A Fog Computing-Based Cost-Effective Smart Health Monitoring Device for Infectious Disease Applications ^†

Abstract

The COVID-19 epidemic has raised awareness of exactly how crucial it is to continuously observe issues and diagnose respiratory problems early. Although the respiratory system is the primary objective of the disease’s acute phase, subsequent complications of SARS-CoV-2 infection might trigger enduring respiratory problems and symptoms, according to new research. These signs and symptoms, which collectively inflict considerable strain on healthcare systems and people’s quality of life, comprise, but are not restricted to, congestion, shortage of breath, tightness in the chest, and a decrease in lung function. Wearable technology offers a promising remedy to this persistent issue by offering continuous respiratory parameter monitoring, facilitating the early control and intervention of post-COVID-19 issues with respiration. In an effort to enhance patient outcomes and reduce expenses related to healthcare, this paper examines the possibility of using wearable technology to provide remote surveillance and the early diagnosis of respiratory problems in individuals suffering from COVID-19. In this work, a fog computing-based cost-effective smart health monitoring device is proposed for infectious disease applications. Further, the proposed device consists of three different biosensor modules, namely a MAX90614 infrared temperature sensor, a MAX30100 pulse oximeter, and a microphone sensor. All these sensor modules are connected to a fog computing device, namely a Raspberry PI microcontroller. Also, three different sensor modules were integrated with the Raspberry PI microcontroller and individuals’ physiological parameters, such as oxygen saturation (SPO2), heartbeat rate, and cough sounds, were recorded by the computing device. Additionally, a convolutional neural network (CNN)-based deep learning algorithm was coded inside the Raspberry PI and was trained with normal and COVID-19 cough sounds from the KAGGLE database. This work appears to be of high clinical significance since the developed fog computing-based smart health monitoring device is capable of identifying the presence of infectious disease through individual physiological parameters.

1. Introduction

The COVID-19 pandemic shone a spotlight on the critical requirement of prompt identification and the efficient monitoring of respiratory problems, since patients may have both short-term symptoms and long-term consequences. Although COVID-19’s acute phase largely affects the respiratory system, a new study indicates that post-infection complications may result in ongoing respiratory issues, placing a heavy cost on healthcare systems and negatively affecting people’s quality of life. In order to effectively manage these respiratory disorders, there is a pressing need for novel methods that permit ongoing monitoring and prompt intervention. Improved ways to perform early detection, monitoring, and therapeutic intervention require advancement because transmissible diseases cause severe challenges for healthcare systems across the world. Fog computing is a new method whose abilities broaden the potential of cloud-based computing to the peripheral regions of the network, paving the way to promote the formation of secured and profitable medical solutions. This investigation aims to enhance real-time monitoring and diagnosis, along with treatment plans by designing a fog computing-based smart health monitoring device essential for communicable disease applications [1].

This study suggests an exemplary fog computing-based smart health monitoring device designed exclusively for COVID-19 applications as an attempt to aid in fulfilling this pressing requirement [2]. The three physiological modules were implemented into the device in order in order to constantly track and conserve the patient’s medical data: the sound sensor, the MAX30100 pulse oximeter, and the MAX90614 infrared radiation temperature sensor. The monitoring devices can be used together with a Raspberry Pi microprocessor-based fog computing device which enables real-time information for interpretation and analysis. Regardless of this, the suggested device relies on a deep learning system which employs the extensive use of convolutional neural networks (CNNs), which at first were trained on a database of cough signals from both COVID-19 patients and healthy people, to assess the probability of an infectious disease being present, based on agreed medical standards. Through the implementation of fog computing technology, the device may offer instantaneous data which promote early identification, thereby enhancing the quality of life and retaining the medical expenses associated with COVID-19 care. Traditional medical monitoring remedies generally depend upon an infrastructure that uses cloud-based computing, providing the centralized processing of information [3]. Inevitably, this centralized management infrastructure has limitations like bandwidth limitations due to congestion or lag and issues with confidentiality. The previously addressed constraints are worse during pandemics of infectious diseases, when real-time data analysis and treatment are crucial. Fog computing allows distributed data analysis and processing; improves speed of response, flexibility, and latency; and moves computation towards a data source with the goal of conquering the aforementioned challenges.

Considering the fundamental concepts of fog computing, our suggested smart health monitoring device determines an interconnected computing structure containing wearable sensors and localized fog nodes that may be lined up as devices on the edges within communities or medical centers [4]. These cutting-edge technologies periodically receive biological markers and indicators associated with infectious diseases, as well as plenty of additional health-related information. In fog computing, the information gathered is actively analyzed using fundamental protocols to remove extraneous data and recover relevant data. No centralized cloud resources are needed as a result of this decentralized examination, minimizing concerns about network performance, data security, and privacy. Also, by improving the early identification of pandemics of infectious diseases and by implementing preventive measures in a timely manner, the fog computing system enhances real-time data analytics and decision-making [5]. By searching for similarities in the information that has been gathered, powerful machine learning algorithms composed of fog nodes have the ability to recognize deviations that might represent warning signs of diseases or advances in the treatment of diseases. For better unified patient management and monitoring, the medical device may interact with up-to-date medical systems, especially electronic health record (EHR) systems [6].

Our proposed method offers numerous positive aspects, of which the most significant being its affordability: fog computing promotes an effective use of already existing assets and minimizes the requirement for cost-inefficient cloud infrastructure. Leveraging the implementation of economical, cutting-edge technology and locally deployed fog nodes, our smart health monitoring device provides a flexible and economical solution that is suitable for utilization in resource-constrained instances, especially nations with limited resources and low-income communities.

This study defines a new approach to infectious illness control and monitoring by employing fog computing in order to create an affordable smart health monitoring device. By using the decentralized management of data processing, the improvements of real-time analytics, and the optimization of resource utilization, the solution that we suggest is designed to transform the delivery of healthcare, specifically in the domain of infectious disease control and prevention. In summary, this paper suggests an in-depth overview of the difficulties that arise from respiratory issues in the context of COVID-19 and suggests a novel approach that incorporates wearable technology and fog computing. Considering its ability to deliver continuous monitoring, timely detection, and feasible care, the suggested smart health monitoring device has the potential to improve the treatment of infectious diseases and improve patient care in this phase of the current pandemic.

2. Literature Survey

The COVID-19 pandemic has highlighted the vital requirement for novel approaches in the healthcare industry, most notably when it comes to the diagnosis and treatment of infectious diseases. As the global pandemic expands, more research is being carried out on the advancement and utilization of cutting-edge technology to deal with the issues produced by respiratory disorders like COVID-19. Considering these medical devices are equipped with sensors to track vital metrics like heart rate, respiration rate, and oxygen level, they can continuously monitor patients’ health status outside of normative clinical settings. Wearable technology has gained a lot of interest as a potentially valuable tool for monitoring physiologic parameters among individuals with respiratory disorders. Numerous studies [2,4] suggest that wearable technology is both feasible and beneficial for tracking respiratory symptoms and identifying early indicators of decline in COVID-19 patients.

Fog computing is an essential approach that has been developed that can assist with the effective use of wearable technology in healthcare applications. By broadening cloud computing abilities towards the network edge, fog computing provides the ability to handle, evaluate, and make decisions in real time nearer to the data source. Advantages associated with cloud computing technology include enhanced flexibility, less latency, and improved data protection and confidentiality. Recent investigations [3,4] have shown that fog computing provides an abundance of possibilities for use in medical applications. This can be particularly true when it comes to controlling and observing illnesses that are contagious.

Artificial intelligence (AI), mainly convolutional neural networks (CNNs), have been proven to be successful in identifying and diagnosing the existence of numerous illnesses, including infections of the respiratory tract. CNNs examine large amounts of data, including clinical and radiological signals, in order to identify similarities that indicate the existence or development of diseases. Several studies indicate the potency of deep learning algorithms in identifying breathing-related disorders through the analysis of cough sounds and other physiological signals, highlighting the potential for early disease identification and treatment [5,6].

Wearable devices, fog computing, and artificial intelligence altogether could significantly enhance the detection, management, and prevention of infectious diseases. Researchers can develop advanced technologies which enable early identification, diagnosis, and rehabilitation through the integration of wearable devices for continual physiological evaluation and fog computing for instant data processing, alongside deep learning for identifying diseases. In accordance with recent research, these integrated systems are practical as well as effective for keeping track of and managing diseases that are infectious, among other medical applications [1,2].

3. Methodology

3.1. Proposed System

The proposed system is a smart health monitoring gadget that utilizes fog computing for identifying and observing respiratory conditions in real time, particularly those affiliated with COVID-19 (Figure 1). In general, there are three primary components/layers that contribute to the proposed framework, based on [7,8]. The first layer consists of biosensor modules, which collect data and transfer them to the middle layer. Furthermore, the middle layer is the fog layer, which computes the input data and sends the output to the cloud layer. The cloud layer presents the output in the form of decision support for the caretaker/physician of the patients.

• Biosensors Modules: The medical device incorporates three biosensor devices, namely a microphone sensor for capturing cough noises; an infrared sensor module (MLX90614), which acts as an IoT edge device to measure temperature; and a pulse oximeter module (MAX30100) to measure a patient’s pulse.
• Fog Computing Node: The fog computing node aggregates and evaluates knowledge collected by the sensor technology which employs a Raspberry Pi microcontroller.
• Deep Learning Model: The Raspberry Pi microcontroller was designed using a proprietary convolutional neural network (CNN), which investigates cough sounds and distinguishes prospective COVID-19 infections.

3.2. Data Acquisition

The process of gathering sensor data involves the following:

• The recording of the internal temperature of the body using the MAX90614 sensor.
• The measurement of the pace of the heart and the saturation level of oxygen (SpO₂) utilizing the MAX30100 sensor.
• The recording of cough noises using a microphone sensor which is vital in diagnosing problems associated with breathing.

The gathered data include a collection of an ample number of cough sounds from COVID-19 patients and individuals who are healthy; these were obtained through a publicly available database, namely the KAGGLE database. Also, by including data from individuals of a variety of ages and both sexes, as well as data of COVID-19 symptoms from various phases, data heterogeneity is guaranteed.

3.3. Data Pre-Processing and Feature Extraction

The evaluations concerning temperature and SpO2 were calibrated according to the requirements of this section. Also, the coughing noises were processed before being transmitted as audio signals for the purpose of mitigating noise and enhancing the signal quality. It is essential to extract appropriate characteristics, irrespective of the biosensor data, such as the fluctuation of temperatures, times, and variances in cardiovascular activity and SpO2. Leverage techniques were adopted from signal processing for extracting harmonic contradiction, luminescence features, and Mel-frequency cepstral coefficients (MFCCs) from coughing sound effects.

3.4. Deep Learning Algorithm

An especially effective type of deep learning process for performing the analysis of visual information is the convolutional neural network (CNN). Its design architecture, which seamlessly and flexibly integrates the spatially structured arrangements of attributes from images that are input, is based on the visual cortex of primates. Figure 2 shows the block diagram for the proposed CNN model. Convolutional design, pooling, and fully associated layers represent some of the various stages that comprise a CNN in the majority of circumstances. Convolutional layers formulate feature databases and capture significant characteristics like the edges, shapes, and different variations by adding multiple kinds of filters to the input data. The statistical dimensions contained in these representations of features are then compressed by pooling layers, which conserves the most valuable information while reducing data processing complexity. For instance, to yield a final output, which might reflect a regression coefficient value, the classification labels, or any other intended result, layers that are fully integrated synthesize the information and gathered attributes. For services like the identification of objects, perception detection, and image analysis in healthcare, convolutional neural networks (CNNs) have proven to be extremely profitable when considering their tendency to acquire and generalize information from huge quantities of data.

CNNs are indispensable for obtaining a statistical evaluation of cough patterns in the suggested fog computing-based smart health monitoring system, particularly when using them to uncover the presence of COVID-19. The algorithm employed by the CNN is able to identify particular sound rhythms correlated to the ailment by being trained on a dataset of coughing sounds originating from COVID-19 sufferers and individuals who are well. To make it possible to uncover significant components in the recordings of cough noises have been transmitted into the CNN, and the program processes the recorded audio information using its layers. Whether a coughing sound is a warning sign of a COVID-19 infection is apparent by the network’s final output. Real-time, portable devices’ interpretation of sounds caused by breathing has been rendered possible by integrating this CNN-based data with the help of the Raspberry PI microcontroller. This makes the early detection and continual monitoring of breathing-related issues feasible. This technique enables the device to be more beneficial and efficient in dealing with transmissible illnesses by strengthening its diagnosis capabilities, guaranteeing rapid intervention times, and reducing the demand on centralized cloud infrastructures

4. Results and Discussion

In this work, the recordings of coughs from healthy and COVID-19-positive patients were obtained from the COUGHVID crowdsourcing dataset [9]. Furthermore, the recordings of healthy individuals and COVID-19-positive patients along with the human physiological parameters, such as temperature, oxygen saturation (SPO2) and heart rate, were stored inside the Raspberry PI-based fog computing device. Also, these data were considered and utilized for the training and testing of the proposed CNN model. The proposed CNN algorithm was coded inside the fog computing device using the Python programming language. A total of 100 cough sounds of normal and COVID positive patients were utilized, and 80% of data was utilized for training and 20% was used for the testing phase. Firstly, 80% of the input data was assigned to train the proposed CNN algorithm. As a result, a knowledge base was created in the form of a model file. The testing of the proposed CNN model was carried out following the training of the proposed CNN. Figure 3 shows the heat map of the proposed CNN model. Also, the performance matrices of the proposed CNN model for the COVID-19 cough signal classification are shown in Figure 4.

From Figure 3, it can be seen that the confusion matrices were plotted between the predicted and actual results. Also, the confusion matrix elements, such as true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN), are clearly shown in Figure 4. From these confusion matrix elements, the performance parameters, such as accuracy, sensitivity and specificity, were derived. Further, it is evident that the accuracy of the proposed CNN is 94%. Also, it can be observed that the sensitivity and specificity values of the proposed model are 94% and 94%, respectively. In addition, it is evident that the proposed device records various human physiological parameters, such as cough sounds, temperature, and pulse rate, which in turn predicts health conditions. Moreover, to increase reliability, the prediction results will be provided to the caretaker or physician of a patient through a mobile application in the near future.

5. Conclusions

In this work, a Raspberry PI-based fog computing device was utilized to predict infectious diseases in the human population. Also, different sensor parameters, namely temperature, SPO2 level, and heart rate, as well as cough sounds, were used to predict the health status of healthy individuals and COVID-19-affected patients. Cough sounds were obtained from the online KAGGLE database, and other parameters were used to form a dataset for the proposed CNN-based deep learning model. Performance metrics, such as accuracy, sensitivity, and specificity, were determined for the CNN model. From the results, it can be observed that the accuracy of the proposed CNN model is 94%. Furthermore, the sensitivity and specificity values of the proposed model are 94% and 94%, respectively. Also, the proposed CNN model was tested in the Raspberry PI-based fog computing device, and the proposed device is capable of diagnosing infectious diseases.