A novel methodology for patient prescreening using wireless body area networks (WBANs)

Introduction

Background

In contemporary society, individuals are increasingly susceptible to and impacted by various diseases as a result of numerous internal and external factors (1). Upon experiencing symptoms, individuals seek medical attention at hospitals where they are required to endure extended waiting periods in queues for triage. Subsequently, hospital personnel conduct vital checks, including temperature assessment, saturation of peripheral oxygen (SpO₂) measurement, heart rate evaluation, blood pressure examination, and diabetes screening (2). Notably, the assessment of their pain intensity involves presenting them with a manual chart and soliciting their subjective estimation of pain level. After completion, the data is manually inputted into expensive proprietary information technology (IT) systems owned by the hospitals and is accessible to both the doctors and the hospital staff. The patient is granted access to their medical record data upon request; otherwise, they are not afforded visibility of it. In addition, it should be noted that patients presenting themselves for triage, whether outdoors or in the emergency department, exhibit varying degrees of disease severity. Consequently, hospital personnel must manually prioritize and administer treatment to these patients, resulting in a lack of efficiency and inconvenience for those in the triage process who may receive treatment out of sequence.

Wireless body area networks (WBANs) as shown in Figure 1 have been recognized as a revolutionary technology within the healthcare field, as they allow for the uninterrupted monitoring of physiological parameters and the provision of personalized healthcare services. WBANs are comprised of diminutive, energy-efficient wearable or implantable sensors that are strategically positioned either on or within the human body (3). The sensors are responsible for gathering essential physiological indicators and other health-related information, which are subsequently transmitted wirelessly to a centralized monitoring system. The ability to monitor in real-time and remotely provides several benefits in terms of early identification of anomalies, enhanced provision of healthcare to patients, and improved overall quality of life (4).

Figure 1 WBAN system structure. BAN, body area network; DMU, data monitoring unit; WBAN, wireless body area network.

The emergence and advancement of WBANs have been influenced by various contributing factors. The continuous monitoring of health is imperative due to the rising incidence of chronic illnesses and the aging demographic. WBANs offer a non-intrusive and inconspicuous means of monitoring essential physiological indicators, including heart rate, blood pressure, body temperature, and respiratory rate. The implementation of continuous monitoring in healthcare settings enables healthcare providers to promptly identify deviations from normal physiological parameters and promptly intervene, thereby enhancing disease management and mitigating the risk of complications. Furthermore, the progress in wireless communication technologies and the miniaturization of electronic components have played a pivotal role in enabling the creation of compact and portable sensors that consume minimal power. The integration of these sensors into garments or their attachment to the human body enables individuals to perform their daily tasks without any disruption. The utilization of wireless data transmission obviates the necessity for unwieldy cables, thereby facilitating enhanced mobility and unrestricted physical motion (5). In addition, the widespread adoption of intelligent devices and the internet of things (IoT) has expedited the incorporation of WBANs into healthcare infrastructures. The data obtained from WBAN sensors can be transmitted in a secure manner to the systems of healthcare providers, thereby enabling the provision of remote patient monitoring and telemedicine services. The implementation of remote monitoring technology allows healthcare practitioners to evaluate the health condition of patients, offer immediate feedback, and make necessary modifications to treatment strategies (6). Nevertheless, the implementation of WBANs within healthcare environments presents numerous technical obstacles. A significant obstacle that must be addressed pertains to the establishment of dependable and effective communication channels between the sensors and the central monitoring system. The design of communication protocols should account for the distinctive attributes of WBANs, including but not limited to low power consumption, limited bandwidth, and the possibility of interference. Several studies have addressed this challenge. For instance, Naeem et al. (7) proposed an adaptive routing algorithm for WBANs with the objective of enhancing energy efficiency and prolonging the network’s operational lifespan. Power management is an essential component of WBANs. In order to optimize the battery life, it is necessary to implement energy-efficient techniques for wearable sensors, given their limited power resources. Several scholarly investigations have put forth power-saving strategies, including duty cycling and dynamic voltage scaling. For example, Zhao et al. (8) introduced a duty cycling scheme that is energy-efficient for WBANs. This scheme aims to extend the battery life of sensor nodes while maintaining reliable data transmission. In conjunction with the aspects of communication and power management, security and privacy emerge as significant considerations within WBANs. Given the sensitive nature of the health data handled by WBANs, it is imperative to prioritize the safeguarding of patient privacy and the maintenance of data integrity throughout the processes of transmission and storage. Numerous scholarly investigations have put forth frameworks for secure data transmission, encryption algorithms, and access control mechanisms as potential solutions to address these aforementioned concerns. For instance, Alrawais et al. (9) implemented a lightweight encryption scheme designed to ensure secure data transmission within WBANs. Moreover, the utilization of data fusion techniques is of utmost importance in WBANs as they serve the purpose of extracting significant and valuable information from various sensor inputs. Data fusion algorithms are utilized to integrate data from multiple sensors in order to improve the precision and dependability of health parameter measurements. For example, Gu et al. (10) proposed a multi-sensor fusion algorithm for heart rate monitoring in WBANs, combining data from photo plethysmography (PPG) and accelerometer sensors to improve accuracy. Electronics and automated health systems have used a classification approach to categorize the amount of risk associated with heart disease. The support vector machine (SVM) algorithm has shown remarkable effectiveness in illness detection, while the use of feature selection techniques further enhances its performance (11). The heart disease’s presence has been predicted by the use of SVM and convolutional neural network (CNN) (12,13) for classification and feature extraction. Research has shown that the 1-D CNN achieves higher accuracy in categorizing electrocardiogram (ECG) signals (14). Logistic regression (LR) demonstrates comparable efficacy to machine learning models in forecasting the likelihood of significant chronic illnesses characterized by low occurrence rates and uncomplicated clinical indicators. This paper aims to design the WBAN system, which utilizes machine learning, particularly LR. The choice of LR is based on its demonstrated superiority over SVM and CNN. The system is implemented using a Raspberry Pi and other related sensors to measure patient health. The patient data and severity levels are monitored through a custom-designed dashboard.

Methods

Dataset description and processing

The dataset provided in this study includes a total of 236 recorded instances of patients cases. These cases have been scientifically recorded and may be used to categorize the intensity of symptoms experienced by patients. This categorization is based on a set of specified standard symptoms. The dataset comprises 53 instances of mortality and 183 instances characterized by varying degrees of severity. This study investigated the problem of classifying patients into three categories: mild, medium, or severe symptoms risk, with the last potentially resulting in fatal outcomes. The dataset, sourced from BMC Emergency Medicine, has 16 significant characteristics that impact the presence of dangerous symptoms. Each feature is characterized as outlined in Table 1.

It is also obvious that the dependent variable is death and all other columns are independent variables.

The first step in the data cleaning process is the elimination of non-correlated columns by the use of expert judgment. In this particular scenario, it is apparent that the first column is associated with the count of individuals receiving medical treatment, a variable that is considered inconsequential in the process of constructing the model.

Furthermore, statistical methodologies, such as the use of the median, might be employed to address the issue of missing data. However, the current investigation identified a total of five cases of missing data within the dataset. These occurrences were especially seen in the oxygen saturation variable, as shown in Figure 2. The median oxygen saturation rate is 92 breaths per minute. All missing data is replaced with the value 92, and the resulting dataset after filling in the missing values is shown in Figure 3.

Figure 2 Detecting the missing data. BP, blood pressure; DM, diabetes mellitus; HIV, human immunodeficiency virus; COPD, chronic obstructive pulmonary disease.

Figure 3 Missing data test in python. BP, blood pressure; DM, diabetes mellitus; HIV, human immunodeficiency virus; COPD, chronic obstructive pulmonary disease.

An outlier refers to an observation point within a dataset that deviates significantly from the majority of readings. The Data Science project proceeds with the process of data collecting, during which outliers were first included into the population. During the selection process, it is possible that the existence of outliers may not be known. During the process of data analysis, the presence of outliers may be attributed to either accidental occurrences or serve as indicators of inherent variability within the dataset. This investigation did not identify any outliers. The dataset was partitioned in a stratified fashion, with 80% allocated for training purposes and the remaining 20% reserved for testing.

The next phase entails the process of feature selection, where the dataset, resembling a spreadsheet in the format of rows and columns similar to Excel, is prepared for machine learning purposes, namely for classification or regression tasks. In the domain of data analysis, the terminology “bars” may be used synonymously with the phrase “samples”. Likewise, the word “attributes” may be used to denote the characteristics of an observation inside a certain area of concern, typically referred to as “columns”. Feature selection is a methodology that entails the identification and selection of a subset of the most relevant characteristics (columns) from a provided dataset. The use of this procedure is often seen in methodologies designed to enhance the efficacy of machine learning models. The efficiency and accuracy of machine learning algorithms may be improved by reducing the number of features, which in turn decreases their space or time complexity. In the field of machine learning, the existence of extraneous input characteristics may cause ambiguity in certain algorithms, eventually leading to less than ideal predicting results. The Recursive Feature Elimination (RFE) technique is a filtering algorithm often used for characteristics presented in a wrapper style. The suggestion is that a certain machine learning algorithm be applied as the central component of the process, which is encompassed by RFE and used to aid in the selection of features. This stands in opposition to the use of filter-based features, which assess each function and choose the characteristics with the most elevated (or lowest) score. The RFE algorithm is a frequently used method for wrapper-style feature selection. It often incorporates filter-based feature selection as an internal mechanism. The RFE technique works by first examining all features present in the training dataset and then eliminating features in an iterative manner until the required number of features is achieved, relying on a subset of features. The procedure entails using the machine learning algorithm integrated into the heart model to assess the importance of characteristics, eliminating the least important ones, and then retraining the model. The aforementioned method is repeated till the number of features listed remains the same. The employment position uses the RFE technique in Python to determine the suitable features.

Following the use of RFE to identify and eliminate the least significant features, the validation of the data collection was conducted. Subsequently, an experiment was done using SVM with k-fold cross validation. The findings indicate that the five most significant variables are age, gender, temperature, heart rate, and oxygen saturation level. The model’s accuracy was roughly 80%, a level of performance that is similar to the accuracy obtained when using all available features.

Machine learning models

The system under consideration has been developed using machine learning techniques. The present research included the development and assessment of models based on LR, SVM, and CNN methodologies. The next section is a comprehensive explanation of the definitions of the three models.

LR

LR is a statistical technique used for the purpose of forecasting a binary result, relying on independent variables that may take the form of categorical, continuous, or a mix of both. LR is a frequently used statistical methodology for the purpose of modeling a dependent variable that exhibits two or more discrete outcomes. When confronted with a collection of independent factors, the resulting conclusion may be represented as a binary answer, characterized by a yes/no, 1/0, or real/wrong outcome, or alternatively as a high/low number. Figure 4 presents a conceptual framework that visually represents the correlation between a continuous indicator and the probability of an occurrence or result. If the observed correlation between variable X and the probability is valid, it may be concluded that the linear model is not an appropriate choice for fitting the data. To accurately represent this interaction, it is necessary to use a non-linear equation. One of the characteristics exhibits the tale. The sigmoid function is used to depict the characteristic S-shape of the feature (11,12).

Figure 4 Logistic regression model.

A modified version of LR utilizes a transformation of the logit function to estimate probabilities. The logit function refers to the natural logarithm of the chances (Eq. [1]) and the LR (Eq. [2]) respectively are:

Logit(p)=ln(p1−p)[1]

ln(p1−p)=β0+β1x1+⋯+βkxk[2]

Where P is the probability of an occurrence, x_k is the input and β_k is the regression coefficients.

SVM

The SVM is a supervised machine learning technique that is proficient in handling both classification and regression tasks. In the context of classification difficulties, it is mostly used (13,14). The operational fundamentals of the SVM method are shown by representing data items as points in an n-dimensional space, where n corresponds to the number of features. Each function is associated with a particular coordinate, representing its value. Following this, the process of differentiation is carried out by determining the hyperplane that effectively distinguishes the two groups, as seen in Figure 5.

Figure 5 Hyperplane with two vectors.

The present matter concerns the determination of a suitable hyperplane definition as per equation (3). Three hyperplanes, denoted as A, B, and C, have been used to distinguish between two distinct groups, as seen in Figure 6. When attempting to choose the most suitable hyperplane, it is crucial to stick to the following principle: “Choose the hyperplane that offers enhanced discrimination between the two classes”. In this specific case, hyper-plane “B” successfully achieved its targeted objective.

H:wt(x)+b=0[3]

Figure 6 Three hyperplanes.

Where w^t(x) is the normalized input and b is the equation intercept of hyperplane.

The finding of the optimal hyperplane may be accomplished by maximizing the distances between the nearest data points, irrespective of their class, and the hyperplane. This phenomenon is seen in Figure 7. The area around the content of a document, also known as the margin, is the space between the content and the document’s edge. The margin of hyperplane C is larger than that of both hyperplanes A and B, as seen in the diagram provided. The hyperplane labeled as C is often known as the proper hyperplane. The rationale for choosing a hyper-plane with a greater margin is based on its resilience, which is a very desired attribute for lightning phenomena. The act of choosing a hyperplane with a narrow margin is likely to result in a higher likelihood of misclassification.

Figure 7 Hyperplanes distances.

The method of reflecting data points and connecting them with a line on an individual basis has been successfully demonstrated. The determination of a line’s selection is contingent upon its distance, which may be quantified by the use of Eq. [4].

dH(Φ(x))=|w(Φ(x))+b||w|2[4]

Where Φ(x) is the input (x) feature mapped basis vector and |w|₂ is called Euclidean norm.

CNN

A CNN is a specialized variant of a feed-forward neural network that has been specially developed for the purpose of processing visual data (15,16). The use of a grid-like topology is often employed in order to comprehend the incoming data. CNNs have also been subjected to categorization.

The use of a CNN is done with the objective of recognizing and identifying artifacts that are present within a picture. Figure 8 illustrates a neural network that has been specifically developed to differentiate between two distinct categories of flowers, namely roses and orchids. The convolution process is an essential element of CNNs. The convolution process may be understood by using two one-dimensional matrices, specifically referred to as matrices a and b. In this scenario, let us consider two arrays: array a, which consists of the elements [5,3,7,5,9,7], and array b, which consists of the elements [1,2,3].

Figure 8 Convolutional neural network.

During the process of convolution, arrays undergo element-wise multiplication, which is then followed by the summing of the resultant values. The aforementioned procedure yields a novel array that symbolizes the multiplication of variables a and b. The first three elements of matrix b undergo a multiplication operation. In order to get the answer, the product is consolidated and the procedure may be carried out, as seen in Figure 9.

Figure 9 CNN convolution process (elements multiplication). CNN, convolutional neural network.

The convolution process entails the multiplication of the consecutive three entries from matrix a with their corresponding elements in matrix b. The resultant goods are then aggregated, as seen in Figure 10. The aforementioned procedure is iterated until the convolution operation reaches completion.

Figure 10 CNN convolution process. CNN, convolutional neural network.

A neural convolutional network is composed of numerous hidden layers that enable the extraction of information from the input, such as an image. The input picture is subjected to a sequence of convolutional layers using filters (also known as kernels), pooling operations, fully connected layers (FC), and the Softmax activation function. This process aims to identify an item and assign probability values within the range of 0 to 1. CNN models, which are based on deep learning techniques, are used for the objectives of training and analysis in theoretical contexts. Figure 11 depicts the comprehensive procedure of the CNN for creating an input picture and classifying artifacts according to their corresponding values.

Figure 11 CNN network processes. CNN, convolutional neural network; ReLU, Rectified Linear Unit.

The first step in the process of extracting features from an input picture is the use of the convolution layer. Convolution is a computational method that maintains the association among pixels via the examination of tiny input data squares, enabling the acquisition of picture features. The mathematical action shown in Figure 12 involves the use of two inputs, namely an image matrix and a kernel or filter.

Figure 12 Example of image matrix multiplies by filter matrix.

The use of filters enables the convolution of an image with several filters, hence facilitating tasks such as edge detection, blurring, and sharpening. The word “stride” denotes the magnitude of pixel displacement applied to the input matrix during the processing stage in order to generate the output matrix. When the phase is set to 1, the filters are calibrated to shift by 1 pixel on each cycle. When the phase reaches a value of 2, the filters undergo a transfer procedure that takes place in discrete steps of two pixels. Based on the information shown in Figure 13, it can be inferred that the use of a phase of 2 will result in the effectiveness of convolution.

Figure 13 Stride operation.

There exists a potential for the filter to exhibit imperfect correspondence with the input picture. There are two alternatives at one’s disposal. One such method involves applying zero-padding to the picture, either to expand it to match the dimensions of the filter or to remove the region of the image that was not included by the filter. The approach under discussion is sometimes referred to as real padding, in which only half of the picture is retained. The Rectified Linear Unit (ReLU) is a commonly used activation function in non-linear computations. The final mathematical statement may be represented as x(x) = max (0, x), where x is a variable. The ReLU function is used in CNNs to include non-linearity. The CNN’s objective is to understand real-world features that are characterized by non-negative linear values. The Pooling Layer component has the ability to decrease the parameter count in the context of handling large-scale photographs. Spatial pooling is a well-recognized method referred to as subsampling or down sampling in academic literature. The aforementioned approach effectively decreases the dimensions of individual images while preserving essential data.

The process of the FC entails transforming a matrix into a vector and feeding it into a neural network as a FC, as seen in Figure 13. Figure 14 illustrates the conversion of the map matrix function into a vector form, denoted as (x1, x2, x3, ...). The aforementioned characteristics are included into the fully linked layers to establish the model. Activation functions, such as Softmax or Sigmoid, may be used for the purpose of identifying the outputs.

Figure 14 Fully connected layer.

System configuration and implementation

Hardware architecture

The proposed smart healthcare monitoring system is designed to monitor essential indicators such as SpO₂ (blood oxygen saturation levels), heart rate, body temperature, and blood pressure. The system incorporates a very precise sensor for the measurement of the existing characteristics (Table 1). The design parameters of the proposed system are determined by many essential elements, including mobility, affordability, accuracy, quality, and convenience of use. The use of sensors is often seen in conjunction with the Raspberry Pi 4 (Rpi-4) Board. Various communication protocols are used for the purpose of transmitting and receiving data between sensors and the Raspberry Embedded system. Figure 15 depicts the implementation of a WBAN topology for the purpose of collecting and analyzing patients’ information.

Figure 15 System block diagram.

The Rpi-4 serves as the central processing unit (CPU) of the system, using its superior performance and exceptional hardware and software capabilities. The Rpi-4 model utilizes a Quad-Core Cortex-A72 (ARM v8) 64-bit processor, including CPUs operating at a frequency of 1.5 GHz. Additionally, the device has a total of 40 input/output pins, which may be programmed to function as Inter-Integrated Circuit (I2C) ports, Serial Peripheral Interface (SPI) ports, High-Definition Multimedia Interface (HDMI) ports, and a subset of these pins are capable of delivering a consistent voltage of 3.3 and 5 V. The AMG8833, an affordable InfraRed thermal camera, is used for the purpose of measuring body temperature. The AMG8833 refers to a set of heat sensors manufactured by Panasonic, with an 8×8 resolution. The device has the capability to measure the body temperature of a person from a maximum distance of 7 meters, which is approximately equivalent to 23 feet. The highest frame rate is 10 Hz.

Furthermore, the use of photoplethysmography (PPG) technology using Picamera, as seen in Figure 16, is employed for the acquisition of both typical and atypical human cardiovascular and hemodynamic measurements. The MAX32664D is a sensor capable of quantifying pulse oximetry SpO₂, heart rate, and blood pressure by means of finger contact. The sensor technology employs a combination of two light-emitting diodes (LEDs), a photodetector, enhanced optical components, and a low-noise analog signal processing system to accurately detect pulse oximetry and heart rate data. The use of OpenCV with deep learning models enables the precise determination of an individual’s gender and age based on a live video feed. The deep learning model used in this system for age and gender detection was implemented and trained according to the methodology outlined in reference (17).

Figure 16 Picamera.

Software architecture

A specialized graphical user interface (GUI) is used as a localized dashboard, enabling the user to engage with a WBAN over a WiFi network. The cloud-based server, Thingspeak (18), built by MatLab, is used for the purpose of storing data collected at regular intervals of 1 minute from various sensors. Additionally, the data will be presented on a Thinkspeak website for the healthcare institution. The operational mechanism of the cloud service is the establishment of distinct channels, whereby data from the sensors is autonomously received for each individual channel.

The installation of the Raspbian operating system (19) has been completed. Raspbian is an operating system based on the Linux kernel, designed to provide comprehensive support for a range of programming languages including Python, C, C++, and others. In the context of this particular research endeavor, the Python programming language has been used entirely for the development of the complete system. The Python implementation of the WBAN system utilizes many packages and libraries, as shown in Table 2.

The customized GUI serves as a localized dashboard, enabling users to assess pulse oximetry SpO₂, heart rate, and blood pressure by means of finger touch. Additionally, it utilizes the camera and deep learning algorithms to estimate the user’s age and gender. The “Analyze” button enables the user to execute the deep learning program that was previously stated and evaluated. Deep learning has robust learning skills that enable the classification of patients based on their healthcare data. Figure 17 displays the flow diagram of the proposed system.

Figure 17 WBAN process flow chart. WBAN, wireless body area network.

Results

Machine learning plays a significant role in our everyday lives, particularly in its application to scientific and market-oriented domains. In order for a particular business to get actual advantages, it is crucial that machine learning models provide precise predictions. The computation of the mean absolute error in regression problems. In the context of model assessment, the use of model evaluation metrics is crucial for assessing the efficacy of a model. The choice of measurement criteria is dependent on the particular machine learning job being considered, which encompasses classification, regression, ranking, clustering, topic modeling, and other associated applications. To effectively complete a range of activities, it is advantageous to use distinct metrics, such as time, accuracy, and recall, among others. The predominant machine learning programs consist of supervised learning functions, namely classification and regression. This section pertains to the metrics used in classification models (20,21).

The assessment of a classification algorithm’s accuracy in the field of machine learning involves evaluating the extent to which the algorithm accurately assigns a given data point to its appropriate class. Accuracy is defined as the ratio of accurately predicted data points to the total number of data points. The LR model demonstrated the greatest level of accuracy, achieving a score of 85 percent. This was closely followed by the SVM model, which reached an accuracy of 81 percent. The CNN model, on the other hand, earned a comparatively lower accuracy of 56 percent. The degree of accuracy is determined by the proportion of relevant cases to the total number of retrieved instances. In the present study, the LR model demonstrated the greatest accuracy, achieving a relative score of 88%. In comparison, the SVM and CNN models obtained scores of 88% and 50% respectively. In the domain of machine learning, the term “recall” is well recognized and referred to as “sensitivity”. The metric signifies the proportion of accurately recovered instances in relation to the overall number of relevant examples. The results of this research indicate that the LR model had the greatest recall rate of 85% when compared to the other tested models. Based on the data shown in Table 3, it can be seen that the LR model exhibited a higher level of performance when compared to both the SVM and CNN models.

Discussion

Machine learning models are often assessed by the use of k-fold cross-validation, and their selection is determined by their average or median performance metrics. Based on the forecast, it is anticipated that the algorithm with the greatest mean/median output would outperform algorithms with lower mean performance. When faced with a statistical anomaly that leads to a deviation in the average results, it becomes imperative to use a statistical hypothesis test in order to determine the credibility of the disparity in average or median performance between two algorithms (22). The determination of the suitable test format should be predicated upon the distribution of outcomes. The Shipro test was used in this research to analyze the distributions of the dependent variable. The P values for the dependent variables of LR, SVM, and CNN were found to be 5×10⁻⁵, 3.1×910⁻⁵, and 0.00489, respectively. The findings of this study suggest that all P values are below the threshold of 0.5, indicating statistical significance. Additionally, the analysis reveals that the variables under investigation do not exhibit a normal distribution. Consequently, non-parametric statistical test methodologies were used. The Mann-Whitney U test, a very successful and extensively used procedure, was chosen for implementation in this particular investigation.

The LR, SVM, and CNN models were assessed for their performance at a significance threshold of 95%. The null hypothesis was subjected to testing, and P values were computed in order to ascertain its significance at a 95% confidence level, with a threshold of less than 0.05. In the event that the P values were found to be below the significance level of 0.05, it would be feasible to reject the null hypothesis and instead accept the alternative hypothesis via the utilization of the Mann-Whitney test. This remark implies that the statistical study has established a 95% level of significance in the observed disparities between the models. The training and testing durations for all models are documented and shown in Table 4. Based on the data provided in Table 4, it can be seen that the LR model exhibited a higher level of performance when compared to both the SVM and CNN models.

The health monitoring dashboard shown in Figure 18 enables healthcare providers to effectively monitor and evaluate patient data in a consolidated and structured way.

Figure 18 WBAN system dashboard. WBAN, wireless body area network.

This technology enables the ongoing observation of essential physiological indicators and other health metrics in real-time via the use of medical sensors. The dashboard presents a range of data that healthcare practitioners may use to monitor and swiftly address any abnormal or important developments. The dashboard incorporates analytical tools that use deep learning techniques for the processing and analysis of patient data. This methodology has the potential to facilitate the identification of correlations, the prediction of trends, and the generation of detailed reports for each individual patient.

Conclusions

Within the present global context characterized by the ongoing pandemic, which has exerted a substantial impact on the global community and necessitated the implementation of novel contactless measures, the identification, surveillance, and management of health issues present considerable challenges on a global scale. Further changes are needed for the current manual illness detection procedure in order to promote productivity, health, safety, and the timely reaction of workers engaged in disease management via automation. This research aims to develop a gadget that provides a holistic, environmentally friendly, and power-efficient approach to identifying, tracking, and overseeing the well-being of persons who are experiencing illness or adversity. The proposed method involves the use of noninvasive body area network (BAN) nodes for the purpose of data collection and transmission to a central control server. The system under consideration utilizes machine learning algorithms to examine the medical profiles of patients, through employs noninvasive components of a BAN to continually monitor critical health metrics of patients, such as temperature, heart rate, blood pressure, etc., hence the system endeavors to ascertain the degree of illness severity in patients. Afterwards, monitoring and management procedures are implemented on the patients, ensuring that the central dashboard is consistently updated and appropriate people are swiftly informed. The system will continually monitor people who need ongoing surveillance as a result of their severity of illness. It is expected that the system would attract a significant number of users from many businesses and aspects of human life. The purpose of this system is to address the varied needs of many businesses and departments, including but not limited to hospitals, factories, educational institutions, and the hospitality industry.

Acknowledgments

Funding: None.

Footnote

Data Sharing Statement: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-24-2/dss

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-24-2/prf

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-24-2/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. IRB and informed consent are not applicable to this study as it doesn’t involve any human experiment.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sriraam N, Chakravarty P. Review of the Requirements for a Wearable, Low-Cost External Cardiac Loop Recorder with WBAN in Resource-Constrained Settings. Crit Rev Biomed Eng 2023;51:27-41. [Crossref] [PubMed]
2. Demrozi F, Turetta C, Kindt PH, et al. A Low-Cost Wireless Body Area Network for Human Activity Recognition in Healthy Life and Medical Applications. IEEE Trans Emerg Top Comput 2023;11:839-50. [Crossref]
3. Kolluri SSK, Mukherjee A, Sundararajan ADD. Design and Implementation of Wearable Patient Health Monitoring System Using WBAN and IoT. ECS Trans 2022;107:7821-33. [Crossref]
4. Ivanov S, Foley C, Balasubramaniam S, et al. Virtual groups for patient WBAN monitoring in medical environments. IEEE Trans Biomed Eng 2012;59:3238-46. [Crossref] [PubMed]
5. Noman SA, Noman HA, Al-Maatouk Q, et al. A Survey of IEEE 802.15. 6: Body Area Networks. International Journal of Computing and Digital Systems 2023;14:691-705. [Crossref]
6. Aski VJ, Dhaka VS, Parashar A, et al. Internet of Things in healthcare: A survey on protocol standards, enabling technologies, WBAN architectures and open issues. Physical Communication 2023;60:102103-18. [Crossref]
7. Naeem MA, Shah GA, Din IU. An Adaptive Routing Algorithm for Wireless Body Area Networks in Smart Health. IEEE J Biomed Health Inform 2018;22:748-57.
8. Zhao H, Guo Y, Yu X, et al. Energy-efficient Duty Cycling for Wireless Body Area Networks. IEEE Trans Mob Comput 2016;15:1470-83.
9. Alrawais A, Alhothaily A, Hu C, et al. LEEF: Lightweight Encryption and Efficient Forwarding for Wireless Body Area Networks in e-Healthcare Systems. IEEE Transactions on Information Forensics and Security 2017;12:144-57.
10. Gu J, Wang Z, Kuen J, et al. Recent advances in convolutional neural networks. Pattern Recognition 2018;77:354-77. [Crossref]
11. Wilson PW, D'Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk factor categories. Circulation 1998;97:1837-47. [Crossref] [PubMed]
12. Kausar N, Palaniappan S, Samir BB, et al. Systematic analysis of applied data mining based optimization algorithms in clinical attribute extraction and classification for the diagnosis of cardiac patients. In: Hassanien AE, Grosan C, Fahmy Tolba M. editors. Applications of Intelligent Optimization in Biology and Medicine. Intelligent Systems Reference Library, vol 96. Springer, Cham; 2016:217-31.
13. Nurmaini S, Partan RU, Caesarendra W, et al. An automated ECG beat classification system using deep neural networks with an unsupervised feature extraction technique. Appl Sci 2019;9:2921. [Crossref]
14. Nusinovici S, Tham YC, Chak Yan MY, et al. Logistic regression was as good as machine learning for predicting major chronic diseases. J Clin Epidemiol 2020;122:56-69. [Crossref] [PubMed]
15. Li Z, Liu F, Yang W, et al. A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects. IEEE Trans Neural Netw Learn Syst 2022;33:6999-7019. [Crossref] [PubMed]
16. Handelman GS, Kok HK, Chandra RV, et al. Peering Into the Black Box of Artificial Intelligence: Evaluation Metrics of Machine Learning Methods. AJR Am J Roentgenol 2019;212:38-43. [Crossref] [PubMed]
17. Liu Y, Xu J, Zhan Q, et al. Multi-sensor Fusion Algorithm for Accurate Heart Rate Monitoring in WBAN. IEEE Internet of Things Journal 2018;5:4597-607.
18. Available online: https://thingspeak.com/
19. Available online: https://www.raspbian.org/
20. Vujovic Z. Classification Model Evaluation Metrics. International Journal of Advanced Computer Science and Applications 2021;12:599-606. [Crossref]
21. Trawiński B, Smętek M, Telec Z, et al. Nonparametric statistical analysis for multiple comparison of machine learning regression algorithms. Int J Appl Math Comput Sci 2012;22:867-81. [Crossref]
22. Khadse VM, Mahalle PN, Shinde GR. Statistical Study of Machine Learning Algorithms Using Parametric and Non-Parametric Tests: A Comparative Analysis and Recommendations. International Journal of Ambient Computing and Intelligence 2020;11:80-105. [Crossref]