Smart Health

Proposed in recent years, smart health has gained great attention after its debut. Undoubtedly, wearable devices will bring great convenience to the development of smart health. With the aging trend, increase of empty families and solitary elderly people, and the growth spurt of chronic disease, wearable devices have come to its spring. But due to the lack of standards in the industry, the data collected by wearable devices is unreliable, and is difficult to be accepted by doctors. Besides, there are also some problems concerning with privacy. In this paper, the advantages, disadvantages, opportunities and obstacles of wearable devices are analyzed, and some development suggestions are provided.

Data classification is one of the core technologies in the field of pattern recognition and machine learning, which is of great theoretical significance and application value. With the increasing improvement of data acquisition, storage, transmission means and the amount of data, how to extract the essential attribute data from massive data, data accurate classification has become an important research topic. Inverse n^th n order gravitational field is essentially a generalization of the n order in the physics, which can effectively describe the interaction between all the particles in the gravitational field. This paper proposes a new inverse n^th power gravitation (I-n-PG) based clustering method is proposed for data classification. Some randomly generated data samples as well as some well-known classification data sets are used for the verification of the proposed I-n-PG classifier. The experiments show that our proposed I-n-PG classifier performs very well on both of these two test sets.

Clinical pathway can be used to reduce medical cost and improve medical efficiency. Traditionally, clinical pathways are designed by experts based on their experience. However, it is time consuming and sometimes not adaptive for specific hospitals, and mining clinical pathways from historic data can be helpful. Clinical pathway naturally can be regarded as a kind of process, and process mining can be used for clinical pathway mining. However, due to the complexity and dynamic of medical behaviors, traditional process mining methods often generate spaghetti-like clinical pathways with too many nodes and edges. To reduce the number of nodes in the resulting models, we put correlated events into clinical-event-packages as new units of log event for further mining. The experiment results has shown that our approach is a good way of generating more comprehensible clinical process as well as packages with better quality according to medical practitioners.

In recent years, Data Mining techniques such as classification, clustering, association, regression etc. are widely used in healthcare field to help analyzing and predicting disease and improving the quality and efficiency of medical services. This paper presents a web-based platform for big data analysis of healthcare using Data Mining techniques. The platform consists of three main layers: Apache Spark Layer, Workflow Layer and Web Service Layer. Apache Spark Layer provides basic Apache Spark functionalities as regular Resilient Distributed Datasets (RDD) operations. Meanwhile, this layer provides a cache mechanism to maximize the use of the results as much as possible which were calculated before. Workflow Layer encapsulates a variety of nodes for Data Mining, which have different roles such as data source, algorithm model or evaluation tool. These nodes can be organized into a workflow which is a directed acyclic graph (DAG), and then it will be submitted to Apache Spark Layer to execute. And we have implemented many models including Naïve Bayes model, Decision Tree model and Logistic Regression model etc. for healthcare big data. Web Service Layer implements rich restful API including data uploading, workflow composition and analysis task submission. We also provide a web graphical interface for the user. Through the interface users can achieve efficient Data Mining without any programming which can greatly help the medical staff who don’t understand programming to diagnose the patients’ condition more accurately and efficiently.

Medical entity relation extraction is of great significance for medical text data mining and medical knowledge graph. However, medical field requires very high data accuracy rate, the current medical entity relation extraction system is difficult to achieve the required accuracy. A main technical difficulty lies in how to obtain high-precision medical data, and automatically generate annotated training sample set. In this paper, a medical entity relation automatic extraction system based on weak supervision is proposed. At first, we designed a visual annotation tool, it can automatically generate crawl scripts, crawling the medical data from the site where the entity and its attributes are Separate stored. Then, based on the acquired data structure, we propose a weakly supervised hypothesis to automatically generate positive sample training data. Finally, we use CNN model to extract medical entity relation. Experiments show that the method is feasible and accurate.

Multilingual information retrieval is very important to the persons who need to consolidate information from different languages posts and forums. However, it is not an easy job to find appropriate citations for a given context, especially for citations in different languages. In this paper, we define a novel computing framework of massive posts data and user behavior data to realize multilingual information retrieval and key technologies of multilingual information retrieval. This task is very challenging because the posts data are written in different languages and there exists a language gap when matching them. To tackle this problem, we propose the multilingual posts matching technology, source information handling technology, and personalized feed or smart feed technology. We evaluate the proposed methods based on a real dataset that contains Chinese posts data and English posts data. The results demonstrate that our proposed algorithms can outperform the conventional information retrieval scheme.

Due to increased longevity, there has been a significant growth in the aging population world over. Caring for this burgeoning class has become a pressing challenge faced by many developed and emerging countries, including the US (the aging baby boomer) and China (the reverse 4-2-1 family pyramid due to one child policy). Despite failing health, most senior citizens prefer to live independently at home and hence the focus of current healthcare technologies has shifted from traditional clinical care to “at-home” care for the senior citizens. We propose to develop SilverLink, a system that is unique in its smart and connected technologies and will offer: (1) affordable and non-invasive home-based mobile health technologies for monitoring health-related motion and daily activities; (2) advanced mobile health analytics algorithms for fall detection, health status progression monitoring, and health anomaly detection and alert; and (3) a comprehensive health activity portal for reporting user activity and health status and for engaging family members. This review discusses the SilverLink system in detail along with some of the technical challenges faced during the development of this system and future opportunities.

In this paper, we present a clinically pediatric disease prediction system based on a new efficient hybrid knowledge graph. Firstly, we automatically extract a set of triples by modeling and analyzing 1454 clinically pediatric cases, building a weighted knowledge graph based Naïve Bayes. Secondly, to extract new prediction opportunities from heterogeneous data sources, we model and analyze both classically professional pediatrics textbooks and clinical experiences of pediatric doctors respectively in order to derive prediction rules. Thirdly, we mix up those rules with the weighted knowledge graph we built to propose a new hybrid knowledge graph which can carry on both the Bayesian reasoning and the logic calculation at the same time. Fourthly, in term of that hybrid knowledge graph, we further design a new multi-label classifier based on the well-known Bayesian Ranking for the disease prediction. Finally, we implement such a hybrid knowledge graph based disease pediatric prediction system (HKDP) which uses the descriptions of the patients’ symptoms as the inputs so as to return the predicted candidates of diseases for a child. In our experiments, the comparisons with classical prediction methods prove the validity and advantage of our system, especially guaranteeing good balance between the interpretability and precision of predictions in HKDP.

Data collection is the primary prerequisite in health sector whenever an organization craving to fortify and improve its health system. In early phases of expansion and advancement in health sector data collection was paper based, which repel in analysis on daily basis because of gigantic data is available on daily basis and manual calculation and analysis is not possible. So for removing that encumbrance of analysis the manual paper based data collection is then entered into health MIS system which help a lot in analysis purpose. But that was not as faster in development and analysis phase which help analytical analysis of trends of Diseases, Equipment’s, Vaccines and punctuality of Staff of health organization also the problems of developing countries cannot afford MIS systems in rural areas thus there is a need of solution to tackle information collection and timely analysis with submissions. In this paper, the concept of Smartphone based healthcare strengthen system is shared which will reduce the burden of data collection also speed up the process of data analysis and reduce the burden. Paper work can be reduce and no need of punching the data in some health MIS system. Smart phone application also gives the concept how the monitoring staff performance and healthcare organization regularity and punctuality can be analyzed. Complete this system is developed using open source frameworks and developed application is used on smartphones having android OS. Project is implemented in Pakistan and is being used successfully in 24 districts and this project brought a drastic change in the organization in the decision making process and also for improving the poor healthcare indicators.

Biometric features of humans, which include behavioral features and biological features, have two main popular application areas: 1. Biomedical application as in human disease diagnosis and health monitoring; 2. Security application as in identity authentication.

Nowadays, biometric features have great, sometimes even better, performance in certain medical diagnosis areas in comparison to modern medical system, such as heart beats, human breath, tongue image, pulse signal, etc. By using biometric features of human body, biomedical techniques own inherent superiority over traditional medical system on accuracy, speed, sanitation, maintenance and security.

Finger-vein authentication research has been rising since 2003, and has soon become one of the most popular biometric authentication techniques. However, given its inherent advantage as a human feature and great performance in security area, yet by far little has been revealed about whether finger-vein can be used as a human health indicator. In this paper, we discuss what does it take to become a valid health indicator and proceed detailed evaluation on finger-vein features as one. The result shows finger-vein feature is a potential effective indicator for disease early detection and further real-time monitoring on physical condition of human body.

Blood pressure (BP) is an important physiological signal of human body. How to measure blood pressure is a meaningful problem for detection of human health. The most commonly used method is cuff based method. But this method can not used for continuous blood pressure measurement. For this concern, a Photoplethysmogram-based method for continuous blood pressure measurement was presented. Many researches have found that there are some relations exist between some Photoplethysmogram (PPG) signal features and human blood pressure. We use an artificial neural network model and a cluster method to make some estimation on human blood pressure based on Photoplethysmogram signal, and our result shows that this method can be used for noninvasive continuous blood pressure measurement in future.

Effective features extracted from Photoplethysmogram (PPG) are the central for estimating accurately blood pressure (BP). To make extracted features have a strong correlation with real blood pressure, a model based on feature fusion is presented to evaluate blood pressure. To divide pulse wave into two types of dicrotic wave and non-dicrotic wave, different types of waveforms use different extracted features, and wavelet transform is used to remove the noise from the extracted features. Linear regression model and neural network are evaluation models, and Matlab system identification toolboxes are used to recognize the model parameters. The experiment results have shown that extracted features have a correlation with systolic pressure (SP) and diastolic pressure (DP). The value of blood pressure can be calculated based on features extracted from PPG. What’s more, the accuracy of the fusion feature model is improved compared with the traditional method only by using one type of extracted feature method for all the pulse waveforms.

In “China Cardiovascular Disease Report 2015”, research shows that mortality of Acute Myocardial Infarction (AMI) was rapidly increased since 2005. The mortality in 2014 was 123.92/lakh, which was 4.4 times higher than in 2002. Cardiovascular disease is ranked No. 1 in cause of death in China right now, in both rural and urban areas. This paper presents a medical information sharing platform based on mobile Internet, cloud computing and big data mining. It is designed to support the PB-level data management and analysis, and millions of concurrent instant messaging. The platform has the following functions: intelligent transportation decision support based on FMC-D time, built-in medical communication unit, built-in medical information sharing unit and quality control system of PCI hospital interventional images. The platform is divided into two parts - medical unit terminals (including EMS terminal, non-PCI hospital terminal and PCI hospital terminal) and cloud computing server, in which data is exchanged via 3G/4G wireless networks. The system has the following characteristics: (1) Timeline, which is a collection of key nodes that describe the AMI patient care process, (2) Smart recommendation technology, for example recommending hospitals based on the distance, medical care ability, idle resource. (3) Capacity to support, such as large number of concurrent collaborative treatment process among multiple PCI hospitals, multi-non-PCI medical institutions, and multi-EMS institutions, as well as the PB level data which are generated in the process.

Pharmacovigilance is the research related to the detection, assessment, understanding, and prevention of adverse drug events. Despite the research efforts in pharmacovigilance in recent year, current approaches are insufficient in detecting adverse drug reaction (ADR) signals timely across different datasets. In this study, we develop an integrated and high-performance AZ Pharm MetaAlert framework for efficient and accurate post-approval pharmacovigilance. Our approach extracts adverse drug events from patient social media, electronic health records, and FDA’s Adverse Event Reporting System (FAERS) and integrates ADR signals with stacking and bagging methods. Experiment results show that our approach achieves 71% in precision, 90% in recall, and 80% in f-measure for ADR signal detection and significantly outperforms the traditional signal detection methods.

Chinese medicine is increasingly being used with Western medicine in practice, especially for treatment of chronic diseases. In this integrative medicine process, it is necessary to understand the interactions between Chinese and Western medicine to reduce adverse events. However, compared to Western medicine, there are limited studies that summarize findings on Chinese medicine interactions and effectively present such findings to practitioners. Built upon the MedC literature analysis system, this paper proposes a Chinese medicine analysis and prediction framework that can effectively present Chinese medicine interactions and connect them with the clinical evidence documented in literature. The system can support Chinese medicine scholars and facilitate research on integrative Chinese medicine.

Cataract is a dulling or clouding of the lens inside the eye. Which is one of the most common diseases that might cause blindness. Considering the damage impact of cataract, we propose to use retinal vascular information for automatic cataract detection, which based on the classification of retinal image. This method focus on the preprocessing step of retinal image. Firstly, we use the maximum entropy method to enhance the contract level of fundus image. Next, in order to collect vessel information based on the Kirsh template of multi-layer filter is used. Last, adaptive weighted median filter has proposed to reduce the noise of the image. Then, according to the retinal blood vessel image, we extracted wavelet features, texture features for cataract classification. For each set of features, SVMs (support vector machines) is used for cataract classification. Finally, cataract image classified into normal, slight, medium or severe four-class. Through comparing the result of classification, three of four classes obtains the better accuracy than former. At the same time, the time that spend on feature extract is greatly reduced. The result demonstrate that our research on classification system is effective and has practical value.

Machine Translation is increasingly applied to medical cross-lingual data processing. In order to evaluate the quality of machine translation, automatic evaluation approaches like BLEU and NIST, most of which are n-gram based metrics, are widely used besides costly human evaluation. Current evaluation approaches merely make surface linguistic comparisons between the candidate and reference translations. Furthermore the domain features such medical terms, dependent and cohesive relations in and among sentences in medical documents should be taken into account when evaluating translations. However severe noises are imported into the procedure when faulty machine translations are parsed using the current syntactic parsers. Therefore using the noisy parsing results to compare with references affects the improvement of evaluation even though the deep processing is incorporated. To lessen noises as well as grasp the main meaning of a sentence, the paper proposes to extract the dependency n-grams only based on dependency parsing of reference translations. Dependency n-grams are stemmed and extended according to linguistic rules and then viewed as the key points for quality evaluation. The score of candidate translation is computed according to the count of dependency n-grams loose matching. Also the penalty of short translation and the clip count of the highest frequency of dependency n-grams are incorporated in the final score of the candidate. Experiments on our translation datasets show the evaluation based on dependency n-grams significantly outperforms the metric of BLEU and NSIT. This approach is also significantly better than the related research which employs dependency relation parsing in evaluation.

With the development of electronic healthcare, more and more medical institutions begin to use the information system to manage their patient’s health records as well as other healthcare data. Electronic medical records (EMR) contain the patient’s personal information, medical history, clinical examination, treatment process, and other information, which have large research value. Today, enormous number of electronic medical records accumulated through the hospital information system all over the world. Analyzing these EMRs can effectively assist doctors in clinical decision-making, provide data support for clinical research as well as personalized healthcare service for patients. This paper presents a EMR similarity computation system. The system accepts EMRs collected from hospitals as input, go through a series of process, and eventually calculates the similarity of any two EMRs. An diseases classification experiment was designed to illustrate the effectiveness of the method. This system lays the foundation for further analysis of electronic medical records.

Objective: This study aims to investigate the self-regulation of the autonomic nervous system following the cognitive stress tests after Heart Rate Variability (HRV) biofeedback therapy in post-stroke depression (PSD) patients. Methods: Twenty-four patients with PSD were randomly divided into feedback and control groups. Feedback patients were given HRV biofeedback therapy, while the control patients only received relaxation therapy without feedback signal. HRV parameters were tracked during the cognitive stress test in quiet baseline state, cognitive stress state, and resting state, to compare the therapeutic effects of the two groups, before and after treatment. Results: Under the stress conditions, LF of both groups increased, but there were significant differences in the increasing rate (P = 0.02): LF of the feedback group increased slowly, while that of the control group increased rapidly. HRs of both groups increased during the second cognitive test, with HR increasing slowly in the feedback group and faster in the control group (P = 0.05). After rest, the HR of the feedback group decreased significantly faster than that of the control group (P = 0.05). HF of the both group increased during the stress test but showed no significant difference. Conclusion: In this paper, we show that during the cognitive stress test, patients that have received HRV biofeedback therapy can achieve a dynamic balance between sympathetic and parasympathetic nerves by reducing sympathetic sensibility, which improved patients’ adaptive capacity to cope with their internal physiological environment and external environmental pressures.

The medical image needs to be encrypted before storing in cloud platform to protect against leaking the personal private information of medical image. And we expect the encrypted medical image can be retrieved automatically in cloud computing platform, but traditional medical image retrieval is based on the visual feature, which is difficult to identify with the naked eye after encryption. In this paper, we propose an algorithm with strong robustness—medical image retrieval algorithm based on DFT encryption domain. We encrypt the image in frequency domain and extract its feature vector to establish a feature database, and then automatically compute the NC (Normalized Cross Correlation Coefficient, NC) between the feature vector of the image to be retrieved and each one stored in the feature database. Finally, the corresponding encrypted image with the greatest value of NC is returned. The experimental results show that this algorithm has ideal ability to resist the conventional attack, such as interference of Gaussian noise, JPEG compressing and median filtering, and geometric attack, such as rotation, scaling, translation, cutting.

Most of the existing robust watermarking schemes were designed to embed the watermark information into the plaintext images, which leads to a latent risk of exposing information and are vulnerable to unauthorized access. In addition, the robustness of watermarking in the encrypted domain is another issue that should be taken into account. Based on Discrete Fourier Transform (DFT) and Logistic chaotic map, we proposed a robust zero-watermarking algorithm in the DFT encrypted domain, which achieves good safety in the protection of both watermark information and the original image itself. Firstly, we encrypt the watermark and the original medical image in DFT encrypted domain. Then, the DFT is performed on the encrypted medical image to acquire the feature vector. In watermark embedding and extraction phase, zero-watermarking technique is utilized to ensure integrity of medical image. Experimental results demonstrate good robustness against both common attacks and geometric distortions.

Objective: To explore the effect of a mobile Internet regional cooperative rescue system on reperfusion efficiency and prognosis of patients with ST segment elevation myocardial infarction (STEMI). Methods: The patients were divided into two groups: the regional transported group (experimental group) and routine transported group (control group) according to whether the first medical contact (FMC) unit was equipped with a regional cooperative rescue system. Every time point during transport process, the time to peak and peak value of cardiac troponin I (cTNI), the rate of heart failure or cardiac death during hospitalization, the value of ejection fractions (EF) measured in 24 h, and indicators of health economics(total hospital charges, days) were observed. Results: The difference of time delay between two groups were all statistically significant P(<0.05); the peak time of cTnI was earlier in experimental group than control group ((14.2 ± 3.4) h vs. (16.3 ± 4.6) h, P < 0.01), the peak value of cTnI was decreased in experimental group compared with the control group ((9.3 ± 2.9) ng/ml vs. (12.3 ± 3.2) ng/ml, P < 0.01); values of EF within 24 h after admission were significantly lower in control group than experimental group (t = 2.37, P < 0.05); in-hospital heart failure rate of experimental group was less than that of the control group 2χ(= 4.46, P = 0.03); cardiac mortality rate of experimental group was less than that of the control group, and it was not significant between the two groups (χ 2 = 0.19, P = 0.66); total cost in hospital, total hospital stay were significantly decreased in experimental group compared with the control group ((56711 ± 12083) yuan vs. (65847 ± 14691) yuan, P < 0.01; (6.35 ± 3.68)d vs. the day (8.64 ± 5.19)d, P = 0.01). Conclusions: Regional cooperative rescue system could significantly shorten the time delay of patients with STEMI, improve heart function in acute stage and reduce the time and cost in hospital.

Cloud computing platform is not a fully trusted third party, which may leak the patient’s personal information when we store medical image, so we need to encrypt medical image. Meanwhile, in order to help doctors who can find out historical cases from the medical image database which are similar to the current diagnostic image to make more accurate diagnosis and treatment, this paper proposes an robust algorithm based on 3D DFT for encrypted medical image retrieval. At first, we extract feature vector of 3D encrypted image and establish features vector database. Next, the NC (Normalized Cross Correlation Coefficient) between the feature vector of query medical image and each one in the features vector database is computed automatically. Finally, the corresponding encrypted image with the highest NC value is returned. The results show that this algorithm has strong robustness against common attacks and geometric attacks.