测精度，优于竞争方法。
Abstract Telemonitoring is the use of electronic devices to monitor patients remotely. A model is needed to translate the data collected by a patient’s mobile device into a predicted score for disease severity assessment. Labeled samples are scarce, which makes it difficult to train a supervised learning model. On the other hand, there is an abundance of samples without precise labels but whose relative rank can be known from domain knowledge. We propose a Ranking-based Weakly Supervised Learning (RWSL) model to integrate both types of data. We apply RWSL to predict Parkinson’s disease severity based on mobile-collected tapping activity data of patients. RWSL achieves high predictive accuracy and outperforms competing methods.