Novel framework based on deep learning and cloud analytics for smart patient monitoring and recommendation (SPMR)
Novel framework based on deep learning and cloud analytics for smart patient monitoring and recommendation (SPMR)
复制标题
DOI：
10.1007/s12652-020-02790-6
复制
发表时间：
2021-01-02
影响因子：
--
通讯作者：
Pawar, Mahesh
中科院分区：
文献类型：
--
作者：
Motwani, Anand;Shukla, Piyush Kumar;Pawar, Mahesh
在世界范围内，患有慢性病和生活方式疾病的患者大幅增加，影响社会和经济生活。在这项工作中，首先提出了一个广泛的调查无处不在，智能和网络化的医疗保健系统，用于监测慢性病和生活方式疾病的患者。然后，提出了智能患者监测和推荐，这是一种基于深度学习（DL）和面向云的分析的新框架。根据患者的生命体征和活动环境，通过周围辅助生活设备生成，SPMR监测和预测真实的健康状况，并呼叫辅助服务。SPMR中设计的本地智能处理（LIP）模块和面向云的分析促进了真实的时间处理和智能。LIP是基于预测DL与新的分类交叉熵（CCE）优化。在实验研究中，通过对患有慢性血压疾病的患者的案例研究收集的不平衡数据集，并预测患者的真实的健康状况。SPMR即使在没有互联网和云服务的情况下也能提供真实的预防和护理。它消除了现有工作的缺点，其中机器学习模型和相关方法被复制到本地部分。我们提出的模型与类似的和最近的模型相比，证明了有效性。我们的模型的最高精度提高范围为8- 18%。此外，F-分数平均和F-分数的紧急类分别提高了17%和36%。结果表明，即使在紧急情况下，SPMR的有效性。
In world, patients suffering from chronic and lifestyle diseases are substantially increasing that effects social as well as economic life. In this work, initially a broad survey of ubiquitous, smart and networked healthcare systems for monitoring of patients with chronic and lifestyle diseases is presented. Afterwards, Smart Patient Monitoring and Recommendation, a novel framework based on Deep Learning (DL) and Cloud oriented analytics is proposed. Based on the patients' vital signs and activity context, generated through Ambient Assisted Living devices, SPMR monitors and predicts the real health status and calls assistive services. The real time processing and intelligence facilitated by both Local Intelligent Processing (LIP) module and cloud oriented analytics devised in SPMR. LIP is based on predictive DL with novel Categorical Cross Entropy (CCE) Optimization. In the experimental study, imbalanced dataset collected through a case study on patients suffering from Chronic Blood Pressure disorder is utilized and real health status of patient is predicted. SPMR offers prevention and care in real time even in the absence of internet and cloud service. It eliminates the drawbacks of existing works, in which Machine Learning models and associated methods are copied to local portion. Our proposed model demonstrates the efficacy when compared with similar and recent models. The highest accuracy improvement with our model ranges from 8-18%. Also, F-score average and F-score for emergency class improved up to 17% and 36% respectively. The results show the effectiveness of SPMR even in case of emergencies.