International Journal of Fuzzy Logic and Intelligent Systems 2024; 24(3): 181-193
Published online September 25, 2024
https://doi.org/10.5391/IJFIS.2024.24.3.181
© The Korean Institute of Intelligent Systems
Le Van Hoa and Vo Viet Minh Nhat
Hue University, Hue City, Vietnam
Correspondence to :
Vo Viet Minh Nhat (vvmnhat@hueuni.edu.vn)
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Radio frequency identification (RFID) network planning is the problem of determining where to place RFID readers in a work area so that tags or objects tagged therein can be monitored. With recent developments in RFID technology, RFID-based monitoring systems have been deployed in many fields, including medical asset management. Monitoring medical assets in a hospital is essential for limiting losses and theft, thereby improving the quality of medical treatment and patient care. This study proposes an RFID network planning model for medical asset monitoring in which an artificial neural network (ANN) is used to optimize the placement of readers within a hospital campus with the constraints of a limited number of used readers and different priorities of monitored assets. The case study considered the Family Medicine Center, University of Medicine and Pharmacy, Hue University. The simulation results show that with the ANN-based optimization method, the optimal location of readers is quickly found while satisfying certain constraints related to medical asset monitoring.
Keywords: RFID, Network planning, Hopfield network, Optimization, Healthcare
Medical asset management is essential in hospitals. According to annual statistics, the amount of assets lost owing to misplacement, loss, and theft is significant (approximately 10%). In addition, the time employees spend searching for medical assets for work usually accounts for approximately 30% of their total working time [1]. Asset management is always tiring and time-consuming, which significantly affects the quality of medical services. Intelligent information systems are essential for the management of medical assets.
Medical asset monitoring is a critical function of medical asset management systems. Several monitoring technologies have been developed, among which sensor technologies have attracted the most attention. Specifically, sensor technologies, such as infrared, Bluetooth, ZigBee, smart cards, and radio frequency identification (RFID), have been introduced [2]. However, a suitable technology for medical asset monitoring must meet certain requirements, such as accuracy, compatibility with medical assets, economic cost, and monitoring specifications [3].
RFID uses radio technology to collect identification information from tags or tagged objects. An RFID system consists of tags, readers, and a host (server). The tags store identification data of the tagged object. The reader interrogates the tag’s identification data, receives the response data from the tag, and forwards them to the server for storage and subsequent processing [4, 5]. RFID technology is characterized by fast identification (thousands of tags per second), wireless communication, invisibility, and long-distance communication capabilities. Therefore, an RFID system is not limited by line of sight, is less hindered by obstacles, and adapts well to mobile objects. These characteristics render RFID technology particularly suitable for monitoring medical assets in hospitals.
To monitor tagged objects, a network of RFID readers must be deployed because the coverage radius of each RFID reader is limited. Placing readers in a workspace is the problem of determining the reader location so that the RFID network can cover the entire hospital campus; this is called the RFID network planning (RNP) problem. The RNP problem has proven to be NP-hard [6], and nature-based approaches, such as swarm intelligence and evolutionary computation, are often used to obtain a suitable solution. There have been a significant number of nature-inspired proposals, such as genetic algorithms (GA) [6], particle swarm optimization (PSO) [7], cuckoo search [8], and firefly algorithm [9], and their effectiveness on the RNP problem is quite impressive. An artificial neural network (ANN) is a nature-based approach that can learn from the environment, adaptively adjust behaviors, and provide solutions as prospectively as possible. This study investigates the application of ANNs to solve the RNP problem of monitoring tagged objects.
Among the types of ANNs that can solve optimization problems, the Hopfield network is particularly suitable. By formulating the objective function into a Hopfield energy function, reducing the energy function of the Hopfield network helps to determine the optimal solution to the RNP problem [10, 11]. This study proposes an ANN-based approach to the RNP problem to improve the efficiency of medical asset monitoring in hospitals. Accordingly, a Hopfield network is used to determine the optimal location for readers within a hospital campus. The objective is to maximize the metric of monitored tagged medical assets while meeting certain constraints, such as economic cost, energy efficiency, and monitoring specifications.
The main contributions of the study include:
- Proposing an approach for Hopfield-based optimization of reader placement locations for monitoring tagged assets in hospital environments.
- Developing an objective function that maximizes the covered medical asset value and simultaneously satisfies the requirements of other monitoring specifications such as economic cost and energy efficiency. The objective function is then formulated as a Hopfield energy function.
- Designing a Hopfield network to solve the RNP problem and integrating a placement area restriction technique into the Hopfield optimization process to increase the convergence speed and reduce the number of readers placed.
- Simulation results show that the Hopfield-based RNP optimization method has achieved significant efficiency in maximizing the covered medical asset value, minimizing the number of placed readers, and energy efficiency.
The remainder of this paper is organized as follows: Section 2 reviews several studies related to the implementation of RFID technology in medical asset management in hospitals, focusing on the RNP issue. Hopfield network-based RNP optimization is presented in Section 3, including a description of the case study, medical asset value model, RNP economic model, and RNP energy model, which are formulated into the objective function of RNP optimization. The Hopfield network-based optimization principle, conversion from the RNP objective function to the Hopfield energy function, proposed Hopfield network, and algorithm for RNP optimization are presented in detail. The implementation and evaluation of the simulation results are presented in Section 4. Finally, Section 5 presents our conclusions.
Monitoring in medical asset management is mandatory for medical facilities to effectively manage medical examination and treatment resources, enhance staff work performance, and increase patient satisfaction. Various monitoring technologies have been introduced, in which RFID has emerged as a promising candidate for monitoring medical assets in hospitals. Given a particular hospital campus (workspace), a network of RFID readers can be deployed to monitor tagged assets. The following are the reviews on RNP optimization approaches and RFID applications in the healthcare sector, emphasizing RNP.
GA and PSO are two approaches that have attracted considerable research attention for solving the RNP problem. GA and PSO can be combined, or each approach can be combined with other techniques. Typically, the multi-community GA-PSO model, a combination of GA and PSO, is proposed to solve the RNP problem in large-scale systems [12]. The main idea is to divide a PSO population into several colonies and use genetic selection and mutation strategies to improve the dynamic rules of the particle swarm. The simulation results show that the multi-community GA-PSO achieves a superior solution to the standard PSO.
PSO can be combined with a tentative reader elimination (TRE) operator [13], called PSO-TRE, to minimize the number of readers required, in which the TRE operator temporarily deletes readers during the PSO process and can restore deleted readers after a few generations if the deletion reduces tag coverage. By employing TRE, PSO-TRE can adaptively adjust the number of readers used to improve the overall performance of the RFID network. Furthermore, the mutation operator is embedded into the algorithm to improve the success rate of TRE. To evaluate its effectiveness, six requirements of RNP and real-world RFID operation scenarios are considered. The experimental results show that PSO-TRE achieves higher coverage and fewer readers than some of the compared algorithms.
GA can also be integrated with the redundant reader elimination (RRE) technique in a two-stage process (called GA-RRE) to minimize the number of readers placed [14]. GA first determines the optimal position of readers in terms of maximum tag coverage, minimum number of readers used, and minimum interference. For the RRE technique, a policy is provided to eliminate redundant readers with little or no impact on the tag interrogation efficiency. The number of candidate readers is limited to reduce the computational complexity. The working area is gridded, where each cell is a candidate position for placing a reader. Simulation results show that, with some cases of investigated cell sizes, GA-RRE performs better in terms of tag coverage, interference, and number of placed readers.
A combination of redundant antenna elimination and neural network algorithm, called RAE-NNA, was proposed in [15], in which RAE optimizes the RNP problem by eliminating redundant antennas and NNA minimizes the difference between the target solution and forecasted solutions. The results demonstrate the effectiveness and reliability of RAE-NNA in solving the RNP problem and designing cost-effective networks by minimizing the number of antennas and collisions while maximizing coverage.
On the application of RFID in the healthcare sector, the first proposal for the application of RFID in tracking medical assets in a healthcare zone is to optimize the design of a medical-asset tracking system constrained by a limited number of RFID readers [16]. With the workspace assumed to be a floor plan representation divided into a grid of squares, the research in [16] formulated the maximal covering location problem along with a new criticality index analysis metric (derived from the severity, frequency, and dwell time of critical medical assets) and the optimal placement of the limited number of RFID readers. Genetic algorithms are used to determine the optimal placement locations for the RFID readers. With the representation of allele as the location of reader placement, the chromosome as a set of p reader locations, the crossover at one point, and the mutation at one gene on the chromosome, the proposed method achieved a 72% improvement in coverage compared with the currently utilized expert/heuristic-based placement strategy.
Implementing RFID to monitor and reduce loss and theft has been hailed as having many benefits, but has yet to convince many healthcare providers. This is because of the lack of models that can measure effectiveness. In [17], the authors propose a Markov chain model that can quantify the benefits of RFID applications in reducing asset loss, reducing asset search time, and enhancing asset utilization efficiency in the hospital. The study examined the impact of RFID-based tracking in various hospital environments by comparing the performance of 48 scenarios with RFID with the implementation of 384 scenarios without RFID. Three groups of factors are considered: (1) patient needs and service rates; (2) search efficiency; and (3) preparation and maintenance policies. The study demonstrated that RFID-enabled device tracking systems could significantly increase asset utilization efficiency. In addition, the proposed model can be used to evaluate medical asset preparation and maintenance policies.
Najera et al. [18] propose solutions for two specific healthcare scenarios. One is a medical asset tracking system for healthcare facilities that allows real-time medical asset location and prevents theft. Aspects such as interference potential, technology selection, and RFID data management from hospital information systems are addressed and analyzed. Conversely, solutions for patient care and control in hospitals based on passive RFID are also mentioned. It is an innovative prototype of RFID features that provides a backup data source from a patient’s wristband. This prototype also provides an offline working mode to increase the application reliability when the network fails, thereby improving patient safety. The implementation results show that the proposed solution offers seamless patient identification and medical data retrieval with nearly 100% accurate identification reliability.
In [19], an optimization framework is proposed to plan a sensor network in a hospital for medical asset monitoring. First, an improved model for the statistical simulation of asset movements was developed to determine the criticality of locations within the hospital. The model provides an optimization algorithm that determines the optimal location of sensors and is modeled according to their coverage characteristics. Two network planning problems are considered: (1) maximizing the coverage with a fixed number of sensors and (2) minimizing the number of sensors with the minimum desired coverage. The problem is formulated as a binary set, and the Q-learning is used to calculate the paths of medical assets. A case study was conducted by the Department of Cardiology, San Camillo-Forlanini Hospital in Rome. The results show that by identifying critical areas, the optimization framework allows a significant reduction in the number of placed readers, and thus reduces the sensor network cost by approximately 40% on average.
Patil et al. [1] provide inventory management and asset tracking solutions using RFID technology. A monitoring and tracking system is developed to manage inventory automatically, track assets to prevent theft, establish communication between the reader and the tag via the ISO18000 protocol, and provide an easy-to-use interface for users. The problem of tag conflicts is also reduced by accurately reading multiple tags simultaneously. Through implementation and analysis, the system has proven its advantages in accurately managing inventory, reducing cumbersome processing, and preventing pilferage in the healthcare industry.
In [20], a new method is developed to automate the real-time detection of various objects. Hospital workflows are monitored and controlled using RFID. The process begins by defining the functional area by detecting the optical signs of the rooms. Next, the regions are clustered using the histogram matching technique. RFID network planning is used for monitoring. A density-based scanning algorithm is used to cluster and extract regions. All the collected data are transferred to a firefly algorithm to track drug distribution and assign locations of doctors and nurses. The results show that 87% of tags are covered in real time to manage and monitor hospital staff. The analysis indicates that the system helps to monitor hospital operations.
To track medical assets in hospital environments effectively, an intelligent RFID network planning model with multi-antenna readers (RNP-MAR) is proposed in [21]. The motivation of the study is to use multi-port readers to minimize the total cost of the RFID network and simultaneously maximize the network coverage. To achieve this, three types of readers with one, two, and four antenna ports are considered. Furthermore, different antenna models with varying coverage characteristics are considered for allocation to various ports of RFID readers. The purpose of RNP-MAR is not only to determine the optimal location of RFID readers in a hospital campus, but also to design multi-antenna readers and their antennas. To improve the efficiency of the RNP-MAR model, a hybrid nonlinear-local metaheuristic based on whale optimization and a simulated annealing algorithm (WOA-SA) is proposed. The multi-objective function of WOA-SA is to maximize coverage while minimizing total cost, interference, collision, and power consumption. The simulation results show that the proposal saves (on average) 39.57% of the total cost of the RFID network by efficiently using multi-antenna RFID readers.
In summary, in the healthcare sector, in addition to techniques related to disease diagnosis [22], the application of RFID technology for monitoring has attracted significant attention. RNP has been proven to be NP-hard, and nature-based approaches, such as those in [6–9, 12–21], are often suitable for finding an efficient solution. This study introduces a new approach to the RNP problem in which a type of ANN, the Hopfield network, is used as an optimization method for determining the optimal location of readers while satisfying some constraints related to medical asset monitoring. The following section details the proposal.
The Family Medicine Center (FMC) is a unit at the University of Medicine and Pharmacy, Hue University, where medical students practice and provide medical examination and treatment. Similar to other healthcare units, the loss and misplacement of assets are severe problems for FMC. An adequate medical asset monitoring system is a prerequisite for managing medical assets, while satisfying other medical constraints.
The FMC campus is almost square (Figure 1), and the white areas, including medical examination and treatment clinics, property warehouses, lobbies, and so on, are places where medical assets need monitoring. Gray areas, including electrical rooms, stairs, washrooms, and so on, are positions where medical assets never appear. The white areas, called workspaces, must be monitored; therefore, readers are placed to monitor medical assets.
An RFID tag is attached to each asset to identify and locate it. Suppose that the assets are distributed in the working area, as shown in Figure 2, in which 12 different types of monitored assets are represented by different colors. Each asset is associated with features, including price, frequency of use, quantity, and urgent role in medical treatment, the values of which are provided by the FMC’s medical experts (Table 1). These values are then normalized in the range [0, 1] using
where
The readers are assumed to be uniform and have a circular coverage radius of
Medical assets are associated with certain features, such as price, frequency of use, quantity, and an urgent role in medical treatment. These features are combined to form the medical asset value model. In the simple case, the value model can be a weighted sum of these features, as shown in
where
A reader placed at cell (
The objective of the RNP problem is to maximize the value of covered medical assets, as shown in
The RNP economic model is based on the cost of placing readers. Let
Reader placement costs can be reduced further if redundant readers are eliminated. The redundant readers are those whose elimination does not reduce the number of covered tags. The RRE techniques [13, 14], can be used to eliminate redundant readers. Subsection 3.9 introduces a combination of the RRE technique into the Hopfield network-based RNP optimization process.
Energy consumption in an RFID network originates from tag interrogation. Each reader must consume power to send commands to tags and maintain a downlink channel so that the tags can harvest power to reply to their data. The energy consumption for one tag interrogation cycle is as in
where
Ideally, each tag is covered by only one reader and the energy consumption of the entire system is equal to the total energy consumption for interrogating covered tags. However, some tags may be in the overlapping area of readers (Figure 2). Readers all interrogate tags; therefore, they must spend energy on the redundant interrogation. Minimizing the number of tags in the overlapping area reduces energy consumption, as shown in
where |
From
where
The Hopfield network is a type of associative memory, which is a branch of ANNs. The Hopfield network has an architecture of just one layer of neurons, in which each neuron is connected to all other neurons but has no connection to itself. Considering that only one layer exists, each neuron is both an input and an output. The performance (energy reduction) of the Hopfield network is based on the principle that the network energy does not increase and can be used for optimization [10, 11]. Accordingly, the objective function of the problem to be optimized is formulated in the form of the Hopfield energy function, as shown in
where
After each execution cycle, the change of energy is expressed by
As proven in [24], Δ
From the objective function in
where
By comparing
and
The early determination of weights and activation thresholds allows the Hopfield network to perform faster than other neural networks by skipping the training stage. This feature is an advantage of Hopfield-based optimization over methods based on other neural networks.
The Hopfield network used for the RNP problem at the FMC is a two-dimensional neural matrix, with the number of neurons corresponding to the number of cells. Each neuron represents a cell in a corresponding location (Figure 3). A neuron with the highest activation determines the cell in the corresponding position to place a reader. The set of neurons with the highest activation defines the corresponding cells selected to place the readers, a candidate solution to the RNP problem. The optimal solution is the collection of selected cells on which readers are placed, which maximizes the medical asset metric and satisfies other constraints.
Each cell (
The following are the main steps of the Hopfield network-based RNP optimization (HN-RNPO) algorithm.
The complexity of the HN-RNPO algorithm arises primarily from calculating the activation value of neurons. With a matrix of
As described in
The HN-RNPO algorithm is simulated on a PC with an 11th Gen Intel Core i5-1135G7 @ 2.40 GHz and 8 GB RAM. The case study is conducted at the FMC, which has a square campus of 41.6×41.6 m2. Three cell sizes are gridded: 5.2, 2.6, and 1.3 m, corresponding to 2, 1, and 1
The effectiveness of the HN-RNPO algorithm is evaluated based on criteria including:
- Coverage rate is determined by the ratio of the number of covered assets to total assets.
- Coverage value is measured by the ratio of the value of the covered assets to the value of the entire assets.
- Coverage efficiency is defined by the number of assets covered by each reader. This criterion is considered with and without RRE.
- Energy efficiency is calculated indirectly by the ratio of the number of covered tags to the total number of interrogations performed. This criterion is also considered with and without RRE.
Figures 5 and 6 show that the coverage rate and coverage value vary with different cell sizes. As the cell size decreases, the values of these criteria increase. This is because the smaller the cell size, the finer the gridding, and the greater the number of candidate cells selected to place readers. Consequently, the criteria approach the optimal value. However, the computational time significantly increases (Figure 7).
Also shown in Figures 5 and 6, the threshold
Figures 8 and 9 show the variation in coverage efficiency with different cell sizes, with two
More intuitively, Figures 10(a) and 11(a) show a comparison of the reader placement results for HN-RNPO without and with RRE. The cell size used in this study is 0.65 m. The results show that the coverage rate achieved is very high (99.25%), and the number of readers reduced by RRE is significant (29%).
We also compare the coverage efficiency of HN-RNPO with two other algorithms: GA-based RNP optimization (GA-RNPO) [13] and PSO-based RNP optimization (PSO-RNPO) [14]. The simulation results show that HN-RNPO achieves the highest coverage efficiency–99.25% compared to 91.3% and 89.80% of GA-RNPO and PSO-RNPO in Figure 10(b) and 10(c). When combining RRE, the number of readers reduced by GA-RNPO and PSO-RNPO is also significant (27.4%). However, compared with HN-RNPO, the number of readers required by GA-RNPO and PSO-RNPO is still higher when their coverage efficiency is lower.
Figures 12 and 13 show the variation in energy efficiency with different cell sizes and two thresholds
This study proposes a method for optimizing RFID network planning based on ANNs. The RNP optimization problem is considered to maximize the value of medical assets covered and simultaneously satisfy certain medical monitoring specification requirements, such as minimizing economic costs and energy efficiency. The Hopfield network is used to optimize the RNP objective function based on the non-increasing property of its energy function. Restricting the placement area and eliminating redundant readers are used to reduce the runtime of the algorithm and minimize the need for readers. The case study of the FMC of the University of Medicine and Pharmacy, Hue University, is implemented, and gridding with different mesh sizes is considered. Simulation results show that the Hopfield network helps quickly find an optimal solution for placing readers that maximizes the value of covered medical assets and minimizes costs and energy efficiency.
In practice, specific applications may use different types of directional antennas with different coverage radii. Working areas in noisy environments, such as concrete walls, metal partitions, reflective objects, and so on, also significantly affect the reader placement optimization process. Considering that multiple constraints increase the complexity of the RNP problem, further studies are required to determine the optimal solution.
No potential conflict of interest relevant to this article was reported.
The Family Medicine Center campus is gridded with an normalized distance of 3.69×cos(45°) ≈ 5.2 m.
Table 1. Medical assets and their original (org.) and normalized (norm.) feature values.
No. | Asset | Price ($) ( | Frequency ( | Quantity ( | Urgent role ( | Norm. value | ||||
---|---|---|---|---|---|---|---|---|---|---|
Org. | Norm. | Org. | Norm. | Org. | Norm. | Org. | Norm. | |||
1 | ECG monitors | 1000 | 0.323 | medium | 0.5 | 20 | 0.2 | high | 0.9 | 1.0242 |
2 | First aid kit | 166 | 0.041 | low | 0.1 | 10 | 0 | high | 0.9 | 0.8403 |
3 | Heart-lung stethoscope | 60 | 0.005 | high | 0.9 | 40 | 1 | high | 0.9 | 1.0914 |
4 | Health scale with ruler | 100 | 0.019 | high | 0.9 | 30 | 0.6 | low | 0.1 | 0.3414 |
5 | Infusion pole | 50 | 0.002 | low | 0.1 | 20 | 0.2 | low | 0.1 | 0.1438 |
6 | Oxygen bag | 44 | 0 | low | 0.1 | 40 | 1 | high | 0.9 | 0.9300 |
7 | Pregnancy heart monitor | 430 | 0.131 | medium | 0.5 | 10 | 0 | high | 0.9 | 0.9426 |
8 | Respiratory meter | 3000 | 1 | high | 0.9 | 10 | 0 | high | 0.9 | 1.2400 |
9 | Stretcher | 280 | 0.08 | low | 0.1 | 20 | 0.2 | low | 0.1 | 0.1633 |
10 | Surgical light | 480 | 0.147 | high | 0.9 | 40 | 1 | low | 0.1 | 0.4069 |
12 | Wheelchair | 100 | 0.019 | low | 0.1 | 30 | 0.6 | low | 0.1 | 0.1814 |
International Journal of Fuzzy Logic and Intelligent Systems 2024; 24(3): 181-193
Published online September 25, 2024 https://doi.org/10.5391/IJFIS.2024.24.3.181
Copyright © The Korean Institute of Intelligent Systems.
Le Van Hoa and Vo Viet Minh Nhat
Hue University, Hue City, Vietnam
Correspondence to:Vo Viet Minh Nhat (vvmnhat@hueuni.edu.vn)
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Radio frequency identification (RFID) network planning is the problem of determining where to place RFID readers in a work area so that tags or objects tagged therein can be monitored. With recent developments in RFID technology, RFID-based monitoring systems have been deployed in many fields, including medical asset management. Monitoring medical assets in a hospital is essential for limiting losses and theft, thereby improving the quality of medical treatment and patient care. This study proposes an RFID network planning model for medical asset monitoring in which an artificial neural network (ANN) is used to optimize the placement of readers within a hospital campus with the constraints of a limited number of used readers and different priorities of monitored assets. The case study considered the Family Medicine Center, University of Medicine and Pharmacy, Hue University. The simulation results show that with the ANN-based optimization method, the optimal location of readers is quickly found while satisfying certain constraints related to medical asset monitoring.
Keywords: RFID, Network planning, Hopfield network, Optimization, Healthcare
Medical asset management is essential in hospitals. According to annual statistics, the amount of assets lost owing to misplacement, loss, and theft is significant (approximately 10%). In addition, the time employees spend searching for medical assets for work usually accounts for approximately 30% of their total working time [1]. Asset management is always tiring and time-consuming, which significantly affects the quality of medical services. Intelligent information systems are essential for the management of medical assets.
Medical asset monitoring is a critical function of medical asset management systems. Several monitoring technologies have been developed, among which sensor technologies have attracted the most attention. Specifically, sensor technologies, such as infrared, Bluetooth, ZigBee, smart cards, and radio frequency identification (RFID), have been introduced [2]. However, a suitable technology for medical asset monitoring must meet certain requirements, such as accuracy, compatibility with medical assets, economic cost, and monitoring specifications [3].
RFID uses radio technology to collect identification information from tags or tagged objects. An RFID system consists of tags, readers, and a host (server). The tags store identification data of the tagged object. The reader interrogates the tag’s identification data, receives the response data from the tag, and forwards them to the server for storage and subsequent processing [4, 5]. RFID technology is characterized by fast identification (thousands of tags per second), wireless communication, invisibility, and long-distance communication capabilities. Therefore, an RFID system is not limited by line of sight, is less hindered by obstacles, and adapts well to mobile objects. These characteristics render RFID technology particularly suitable for monitoring medical assets in hospitals.
To monitor tagged objects, a network of RFID readers must be deployed because the coverage radius of each RFID reader is limited. Placing readers in a workspace is the problem of determining the reader location so that the RFID network can cover the entire hospital campus; this is called the RFID network planning (RNP) problem. The RNP problem has proven to be NP-hard [6], and nature-based approaches, such as swarm intelligence and evolutionary computation, are often used to obtain a suitable solution. There have been a significant number of nature-inspired proposals, such as genetic algorithms (GA) [6], particle swarm optimization (PSO) [7], cuckoo search [8], and firefly algorithm [9], and their effectiveness on the RNP problem is quite impressive. An artificial neural network (ANN) is a nature-based approach that can learn from the environment, adaptively adjust behaviors, and provide solutions as prospectively as possible. This study investigates the application of ANNs to solve the RNP problem of monitoring tagged objects.
Among the types of ANNs that can solve optimization problems, the Hopfield network is particularly suitable. By formulating the objective function into a Hopfield energy function, reducing the energy function of the Hopfield network helps to determine the optimal solution to the RNP problem [10, 11]. This study proposes an ANN-based approach to the RNP problem to improve the efficiency of medical asset monitoring in hospitals. Accordingly, a Hopfield network is used to determine the optimal location for readers within a hospital campus. The objective is to maximize the metric of monitored tagged medical assets while meeting certain constraints, such as economic cost, energy efficiency, and monitoring specifications.
The main contributions of the study include:
- Proposing an approach for Hopfield-based optimization of reader placement locations for monitoring tagged assets in hospital environments.
- Developing an objective function that maximizes the covered medical asset value and simultaneously satisfies the requirements of other monitoring specifications such as economic cost and energy efficiency. The objective function is then formulated as a Hopfield energy function.
- Designing a Hopfield network to solve the RNP problem and integrating a placement area restriction technique into the Hopfield optimization process to increase the convergence speed and reduce the number of readers placed.
- Simulation results show that the Hopfield-based RNP optimization method has achieved significant efficiency in maximizing the covered medical asset value, minimizing the number of placed readers, and energy efficiency.
The remainder of this paper is organized as follows: Section 2 reviews several studies related to the implementation of RFID technology in medical asset management in hospitals, focusing on the RNP issue. Hopfield network-based RNP optimization is presented in Section 3, including a description of the case study, medical asset value model, RNP economic model, and RNP energy model, which are formulated into the objective function of RNP optimization. The Hopfield network-based optimization principle, conversion from the RNP objective function to the Hopfield energy function, proposed Hopfield network, and algorithm for RNP optimization are presented in detail. The implementation and evaluation of the simulation results are presented in Section 4. Finally, Section 5 presents our conclusions.
Monitoring in medical asset management is mandatory for medical facilities to effectively manage medical examination and treatment resources, enhance staff work performance, and increase patient satisfaction. Various monitoring technologies have been introduced, in which RFID has emerged as a promising candidate for monitoring medical assets in hospitals. Given a particular hospital campus (workspace), a network of RFID readers can be deployed to monitor tagged assets. The following are the reviews on RNP optimization approaches and RFID applications in the healthcare sector, emphasizing RNP.
GA and PSO are two approaches that have attracted considerable research attention for solving the RNP problem. GA and PSO can be combined, or each approach can be combined with other techniques. Typically, the multi-community GA-PSO model, a combination of GA and PSO, is proposed to solve the RNP problem in large-scale systems [12]. The main idea is to divide a PSO population into several colonies and use genetic selection and mutation strategies to improve the dynamic rules of the particle swarm. The simulation results show that the multi-community GA-PSO achieves a superior solution to the standard PSO.
PSO can be combined with a tentative reader elimination (TRE) operator [13], called PSO-TRE, to minimize the number of readers required, in which the TRE operator temporarily deletes readers during the PSO process and can restore deleted readers after a few generations if the deletion reduces tag coverage. By employing TRE, PSO-TRE can adaptively adjust the number of readers used to improve the overall performance of the RFID network. Furthermore, the mutation operator is embedded into the algorithm to improve the success rate of TRE. To evaluate its effectiveness, six requirements of RNP and real-world RFID operation scenarios are considered. The experimental results show that PSO-TRE achieves higher coverage and fewer readers than some of the compared algorithms.
GA can also be integrated with the redundant reader elimination (RRE) technique in a two-stage process (called GA-RRE) to minimize the number of readers placed [14]. GA first determines the optimal position of readers in terms of maximum tag coverage, minimum number of readers used, and minimum interference. For the RRE technique, a policy is provided to eliminate redundant readers with little or no impact on the tag interrogation efficiency. The number of candidate readers is limited to reduce the computational complexity. The working area is gridded, where each cell is a candidate position for placing a reader. Simulation results show that, with some cases of investigated cell sizes, GA-RRE performs better in terms of tag coverage, interference, and number of placed readers.
A combination of redundant antenna elimination and neural network algorithm, called RAE-NNA, was proposed in [15], in which RAE optimizes the RNP problem by eliminating redundant antennas and NNA minimizes the difference between the target solution and forecasted solutions. The results demonstrate the effectiveness and reliability of RAE-NNA in solving the RNP problem and designing cost-effective networks by minimizing the number of antennas and collisions while maximizing coverage.
On the application of RFID in the healthcare sector, the first proposal for the application of RFID in tracking medical assets in a healthcare zone is to optimize the design of a medical-asset tracking system constrained by a limited number of RFID readers [16]. With the workspace assumed to be a floor plan representation divided into a grid of squares, the research in [16] formulated the maximal covering location problem along with a new criticality index analysis metric (derived from the severity, frequency, and dwell time of critical medical assets) and the optimal placement of the limited number of RFID readers. Genetic algorithms are used to determine the optimal placement locations for the RFID readers. With the representation of allele as the location of reader placement, the chromosome as a set of p reader locations, the crossover at one point, and the mutation at one gene on the chromosome, the proposed method achieved a 72% improvement in coverage compared with the currently utilized expert/heuristic-based placement strategy.
Implementing RFID to monitor and reduce loss and theft has been hailed as having many benefits, but has yet to convince many healthcare providers. This is because of the lack of models that can measure effectiveness. In [17], the authors propose a Markov chain model that can quantify the benefits of RFID applications in reducing asset loss, reducing asset search time, and enhancing asset utilization efficiency in the hospital. The study examined the impact of RFID-based tracking in various hospital environments by comparing the performance of 48 scenarios with RFID with the implementation of 384 scenarios without RFID. Three groups of factors are considered: (1) patient needs and service rates; (2) search efficiency; and (3) preparation and maintenance policies. The study demonstrated that RFID-enabled device tracking systems could significantly increase asset utilization efficiency. In addition, the proposed model can be used to evaluate medical asset preparation and maintenance policies.
Najera et al. [18] propose solutions for two specific healthcare scenarios. One is a medical asset tracking system for healthcare facilities that allows real-time medical asset location and prevents theft. Aspects such as interference potential, technology selection, and RFID data management from hospital information systems are addressed and analyzed. Conversely, solutions for patient care and control in hospitals based on passive RFID are also mentioned. It is an innovative prototype of RFID features that provides a backup data source from a patient’s wristband. This prototype also provides an offline working mode to increase the application reliability when the network fails, thereby improving patient safety. The implementation results show that the proposed solution offers seamless patient identification and medical data retrieval with nearly 100% accurate identification reliability.
In [19], an optimization framework is proposed to plan a sensor network in a hospital for medical asset monitoring. First, an improved model for the statistical simulation of asset movements was developed to determine the criticality of locations within the hospital. The model provides an optimization algorithm that determines the optimal location of sensors and is modeled according to their coverage characteristics. Two network planning problems are considered: (1) maximizing the coverage with a fixed number of sensors and (2) minimizing the number of sensors with the minimum desired coverage. The problem is formulated as a binary set, and the Q-learning is used to calculate the paths of medical assets. A case study was conducted by the Department of Cardiology, San Camillo-Forlanini Hospital in Rome. The results show that by identifying critical areas, the optimization framework allows a significant reduction in the number of placed readers, and thus reduces the sensor network cost by approximately 40% on average.
Patil et al. [1] provide inventory management and asset tracking solutions using RFID technology. A monitoring and tracking system is developed to manage inventory automatically, track assets to prevent theft, establish communication between the reader and the tag via the ISO18000 protocol, and provide an easy-to-use interface for users. The problem of tag conflicts is also reduced by accurately reading multiple tags simultaneously. Through implementation and analysis, the system has proven its advantages in accurately managing inventory, reducing cumbersome processing, and preventing pilferage in the healthcare industry.
In [20], a new method is developed to automate the real-time detection of various objects. Hospital workflows are monitored and controlled using RFID. The process begins by defining the functional area by detecting the optical signs of the rooms. Next, the regions are clustered using the histogram matching technique. RFID network planning is used for monitoring. A density-based scanning algorithm is used to cluster and extract regions. All the collected data are transferred to a firefly algorithm to track drug distribution and assign locations of doctors and nurses. The results show that 87% of tags are covered in real time to manage and monitor hospital staff. The analysis indicates that the system helps to monitor hospital operations.
To track medical assets in hospital environments effectively, an intelligent RFID network planning model with multi-antenna readers (RNP-MAR) is proposed in [21]. The motivation of the study is to use multi-port readers to minimize the total cost of the RFID network and simultaneously maximize the network coverage. To achieve this, three types of readers with one, two, and four antenna ports are considered. Furthermore, different antenna models with varying coverage characteristics are considered for allocation to various ports of RFID readers. The purpose of RNP-MAR is not only to determine the optimal location of RFID readers in a hospital campus, but also to design multi-antenna readers and their antennas. To improve the efficiency of the RNP-MAR model, a hybrid nonlinear-local metaheuristic based on whale optimization and a simulated annealing algorithm (WOA-SA) is proposed. The multi-objective function of WOA-SA is to maximize coverage while minimizing total cost, interference, collision, and power consumption. The simulation results show that the proposal saves (on average) 39.57% of the total cost of the RFID network by efficiently using multi-antenna RFID readers.
In summary, in the healthcare sector, in addition to techniques related to disease diagnosis [22], the application of RFID technology for monitoring has attracted significant attention. RNP has been proven to be NP-hard, and nature-based approaches, such as those in [6–9, 12–21], are often suitable for finding an efficient solution. This study introduces a new approach to the RNP problem in which a type of ANN, the Hopfield network, is used as an optimization method for determining the optimal location of readers while satisfying some constraints related to medical asset monitoring. The following section details the proposal.
The Family Medicine Center (FMC) is a unit at the University of Medicine and Pharmacy, Hue University, where medical students practice and provide medical examination and treatment. Similar to other healthcare units, the loss and misplacement of assets are severe problems for FMC. An adequate medical asset monitoring system is a prerequisite for managing medical assets, while satisfying other medical constraints.
The FMC campus is almost square (Figure 1), and the white areas, including medical examination and treatment clinics, property warehouses, lobbies, and so on, are places where medical assets need monitoring. Gray areas, including electrical rooms, stairs, washrooms, and so on, are positions where medical assets never appear. The white areas, called workspaces, must be monitored; therefore, readers are placed to monitor medical assets.
An RFID tag is attached to each asset to identify and locate it. Suppose that the assets are distributed in the working area, as shown in Figure 2, in which 12 different types of monitored assets are represented by different colors. Each asset is associated with features, including price, frequency of use, quantity, and urgent role in medical treatment, the values of which are provided by the FMC’s medical experts (Table 1). These values are then normalized in the range [0, 1] using
where
The readers are assumed to be uniform and have a circular coverage radius of
Medical assets are associated with certain features, such as price, frequency of use, quantity, and an urgent role in medical treatment. These features are combined to form the medical asset value model. In the simple case, the value model can be a weighted sum of these features, as shown in
where
A reader placed at cell (
The objective of the RNP problem is to maximize the value of covered medical assets, as shown in
The RNP economic model is based on the cost of placing readers. Let
Reader placement costs can be reduced further if redundant readers are eliminated. The redundant readers are those whose elimination does not reduce the number of covered tags. The RRE techniques [13, 14], can be used to eliminate redundant readers. Subsection 3.9 introduces a combination of the RRE technique into the Hopfield network-based RNP optimization process.
Energy consumption in an RFID network originates from tag interrogation. Each reader must consume power to send commands to tags and maintain a downlink channel so that the tags can harvest power to reply to their data. The energy consumption for one tag interrogation cycle is as in
where
Ideally, each tag is covered by only one reader and the energy consumption of the entire system is equal to the total energy consumption for interrogating covered tags. However, some tags may be in the overlapping area of readers (Figure 2). Readers all interrogate tags; therefore, they must spend energy on the redundant interrogation. Minimizing the number of tags in the overlapping area reduces energy consumption, as shown in
where |
From
where
The Hopfield network is a type of associative memory, which is a branch of ANNs. The Hopfield network has an architecture of just one layer of neurons, in which each neuron is connected to all other neurons but has no connection to itself. Considering that only one layer exists, each neuron is both an input and an output. The performance (energy reduction) of the Hopfield network is based on the principle that the network energy does not increase and can be used for optimization [10, 11]. Accordingly, the objective function of the problem to be optimized is formulated in the form of the Hopfield energy function, as shown in
where
After each execution cycle, the change of energy is expressed by
As proven in [24], Δ
From the objective function in
where
By comparing
and
The early determination of weights and activation thresholds allows the Hopfield network to perform faster than other neural networks by skipping the training stage. This feature is an advantage of Hopfield-based optimization over methods based on other neural networks.
The Hopfield network used for the RNP problem at the FMC is a two-dimensional neural matrix, with the number of neurons corresponding to the number of cells. Each neuron represents a cell in a corresponding location (Figure 3). A neuron with the highest activation determines the cell in the corresponding position to place a reader. The set of neurons with the highest activation defines the corresponding cells selected to place the readers, a candidate solution to the RNP problem. The optimal solution is the collection of selected cells on which readers are placed, which maximizes the medical asset metric and satisfies other constraints.
Each cell (
The following are the main steps of the Hopfield network-based RNP optimization (HN-RNPO) algorithm.
The complexity of the HN-RNPO algorithm arises primarily from calculating the activation value of neurons. With a matrix of
As described in
The HN-RNPO algorithm is simulated on a PC with an 11th Gen Intel Core i5-1135G7 @ 2.40 GHz and 8 GB RAM. The case study is conducted at the FMC, which has a square campus of 41.6×41.6 m2. Three cell sizes are gridded: 5.2, 2.6, and 1.3 m, corresponding to 2, 1, and 1
The effectiveness of the HN-RNPO algorithm is evaluated based on criteria including:
- Coverage rate is determined by the ratio of the number of covered assets to total assets.
- Coverage value is measured by the ratio of the value of the covered assets to the value of the entire assets.
- Coverage efficiency is defined by the number of assets covered by each reader. This criterion is considered with and without RRE.
- Energy efficiency is calculated indirectly by the ratio of the number of covered tags to the total number of interrogations performed. This criterion is also considered with and without RRE.
Figures 5 and 6 show that the coverage rate and coverage value vary with different cell sizes. As the cell size decreases, the values of these criteria increase. This is because the smaller the cell size, the finer the gridding, and the greater the number of candidate cells selected to place readers. Consequently, the criteria approach the optimal value. However, the computational time significantly increases (Figure 7).
Also shown in Figures 5 and 6, the threshold
Figures 8 and 9 show the variation in coverage efficiency with different cell sizes, with two
More intuitively, Figures 10(a) and 11(a) show a comparison of the reader placement results for HN-RNPO without and with RRE. The cell size used in this study is 0.65 m. The results show that the coverage rate achieved is very high (99.25%), and the number of readers reduced by RRE is significant (29%).
We also compare the coverage efficiency of HN-RNPO with two other algorithms: GA-based RNP optimization (GA-RNPO) [13] and PSO-based RNP optimization (PSO-RNPO) [14]. The simulation results show that HN-RNPO achieves the highest coverage efficiency–99.25% compared to 91.3% and 89.80% of GA-RNPO and PSO-RNPO in Figure 10(b) and 10(c). When combining RRE, the number of readers reduced by GA-RNPO and PSO-RNPO is also significant (27.4%). However, compared with HN-RNPO, the number of readers required by GA-RNPO and PSO-RNPO is still higher when their coverage efficiency is lower.
Figures 12 and 13 show the variation in energy efficiency with different cell sizes and two thresholds
This study proposes a method for optimizing RFID network planning based on ANNs. The RNP optimization problem is considered to maximize the value of medical assets covered and simultaneously satisfy certain medical monitoring specification requirements, such as minimizing economic costs and energy efficiency. The Hopfield network is used to optimize the RNP objective function based on the non-increasing property of its energy function. Restricting the placement area and eliminating redundant readers are used to reduce the runtime of the algorithm and minimize the need for readers. The case study of the FMC of the University of Medicine and Pharmacy, Hue University, is implemented, and gridding with different mesh sizes is considered. Simulation results show that the Hopfield network helps quickly find an optimal solution for placing readers that maximizes the value of covered medical assets and minimizes costs and energy efficiency.
In practice, specific applications may use different types of directional antennas with different coverage radii. Working areas in noisy environments, such as concrete walls, metal partitions, reflective objects, and so on, also significantly affect the reader placement optimization process. Considering that multiple constraints increase the complexity of the RNP problem, further studies are required to determine the optimal solution.
The Family Medicine Center campus with monitored and unmonitored areas.
The Family Medicine Center campus is gridded with an normalized distance of 3.69×cos(45°) ≈ 5.2 m.
The Hopfield network corresponds to the cells of the gridded workspace.
Example of placement restriction areas (at edges and around placed readers).
Coverage rates with varied cell sizes and
Coverage values with varied cell sizes and
Runtime increases rapidly as cell size decreases.
Coverage efficiency with the threshold
Coverage efficiency with the threshold
The placement results of (a) HN-RNPO, (b) GA-RNPO and (c) PSO-RNPO without RRE.
The placement results of (a) HN-RNPO, (b) GA-RNPO and (c) PSO-RNPO with RRE.
Energy efficiency with the threshold
Energy efficiency with the threshold
Table 1 . Medical assets and their original (org.) and normalized (norm.) feature values.
No. | Asset | Price ($) ( | Frequency ( | Quantity ( | Urgent role ( | Norm. value | ||||
---|---|---|---|---|---|---|---|---|---|---|
Org. | Norm. | Org. | Norm. | Org. | Norm. | Org. | Norm. | |||
1 | ECG monitors | 1000 | 0.323 | medium | 0.5 | 20 | 0.2 | high | 0.9 | 1.0242 |
2 | First aid kit | 166 | 0.041 | low | 0.1 | 10 | 0 | high | 0.9 | 0.8403 |
3 | Heart-lung stethoscope | 60 | 0.005 | high | 0.9 | 40 | 1 | high | 0.9 | 1.0914 |
4 | Health scale with ruler | 100 | 0.019 | high | 0.9 | 30 | 0.6 | low | 0.1 | 0.3414 |
5 | Infusion pole | 50 | 0.002 | low | 0.1 | 20 | 0.2 | low | 0.1 | 0.1438 |
6 | Oxygen bag | 44 | 0 | low | 0.1 | 40 | 1 | high | 0.9 | 0.9300 |
7 | Pregnancy heart monitor | 430 | 0.131 | medium | 0.5 | 10 | 0 | high | 0.9 | 0.9426 |
8 | Respiratory meter | 3000 | 1 | high | 0.9 | 10 | 0 | high | 0.9 | 1.2400 |
9 | Stretcher | 280 | 0.08 | low | 0.1 | 20 | 0.2 | low | 0.1 | 0.1633 |
10 | Surgical light | 480 | 0.147 | high | 0.9 | 40 | 1 | low | 0.1 | 0.4069 |
12 | Wheelchair | 100 | 0.019 | low | 0.1 | 30 | 0.6 | low | 0.1 | 0.1814 |
Gayoung Kim
International Journal of Fuzzy Logic and Intelligent Systems 2024; 24(3): 287-294 https://doi.org/10.5391/IJFIS.2024.24.3.287Jung-Sook Kim and Junho Jeong
International Journal of Fuzzy Logic and Intelligent Systems 2019; 19(4): 342-348 https://doi.org/10.5391/IJFIS.2019.19.4.342Minyoung Kim
Int. J. Fuzzy Log. Intell. Syst. 2017; 17(1): 10-16 https://doi.org/10.5391/IJFIS.2017.17.1.10The Family Medicine Center campus with monitored and unmonitored areas.
|@|~(^,^)~|@|The Family Medicine Center campus is gridded with an normalized distance of 3.69×cos(45°) ≈ 5.2 m.
|@|~(^,^)~|@|The Hopfield network corresponds to the cells of the gridded workspace.
|@|~(^,^)~|@|Example of placement restriction areas (at edges and around placed readers).
|@|~(^,^)~|@|Coverage rates with varied cell sizes and
Coverage values with varied cell sizes and
Runtime increases rapidly as cell size decreases.
|@|~(^,^)~|@|Coverage efficiency with the threshold
Coverage efficiency with the threshold
The placement results of (a) HN-RNPO, (b) GA-RNPO and (c) PSO-RNPO without RRE.
|@|~(^,^)~|@|The placement results of (a) HN-RNPO, (b) GA-RNPO and (c) PSO-RNPO with RRE.
|@|~(^,^)~|@|Energy efficiency with the threshold
Energy efficiency with the threshold