RF-IDH: An intelligent fall detection system for hemodialysis patients via COTS RFID

https://doi.org/10.1016/j.future.2020.06.047Get rights and content

Highlights

  • We have taken the lead in proposing an intelligent fall detection system via RF signals for blood purification center.

  • We design an RFE algorithm based on the hemodialysis patient safety process model which outperforms the compared methods.

  • We implemented our system on COTS RFID and comprehensively evaluated the system performance.

Abstract

Unhealthy habits lead to a growing population of hemodialysis patients. The single treatment of hemodialysis is about four hours long. Therefore, patients usually go to the toilet during treatment and need to be checked for safety. However, existing fall detection techniques are often limited by factors such as privacy, signal interference, and the like. In this paper, we propose RF-IDH tackle the above issues, a dedicated system for detecting falls caused by complications in hemodialysis patients using RF signal. In RF-IDH, after collecting the signal, we process the collected data by three functional module clusters, namely signal preprocessing, residual feature extraction, hemodialysis patient’s fall detection, all of which are well-designed to achieve high performance in patient’s fall detection. In particular, we design a residual feature extraction (RFE) algorithm based on the hemodialysis patient safety process model, and the fall detection of hemodialysis patients is treated as a machine learning problem where four classification models are built via learning residual feature space. We implement our system on commercial off-the-shelf RFID devices and compared the evaluation metrics of four different methods in terms of system performance, efficiency, robustness, and latency. The evaluation results show that our proposed RF-IDH that optimizes the 2NN–RFE method achieves superior performance compared to other methods.

Introduction

With the development of society, excessive work pressure, bad eating habits, a severely polluted environment, and many other bad habits make more and more people accept various tests for their health. Once high blood pressure, high blood fat, diabetes, obesity, and other diseases are found, it is also easy to cause kidney damage. Like this, there are many reasons for chronic kidney diseases (CKD), such as urinating, drinking too little, eating too salty, and drug abuse, etc. At present, various chronic kidney diseases will eventually progress to end-stage renal disease (ESRD). Patients may develop severe uremia symptoms and rely on renal replacement therapy to maintain normal life needs. For patients with end-stage renal disease, the globally accepted treatment is renal replacement therapy, including hemodialysis (HD) and renal transplantation (RTx). However, kidney transplants are very expensive and difficult to find a suitable kidney source. Therefore, hemodialysis as an alternative treatment is an effective means for most people to treat kidney diseases and certain immune metabolism and nervous system diseases. According to the survey data of 2016, 2.96 million people worldwide received dialysis treatment, an increase of 5.7% compared with 2015. Forty-two percent of hemodialysis patients worldwide come from the Asia-Pacific region, nearly 500 thousand Chinese people among them, 25 percent from Europe/Middle East and 23 percent from North America. It is believed that the number of new hemodialysis patients in emerging markets such as the Asia Pacific will maintain the growth rate of 6%7%.

Hemodialysis can indeed relieve symptoms and prolong the survival time of patients. However, the treatment process of hemodialysis is not smooth. The treatment frequency of hemodialysis increases with the time of illness, usually three times a week. The time of single blood purification treatment is long, except for the first two adaptation periods, all reach four to five hours. In addition, there are many precautions and possible complications during the treatment of hemodialysis patients. It is gratifying to note that hemodialysis patients are usually able to take care of themselves, but this also entails many potential safety risks. Intradialytic hypotension (IDH) and muscle spasm are the two most common complications of hemodialysis. The main reason is that hemodialysis is a blood purification technology based on the principle of membrane balance. It allows blood to pass through membranes with many small holes (or channels, medically known as semipermeable membranes), which allow small molecules to pass through, while large molecules cannot. In addition, the contact between semipermeable membranes and dialysates containing certain chemicals can achieve the purpose of removing harmful substances from the body and replenishing substances needed by the body. However, this hemodialysis mechanism often leads to excessive dehydration or excessive dehydration speed, which leads to the decline of blood volume or muscle spasm. When the patient’s body position changes, hemodialysis hypotension or muscle spasm is prone to occur and a typical scene of the real treatment period is that hemodialysis patients are more likely to go to the toilet because a single blood purification treatment lasts for up to four or five hours. After the patients go to the toilet, they suddenly change from a sitting position to standing position. There is a certain probability of hemodialysis hypotension or muscle spasm caused by body position changes, which may lead to the risk of falling. Once the patient falls down, the medical staff must respond in time, otherwise, the patient will be in danger of serious insufficiency of vital organ blood supply. Therefore, taking into account all the factors of the above analysis, the hospital blood purification center does need to deploy a high-performance intelligent fall detection system that meets the privacy requirements of hemodialysis patients in the toilet.

In this paper, we propose an RF-IDH system, an intelligent fall detection system for hemodialysis patients with a complication of hypotension or muscle spasm based on RF signal, to meet the actual needs of hospital blood purification centers. RF-IDH uses a commercial off-the-shelf (COTS) RFID reader with one antenna and three EPCglobal C1G2 standard passive tags attached to a badge with adjustable lanyard length. Each tag is deployed horizontally in parallel to achieve multi-speed sample acquisition. In addition, multi-tag deployment also plays the role of redundant backup in the real test environment, preventing the hemodialysis patient from absorbing the backscattered signal of some labels after the body falls. The RFID reader antenna, which is placed in front of the toilet seat, continuously interrogate the tag array and obtain the backscattered RF signals from each tag. For each antenna–tag pair, the reader obtains a sequence of RF phase values and a sequence of received signal strength indicator(RSSI). Fig. 1 shows an overview of our system.

The basic idea of our RF-IDH system is to collect phase and RSSI information of RF signals and perform a series of signal preprocessing operations on these collected data. Then, through careful analysis of the rules of collecting data sample sets in the safe toilet process of hemodialysis patients, a reasonable and effective objective function for the optimization problem is established to derive the process model of hemodialysis patients safely complete the process of going to the toilet. Then, we design a residual feature extraction (RFE) algorithm based on the hemodialysis patient safety process model, and the fall detection of hemodialysis patients is treated as a machine learning problem where a series of classification models are built via learning residual feature space constructed by our proposed RFE algorithm. We implement our system on commercial off-the-shelf RFID devices and compared the evaluation metrics of the four methods of 2NN–RFE, LR–RFE, RF–RFE and SVM–RFE in terms of system performance, system efficiency, system robustness, and system latency.

There are many technical challenges that we will address in this paper. The first technical challenge is how to deal with the variety of fall downs that can occur when a hemodialysis patient gets up and walk after completing the defecation process. Therefore, in order to find a practical solution to the diversity of hemodialysis patients’ fall actions, we have made many technical attempts in the early stage, including solutions based on statistical feature extraction in sliding windows [1], [2], [3] and solutions based on Kalman filter tracking [4], [5], [6]. After careful comparison and analysis of the experimental data, we first denied the solution through the statistical feature extraction based on the sliding window, because the experimental data we collected needs to take into account the real needs of hospital blood purification centers, which did not deliberately limit the patient’s fall action, but try to ensure that the collected data sample set covers different fall directions and areas, different fall times, fall speeds, and different down postures. Therefore, a sample dataset with random diversity is destined to be too difficult to find common features through the sliding window without manual intervention of the fall action. The solution based on statistical feature extraction in the sliding window is impractical, so what about the solution based on Kalman filter tracking? Researchers familiar with Kalman filter are aware of the two major difficulties in establishing Kalman filter tracking. The first difficulty is how to establish a suitable process model to describe the state transfer of the state vector designed by Kalman filter with a specific state transfer function. The second difficulty is how to establish the corresponding relationship between the state space and the measurement space, and convert the prior value predicted by the process model into the phase or RSSI of the RF signal used in the measurement space with a specific measurement function. Given the diversity of falls in hemodialysis patients, the Kalman filter tracking solution does not seem to be easier or more feasible than the first one based on sliding window statistical characteristics extraction. However, we can completely abandon the specific constraints in Kalman filter design process, not to consider the two specific difficulties mentioned earlier, but to stand on a higher and more abstract level, to learn from the principles of Kalman filter and the ideas behind it. In fact, many members of the Bayesian filtering family, including Kalman filter, use different mathematical methods, but fundamentally, they only do one thing, that is, to make a proper choice between prior prediction and actual measurement. As for the selection of a prior prediction model, it depends on the design objective of the filter. Therefore, we rethink the problem that the RF-IDH system really needs to be solved for the hospital. Nobody cares about the types of falls of hemodialysis patients. In fact, the blood purification center only cares about whether the diversity falls of hemodialysis patients can be detected with high accuracy, high efficiency, high robustness and low delay by our intelligent system. If we think backward, it is easy to find such a rule. Compared with the ever-changing fall movements, the process of hemodialysis patient safety process is relatively simple, just normal standing and walking after completing the defecation process. Then, we established a reasonable and effective objective function for optimization problem and derived hemodialysis patient safety process model from measurement space based on dynamic time warping (DTW) [7].

With the help of the hemodialysis patient safety model, we need to address the second technical challenge of how to extract appropriate features to meet the performance requirements of various aspects of the system. Similarly, here we draw on the idea of adaptive filtering and fusion sensing. By calculating the residual between the feature of the RF signal in the measurement space and the hemodialysis patient safety process model, we propose a residual feature extraction (RFE) algorithm based on DTW. The phase and RSSI information of the RF signal is used as the input of the RFE algorithm to obtain the fusion sensing feature space. The detection accuracy and robustness of the system are guaranteed compared with the traditional way of using the phase or RSSI information of the RF signal alone. The algorithm eventually returns the feature matrix composed of the vector of residual phase feature and residual RSSI feature.

We make four key contributions in this paper. As far as I know, we have taken the lead in proposing an intelligent fall detection system for hemodialysis patients with a complication of hypotension or muscle spasm based on RF signal for hospital blood purification centers. Second, we propose an effective and feasible RFE algorithm for deriving hemodialysis patient safety process model from measurement space based on DTW and constructing the residual feature space of the system by calculating the residual between the phase/RSSI of the RF signal in the measurement space and the derived process model. Third, our proposed 2NN–RFE method takes advantage of the fusion of residual information of phase and RSSI of the RF signal and achieves superior performance over traditional methods using phase or RSSI of RF signal alone. Last, we implemented RF-IDH using COTS RFID reader with one antenna and three EPCglobal C1G2 standard passive tags attached to a badge with adjustable lanyard length for multi-speed sample acquisition and redundant backup in the real test environment, preventing the hemodialysis patient from absorbing the backscattered signal of some labels after the volunteer falls. The results show that RF-IDH achieves a F1 score of more than 0.99 in both the cross-validation stage and the final test data evaluation stage. And as the training sample ratio increases, the evaluation metric approaches the maximum metric it can achieve. Moreover, the experiment results also show that RF-IDH is robust against the influence of individual diversity by different volunteers and has an acceptable fall recognition latency.

Section snippets

Related work

Camera-based fall detection: F. Merrouche, B. Ni et al. [8], [9] proposed new methods for fall detection using depth camera by exploiting the advantages of Kinect. S. Zambanini et al. [10] proposed an approach for the detection of falls based on multiple cameras. T. Zhang et al. [11] proposed a falling detection algorithm based on a humanoid robot to monitor and detect the motion of the elderly who live alone. The Pi Camera system uses low-cost Pi Camera mounted on Raspberry Pi to monitor and

System overview

In this section, we analyze how to properly preprocess the RF signal, and how the proposed RFE algorithm extracts the residual feature vector required by the machine learning model from the preprocessed data. We call this method RF-IDH because it is a dedicated intelligent system for the hospital blood purification center to detect the safety process of hemodialysis patients in the toilet.

Our basic idea is to collect the phase and RSSI information in the backscattered signal of the RFID passive

Implementation and evaluation

After we complete the extraction of residual features and construct the residual feature space, we use a variety of machine learning methods to train the residual feature space, and persist the trained classification estimator, and then evaluate the performance of the system.

Conclusion

In the context of smart healthcare and rehabilitation, meeting privacy requirements plays a vital role. Therefore, the IoT devices participating in patient monitoring must ensure that the collected data information is highly secure. Compared with image data, RF signal data is completely free from the risk of privacy leakage and can gain sufficient trust from users. However, using RF signal characteristics to interpret human activities is a challenging and rewarding task, especially for the IoT

CRediT authorship contribution statement

Yi Chen conceived of the presented idea and contributed to the final manuscript. Fu Xiao was in charge of overall direction and planning. Haiping Huang discussed the experimental parts and polished English. Lijuan Sun proposed some suggestions about writing the manuscript.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Yi Chen received his B.S. and M.S. degrees in Communication Engineering and Electronics and Communication Engineering from Nanjing University of Posts and Telecommunications, Nanjing, China, in 2010 and 2014, respectively. He is currently a Ph.D. candidate in the Information Networking at Nanjing University of Posts and Telecommunications. His research interests include RFID Systems, and Internet of Things.

References (28)

  • XieL. et al.

    Multi-touch in the air: Concurrent micromovement recognition using RF signals

    IEEE/ACM Trans. Netw.

    (2018)
  • ZouY. et al.

    GRfid: A device-free RFID-based gesture recognition system

    IEEE Trans. Mob. Comput.

    (2017)
  • WangL. et al.

    Toward a wearable RFID system for real-time activity recognition using radio patterns

    IEEE Trans. Mob. Comput.

    (2017)
  • Y. Bu, L. Xie, Y. Gong, C. Wang, L. Yang, J. Liu, S. Lu, RF-dial: An RFID-based 2D human-computer interaction via tag...
  • Q. Lin, L. Yang, Y. Sun, T. Liu, X. Li, Y. Liu, Beyond one-dollar mouse: A battery-free device for 3D human-computer...
  • HuangC. et al.

    Real-time RFID indoor positioning system based on Kalman-filter drift removal and Heron-bilateration location estimation

    IEEE Trans. Instrum. Meas.

    (2015)
  • Stan SalvadorP.C.

    FastDTW: Toward accurate dynamic time warping in linear time and space

    Intell. Data Anal.

    (2007)
  • F. Merrouche, N. Baha, Depth camera based fall detection using human shape and movement, in: 2016 IEEE International...
  • B. Ni, C.D. Nguyen, P. Moulin, RGBD-camera based get-up event detection for hospital fall prevention, in: 2012 IEEE...
  • S. Zambanini, J. Machajdik, M. Kampel, Detecting falls at homes using a network of low-resolution cameras, in:...
  • T. Zhang, W. Zhang, L. Qi, L. Zhang, Falling detection of lonely elderly people based on NAO humanoid robot, in: 2016...
  • S.A. Waheed, P.S.A. Khader, A Novel approach for smart and cost effective IoT based elderly fall detection system using...
  • P.S. Ong, C.P. Ooi, Y.C. Chang, E.K. Karuppiah, S.M. Tahir, An FPGA-based hardware implementation of visual based fall...
  • R. Ramezani, Y. Xiao, A. Naeim, Sensing-Fi: Wi-Fi CSI and accelerometer fusion system for fall detection, in: 2018 IEEE...
  • Cited by (8)

    • Editorial for FGCS special issue: Intelligent IoT systems for healthcare and rehabilitation

      2021, Future Generation Computer Systems
      Citation Excerpt :

      Unhealthy living habits lead to more and more hemodialysis patients, while the existing fall detection techniques cannot meet the requirements of patient’s arbitrary movements. Chen et al. [2] presented an intelligent RFID based fall detection system for patient monitoring in hospital blood purification center. The basic idea is to use the collected phase and received signal strength indicator (RSSI) features of RF signals to circumvent the privacy problems caused by the camera-based detection methods and perform a series of signal processing operations along with various machine learning methods for fall detection.

    • Wear-free indoor fall detection based on RFID and deep residual networks

      2023, International Journal of Communication Systems
    • SiFall: Practical Online Fall Detection with RF Sensing

      2022, SenSys 2022 - Proceedings of the 20th ACM Conference on Embedded Networked Sensor Systems
    • Study on the effect of substances in passive RFID contactless sensing applications

      2022, 2022 3rd International Conference on Big Data, Artificial Intelligence and Internet of Things Engineering, ICBAIE 2022
    View all citing articles on Scopus

    Yi Chen received his B.S. and M.S. degrees in Communication Engineering and Electronics and Communication Engineering from Nanjing University of Posts and Telecommunications, Nanjing, China, in 2010 and 2014, respectively. He is currently a Ph.D. candidate in the Information Networking at Nanjing University of Posts and Telecommunications. His research interests include RFID Systems, and Internet of Things.

    Fu Xiao received the Ph.D. degree in computer science and technology from the Nanjing University of Science and Technology, Nanjing, China, in 2007. He is currently a Professor and a Ph.D. supervisor with the School of Computer Science and Technology, Nanjing University of Posts and Telecommunications, Nanjing, China. His research interests include wireless sensor networks. Dr. Xiao is a member of the IEEE Computer Society and the Association for Computing Machinery.

    Haiping Huang received the B.Eng. and M.Eng. degrees in computer science and technology from the Nanjing University of Posts and Telecommunications, Nanjing, China, in 2002 and 2005, respectively, and the Ph.D. degree in computer application technology from Soochow University, Suzhou, China, in 2009. From May 2013 to November 2013, he was a Visiting Scholar with the School of Electronics and Computer Science, University of Southampton, Southampton, U.K. He is currently a Professor with the School of Computer Science and Technology, Nanjing University of Posts and Telecommunications. His research interests include information security and privacy protection of wireless sensor networks.

    Lijuan Sun received the Ph.D. degree in information and communication from the Nanjing University of Posts and Telecommunications, Nanjing, China, in 2007. She is currently a Professor and a Ph.D. Supervisor with the School of Computers, Nanjing University of Posts and Telecommunications. Her main research interests are wireless sensor networks and wireless mesh networks.

    This work was supported in part by the National Key Research and Development Program of China under Grant 2018YFB0803400, in part by the Natural Science Foundation of Jiangsu for Distinguished Young Scientist under Grant BK20170039, in part by the National Natural Science Foundation of China under Grant 61873131, 61932013 and 61872196.

    View full text