Review of Internet of Things-Based Artificial Intelligence Analysis Method through Real-Time Indoor Air Quality and Health Effect Monitoring: Focusing on Indoor Air Pollution That Are Harmful to the Respiratory Organ

Article information

Tuberc Respir Dis. 2023;86(1):23-32
Publication date (electronic) : 2022 October 26
doi : https://doi.org/10.4046/trd.2022.0087
Department of Software Engineering, Jeonbuk National University, Jeonju, Republic of Korea
Address for correspondence Jaehyuk Cho, Ph.D. Department of Software Engineering, Jeonbuk National University, 567 Baekje-daero, Deokjin-gu, Jeonju 54896, Republic of Korea Phone 82-63-270-4771 Fax 82-63-270-4767 E-mail chojh@jbnu.ac.kr
Received 2022 June 7; Revised 2022 September 7; Accepted 2022 October 25.

Abstract

Everyone is aware that air and environmental pollutants are harmful to health. Among them, indoor air quality directly affects physical health, such as respiratory rather than outdoor air. However, studies that have examined the correlation between environmental and health information have been conducted with public data targeting large cohorts, and studies with real-time data analysis are insufficient. Therefore, this research explores the research with an indoor air quality monitoring (AQM) system based on developing environmental detection sensors and the internet of things to collect, monitor, and analyze environmental and health data from various data sources in real-time. It explores the usage of wearable devices for health monitoring systems. In addition, the availability of big data and artificial intelligence analysis and prediction has increased, investigating algorithmic studies for accurate prediction of hazardous environments and health impacts. Regarding health effects, techniques to prevent respiratory and related diseases were reviewed.

Introduction

Many studies have shown that air pollution has a direct negative effect on human health. According to the World Health Organization, there are many toxins adversely impact health. Pollutants with the most substantial evidence for public health concern include particulate matter (PM), carbon monoxide (CO), ozone (O3), nitrogen dioxide (NO2), and Sulfur Dioxide (SO2). The PM is an significant source of health risks, as these very tiny particles can penetrate deeply into the lungs, enter the bloodstream, and travel to organs causing systemic damages to tissues and cells [1].

Chronic respiratory disease is one of the world’s leading causes of death. Every year, more than 3 million people, or 6% of the world’s deaths, die from the chronic obstructive pulmonary disease (COPD). COPD is an incurable, progressive life-threatening chronic condition that restricts airflow in the lungs and causes dysfunction and severe illness [2].

The causes of air pollution vary and appear differently depending on the situation. The main sources of outdoor pollution come from residential energy, vehicles, power generation, agriculture and waste incineration, and cooking and heating industries. The main sources of indoor pollution are cooking inefficient stoves, or burning fuels such as wood and coal in open ovens, resulting in various pollutants, including PM and volatile organic compounds (VOCs) [3]. Indoor air pollution induces significant early impairment of airway function and subclinical cardiovascular damage. A long term PM and black carbon (BC) exposure were associated with a significant burden of COPD and cardiovascular dysfunction in the case of the older participants [4]. Exposure to these environmental pollutants causes wide range of adverse health effects in adults and children [5] and even fetal [1], from respiratory diseases to cancer [6], stroke [7], cardiovascular disease [8], and premature death, and cognitive ability [9]. Many studies have been conducted comparing exposure between cities by collecting information from cohort and public environment sites, and empirical studies with experimental data are insufficient. Because of these data restrictions, there is a lack of explanation for individual-level heterogeneity. The fields of artificial intelligence (AI), such as machine learning (ML), deep learning (DL) are rapidly expanding and impacting wide ranges of industries. Recently AI has been revolutionizing health care. With the development of electronic health records, growth in computing power, continuous monitoring systems, and the availability of big data, AI technologies have the perfect platform to flourish and mature. Personalized diagnoses, solutions, prognoses, and predictions of future health outcomes have become crucial clinical decision-making tool that guides physicians and other stakeholders in doing what is best for their patients [10]. Therefore, this study aims to review internet of things (IoT)-based indoor air quality (IAQ) and health monitoring systems and AI analysis methods for the environment and health prediction.

Indoor Air Quality Monitoring System

Air pollution negatively affects human health, and several studies have confirmed that indoor air is more dangerous than outdoor air [11]. Indoor air pollution is a critical environmental health problem worldwide because half of the world’s population depends on biofuel for cooking and heating indoors [12]. For this reason, health problems caused by the increasing number of indoor air pollutants worldwide are an essential topic for discussion among researchers worldwide. Many researchers have proposed an improved indoor air quality monitoring system (IAQMS) by sensor development and verification. However, reviewing all existing and suggested IAQMS in this paper is difficulty because researchers actively working to improve air quality [13].

Air pollutants are considered the prime threat to global public health. A light scattering method capable of real-time measurement was continuously studied for monitoring PM [14]. Existing methods of measuring PM concentration include gravimetric and β-ray absorption methods and light scattering methods. The gravimetric and β-ray absorption methods are challenging to measure in real-time, and the equipment is large and heavy. In contrast, the light scattering sensor is small, light and inexpensive, and there is no need to collect dust with a filter, so that data can be easily measured [15,16]. However, there is a disadvantage to using the light scattering sensor because particle separation is complex, the error rate is high, and it is more severe when rainfall is low [17].

A compact and low-cost light scattering sensing device was developed to enable separation by PM particle size. A semiconductor laser diode was used inside, a voltage level signal was converted to a frequency level, a fast Fourier transform algorithm uses. As a result, the developed sensor overcomes the difficulty of real-time measurement and miniaturization of the existing β-ray absorption system. In addition, by connecting a smartphone through bluetooth, the PM can be monitored in real-time, and the device can be controlled [18].

To evaluate the accuracy and precision of the low-cost sensors, a standard device, metone Aerocet 531s (NaraeTek, Busan, Korea), which can count dust particles down to 0.3 μm, was compared with three low-cost laser sensors. In the case of PM1.0, the error range of all sensors is very large [19]. This study shows that detecting small particles such as PM1.0 with an inexpensive sensor is difficult. In the future research on the reliability of more accurate and sensitive sensors is needed.

Metal oxide semiconductor (MOS) is the most studied technology for CO detection. MOS sensors are susceptible, selective, robust, lightweight, and durable, and they have a fast response and recovery time, are stable and reversible, have low power consumption and have low manufacturing costs. MOS is widely used to measure and monitor trace amounts of environmentally essential gases as CO and NO2 for environment [20,21]. Both n-type and p-type MOSs are used for gas sensing, but n-type is more popular [22]. Among the n-type MOS, tin dioxide (SnO2) is the most widely used because it provides high sensitivity in the case of CO sensitivity [20].

The interference gas effect of the electrochemical ammonia sensor and NO2 sensor was studied for accurate IAQ evaluation. The ammonia’s sensor is greatly affected by hydrogen sulfide (H2S) and hydrogen (H), and SO2 and nitric oxide (NO) also affects sensor performance. In addition, the operation of the NO2 sensor is affected by all gases except hydrogen chloride (HCl). The H2S was maximal at 14 ppm, while the remaining gas values did not exceed 1 ppm, but were still affected [23]. Therefore, interference gas in the electrochemical sensor may cause an error.

The performance of electrochemical sensors of NO2 and SO2 verifies for accurate IAQ evaluation. The ppm/response time duration was calculated. In NO2 detection, hybrid material-based sensors had the highest average ratio, and in SO2 and H2S, GaN and metal oxide-based sensors were the highest [24].

Ventilation is essential for measuring IAQ. The IAQMS has been developed to identify common IAQ with modern and traditional cooking stoves. In poorly ventilated kitchens, total suspended PM is 100 times higher than the standard due to excessive smoke generation [12].

It detects levels of seven gases, including O3, PM, CO, NO2, SO2, VOCs, and CO2 to create an overall air quality warning system. In order to test the effects of various IAQ factors, the experiment was conducted by dividing the space size into church (big), class room (medium), and living room (small). Wind, location, airflow, human density, and room size affect indoor acoustic quality [25].

AirCloud, is a cloud system developed for comprehensive, low-cost personal air quality monitoring (AQM). Based on the combination of sensor data, we invented an air quality analysis engine that learns and generates air quality models. On the cloud-side, this study develops an air quality analysis engine that learns and creates air quality models based on a combination of sensor data. This engine calibrates AQM and mini-AQMs in real-time, and predicts PM2.5 concentrations. AirCloud can achieve higher accuracy at a much lower cost than previous solutions [26].

A web-based system IAQMS presents by using four sensors: gas, PM, temperature, and humidity. The data measured by each sensor transmits to the base station via the wireless sensor network (WSN) node, and a self-developed server program accessible via the internet stores the collected data [27].

To develop IoT-based indoor air pollution monitoring, CO2, NO2, and CO are monitored using low-cost gas sensors, and Raspberry-Pi processed the obtained values. The system is designed using the python coding language. The monitored values are accessible from the IoT platform. As each sensor interfaces with the Raspberry-Pi module through a different channel, it outputs in ppm. A threshold value was sets such that an alarm generates when the emission gas concentration is high [28].

To test the applicability of the comprehensive Air Quality Index (AQI), a widely used IAQ indicator.

A comprehensive air quality indicator and a close monitoring system have been developed. It responds well to real-time dynamic changes in VOCs, CO, and PM10, and is suitable as an IoT-based small-sized AQM system with low memory usage [29].

According to the developing trend of IAQMS, in the past few years, most researchers have conducted WSN-based designs with Zigbee as the most reliable communication protocol. IoT monitoring systems are considered highly reliable solutions for IAQ measurement with long battery life stable single-hop communication capabilities. Along with low delays and low power consumption, these systems also demand less effort for maintenance [13].

Health Monitoring System

Interest in person-generated wearable device data for research purposes is increasing. With the recent movement toward people (patient)-centered care and the widespread routine use of devices and technologies, person-generated health data has emerged as a promising data source for biomedical research [30].

Also, there is growing interest in reusing person-generated wearable device data for research purposes, raising concerns about data quality. However, the literatures on data quality challenges need to be improved, specifically for person-generated wearable device data [31]. Therefore, this paper reviews the health data collection and usage patterns by categorizing wearable device types.

Technological development in the wearable’s market is increasing exponentially. Personal health monitoring and physical activity (PA) are popular among all age groups and medical communities [32]. In addition, regular PA is effective in preventing and managing chronic diseases such as cardiovascular disease, hypertension, diabetes, and obesity [33]. Among them, the wrist-worn type is the most significant development.

A remote health monitoring and support system was developed using information and communication technology to access and manage physical status and PA levels in home-care patients with COPD. A study using an iPad (Apple, Cupertino, CA, USA) as a computing device and a developed utility dealing with the input and transfer of the following data and six assessment items related to the symptom (cough, phlegm, breathing, sleep, appetite, vitality), number of steps per day, and energy consumption. This application enables remote monitoring for medical personnel such as doctors and nurses, prevents exacerbations. Of COPD and helps with early detection and treatment of acute exacerbations. In addition, the system can provide lifestyle guides tailored to individual lifestyles and medical conditions [34].

Fractional exhaled nitric oxide (FeNO) is a non-invasive indicator of airway inflammation in asthma. Recent studies have shown that FeNO is a potential consequence of COPD. Recently, a new portable FeNO analyzer (NIOX MINO, Aerocrine, Solna, Sweden) has been developed. Qualification of FeNO was compared in COPD patient’s and healthy subjects over a short period using NIOX MINO and standard chemiluminescence analysis (NOA, Sensormedics, Yorba Linda, CA, USA) in COPD patients and healthy people. While there was no significant difference in the short period, COPD patients showed greater variability in the long term. This is associated with a significantly exacerbation rate. Also, a FeNO electrochemical portable analyzer is available in COPD, which shows positive consistency with a stationary chemiluminescence analyzer [35].

Data were monitored for 43 patients wearing basis peak for 5 months. As a result, it was possible to identify physiological differences between health conditions among individuals [36].

The system is proposed to measure environmental factors, notify workers with warnings and vibrations to protect safety in case of danger to danger, and send situational information to the control center to take immediate action. It integrated the Galaxy Watch’ Biosensor (Samsung, Seoul, Korea), Apple Watch (electrocardiogram, heart rate, saturation of percutaneous oxygen [SpO2]), gas sensor (CO2, VOCs), and wireless communication technology. For accurate measurement and analysis, multi-level risk modeling and ML techniques must be included [37].

A surface electromyogram electrode with 10 and 24 mm diameters was developed by screen printing poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) ink on 100% cotton fabric. The larger the diameter, the lower the resistance value [38].

A wireless patch-type wearable pulse oximeter has been developed to measure heart rate and SpO2 by reflecting light sources of two wavelengths (red 625 nm, infrared 865 nm) on a person’s forehead. The size of the flexible circuit is 7 cm×2 cm, interfaced with an 8.5 cm×3.5 cm wireless system board. Weighing only 15 g, it is an elastic band that can be easily wearable on the forehead. The SpO2 calculates the ratio of the amplitude of the infrared and red photoluminescence (PPG) signals by detecting and measuring the amplitude change of the PPG signals to red and infrared light due to changes in blood oxygen saturation. SpO2 values measured using our system were consistent with commercial non-abrasion pulse oximeters in both normal and inhale/exhale conditions [39].

These wearable bio-signal monitoring sensors are used by adhering to the human body, so it is essential to surface-treat materials that are harmless to skin contact. If the surface is treated with a flexible polymer material such as polydimethylsiloxane (PDMS), as the electrode does not directly adhere to the skin, it can safely sense a bio-signal. However, PDMS has a high moisture permeation rate, so when adhered to for a long time, there is a risk that body fluids such as human sweat penetrate and the sensor is oxidized [40].

COPD is one of most common respiratory diseases. A new diagnostic method is developed to detect COPD parameters using a microelectromechanical system based acceleration sensor. It records the acceleration data of the movements of the diaphragm in three axes during breathing. With the proposed device in this work, parameters such as tidal volume capacity, forced vital capacity, and respiratory rate commonly associated with COPD are successfully measured. The measurement results were similar to the spirometer and can be considered as an alternative instrument for the spirometer [41].

There is a study of wearable devices along with clothing type. Conventional clothing is soft and flexible and can be draped to our bodies [42]. It should be washable for reuse, but there are technical difficulties due to the fiber material containing hard and non-washable electronic products and electrical materials [43].

All smart electro-clothing systems consist of hardware and software, primarily electronic, and non-fiber materials such as sensing subsystem, action subsystem, control subsystem, communications subsystem, location subsystem, power subsystem, storage subsystem, and display subsystem and common elements [43,44].

A smart jacket is designed to secure the coal miners’ life. This prototype senses the various health-related parameters, i.e., the presence of hazardous gas, pulse rate of miner, updated temperature/humidity, exact depth, and geographical location of the miner. All of these parameters transmit through Wi-Fi to the internet protocol. All miners were monitored. In addition, in the event of a disaster, the miners’ lives can be secured immediately. This designed wearable embedded system will send the last global positioning system location to a specific internet protocol and continuously update the miner’ pulse rate detected by the pulse sensor to the control system [45].

Photonic textiles that can measure pulse oximetry are sewn into gloves for subtypes. SpO2 is measured to change the amount of transmitted light by placing an optical sensor on the fingertip. By measuring the amount of light transmitted by two different wavelengths (HeNe laser, halogen lamp) with an optical sensor, oxygenated and deoxygenated hemoglobin can be calculated to obtain oxygen saturation in the body. Using photonic textiles is feasible for pulse oximetry [46].

AI Analysis for Air Quality Prediction

Research for air quality prediction has been conducted based on various algorithm models. Most of the studies’ settings are primarily divided into comparing AI analysis techniques and developing new algorithms. Therefore, this paper reviews it in two ways.

First, it reviews the comparison of AI analysis techniques for air quality prediction. The study was conducted to determine a predictive model for determining air pollution based on PM10 and PM2.5pollution concentrations in Tehran as ML methods for air pollution prediction were used by support vector regression (SVR), geographically weighted regression, artificial neural networks (ANN), and autoregressive nonlinear neural networks using external inputs have been used. The most reliable algorithm for air pollution prediction is an autoregressive nonlinear neural network with external input using the proposed prediction model, which has a one day prediction error of 1.79 μg/m3 [47].

To predict PM and BC, a transportation-related air pollution factor, ML model performance was compared. This study investigates the land use regression (LUR)’ boundaries approaches and the potential of two different ML models: ANN and gradient boost. Models were developed for PM performing better than those for BC. For the same contaminants, ANN and extreme gradient boosting (XGBoost) models showed better performance than LUR [48].

In order to assess PM prediction performance, it was compared ANN with multiple linear regression (MLR) models. The model’s input data was PM10 concentration and variables for the weather. As a result of comparing the two models, the nonlinear ANN method showed better performance for predicting of PM10 [49].

The PM10 concentration in Seoul was predicted using weather factors as an input dataset of MLR, support vector machine (SVM), and random forest (RF) models, and the performance of the model was compared and evaluated. The model’s input dataset comprised nine meteorological factors from the automatic weather system (AWS): temperatures, precipitation, wind speeds, wind direction, yellow dust, and relative humidity. The prediction performance of the ensemble model RF was the highest, followed by the relative humidity and yellow dust, which contributed significantly to the predictive performance of all models, and the maximum temperature and average wind speed showed relatively low. In case of Gwanak-gu and Gangnam-daero which are relatively close to air quality monitoring sites (AQMS) and AWS, SVM, and RF models were highly accurate according to the model validations. By contrast, both models could have performed better in Yongsan-gu, which is relatively far from AQMS and AWS. The results indicate that AQMS and AWS adjacencies significantly affect PM10 concentration prediction [50].

To compare the performance of the PM concentration prediction algorithm, MLR, SVR, autoregressive integrated moving average (ARIMA), and autoregressive integrated moving average with explanatory variable (ARIMAX) was compared. It was evaluated by root mean square error (RMSE) using air quality data and meteorological data. In the integrated concentration prediction, the performance of SVR was superior to MLR, and in the time series prediction, the performance of ARIMAX was superior to that of ARIMA [51].

The study performs a traditional model k-nearest neighbors (k-NN) and logistic regression (LR) and a non-traditional long-short term memory (LSTM) network-based DL algorithm for the creation of alert messages regarding to bin status and predicting the amount of air pollutant CO presence in the air at a specific instance. In real-time testing for predicting bin status, LR and k-NN are 79% and 83%, respectively. The accuracy of modified LSTM and simple LSTM models is 90% and 88%, respectively, to predict the future gas concentration in the air. The system provided real-time monitoring of garbage levels along with notifications from an alert mechanism and improved accuracy by utilizing ML [52].

Second, it is a review of developing algorithms for quality air prediction. A data mining algorithm was developed using ANN and k-NN to implement accurate PM prediction models. For ANN operation, a network, constructs with 13 nodes in the input layer, 15 nodes in the hidden layer, and one node in the output layer. The output was classified using the k-NN algorithm and had high accuracy when K=9. The proposed model showed a better prediction rate than ANN and k-NN [53].

A separation prediction model for PM based on a deep neural network (DNN) was designed to improve PM10 prediction accuracy. In order to select the optimal hyperparameters, 3,600 candidate parameters were set for each model through the grid search technique. In this process, to select a hyperparameter value with a high generalization performance, the hyperparameter search was performed by setting the number of folds of k-fold, which is one of the cross-validation methods, to 3. In addition, for performance comparison with the proposed concentration-specific separation prediction model, the hyperparameter optimization of the DNN-based model was performed [54].

MLR and ANN designed prediction models, and the suitability of algorithms for PM prediction was evaluated by comparison with real data. MLR and ANN were compared with real-world data to access algorithm’s suitability for PM prediction. In the case of algorithm PM prediction, ANN was superior in performance. When designing a PM prediction model using ANN, the composition of a hidden layer with an appropriate number of neurons is essential [55].

The PM10 concentration prediction algorithm was modeled using weather and traffic-related air pollutants concentration variables such as CO, NO, and nitrogen (N) data. A generalized additive model was developed and evaluated. This study, identified weather variables such as temperature and wind speed as major control factors for PM10 concentration, but traffic-related air pollutants and PM10 concentrations showed a weak relationship. Therefore road traffic is not the leading cause of PM [56].

Despite the abundance of studies on PM2.5 and PM10 estimations from satellite remote sensing, only a few studies have been conducted on PM1.0 by using satellite observations. Thus, this study estimated hourly PM1.0 concentrations in China using an integrated principal component analysis (PCA) and hybrid generalized regression neural network (GRNN) model that combines ground-based observations of PM2.5 with a geostationary satellite Himawari-8 (Mitsubishi Electric, Tokyo, Japan) aerosol optical depth data. Fusing PM2.5 data was advantageous for the continuous spatiotemporal estimation of PM1.0, and the estimation accuracy of each model was significantly improved. Specifically, the R2 of MLR increased from 0.21 to 0.38, and the GRNN and PCA-integrated GRNN models improved by 8% and 6%, respectively. A comparison of the linear regression model and GRNN models (including PCA-integrated GRNN) showed that the nonlinear model could determine the potential relationship between PM and predictors [57]. Due to the absence of low-cost, high-quality PM1.0 sensors, prediction of PM1.0 AI analysis is an important research topic.

A study proposes an air quality prediction system ‘gated recurrent units (GRU),’ using six atmospheric sensor data (VOCs, CO2, PM, temperature, humidity, and light quantity) and DL models. The predictive accuracy performance of the proposed GRU model compared to other models, such as LSTM networks and linear regression. The proposed system performed better with 85% higher accuracy for various parameters [58].

AI Analysis for Health Effect Prediction

Research for health effect prediction has been conducted based on various algorithm models. The settings of most studies are divided mainly into a comparison of AI analysis techniques and the development of new algorithms. Therefore, this paper reviews it in two ways.

First, it reviews a comparison of AI analysis techniques for health effect prediction. In order to predict sepsis mortality, a study compares conventional context-based LR approaches with four ML techniques: least absolute shrinkage and selection operator regularization, RF, XGbost, and DNN. All four ML models showed higher sensitivity, specificity, positive prediction, and negative prediction values than the LR model [59].

When the most accurate predicted model is the goal, ML algorithms are more advantageous than conventional regression methods when the most accurate predicted model is the goal. When using ML methods, special attention requires in the form of model validation. The usefulness of solving individual problems varies, so comparison with multiple approaches is required, and the criterion for how much flexibility can be allowed becomes the ultimate modeling technique [60].

In order to predict the frequency of asthma, it was analyzed through three predictive models: SVM, neural net, and DL based on asthma-causing lifestyle, eating habits, environmental characteristics, and essential characteristics. The model’s predictive ability was compared based on the model’s accuracy, RMSE, and mean absolute error (MAE). SVM has a significant accuracy of 93.19%, but RMSE 0.320 and MAE 0.300 indicators could be better. The DL’s evaluation results are an accuracy of 74.78%, RMSE 0.252, and MAE 0.120, which are generally good. In contrast, the neural net model is quite good, with an accuracy of 93.19%, and RMSE 0.251 and MAE 0.124 indicators are also quite good. The neural net model is the best prediction model for asthma because the model is learned by feedforward neural networks learned by the backpropagation algorithm [61].

Second, it is a review of the development of algorithms for the prediction of health effects. The automated device for asthma monitoring and management, a wearable IoT sensor smart device, was used to collect general conditions such as the patient’s physical condition, body temperature, emotion, heart rate, respiratory status, and behavior. A patient monitoring system based on Iterative Golden section optimized Deep Belief neural Network (IGDBN) using MATLAB for the collected data is developed. The developed IGDBN guarantees a higher precision and a higher MCC value with a lower error rate than DNN, a hybrid RF with a linear model, a long-short term neural network, and a fuzzy rule-based neural classifier [62].

An electronic stethoscope was developed through AI analysis of acoustic medical data by measuring lung sound. The device divides into three parts: sounds collection module (SCM), e-healthcare home gateway (eHG), and smartphone. SCM records heart and lung sounds, and eHG communicates with data translation, cloud servers, and mobile devices interface. For lung sound analysis, first perform a short-time fourier transformation (STFT) of the sound file and output signals for further classification. Use convolutional neural network (CNN) and k-NN models for classification. After the STFT image is loaded, it is first converted into a gray-tone image, and then used CNN model. The last extracted features use as input to the k-NN model for the final classification [63].

The existing auscultation through the stethoscope may cause interpretation errors due to the subjective approach of a doctor. Therefore, objective evaluation is required using ML to detect wheezing. This study proposed an LSTM-based neural network, a novel wheezing detection model that distinguishes normal and wheezing. Mel-frequency cepstral coefficients (MFCC) uses as the feature extraction method. A simulation performs through MATLAB R2020a. The performance of the proposed model was evaluated and compared with the existing multilayer perceptron, a widely used neural network that has proven efficiency in traditional respiratory sound classification. LSTMs are an up-andcoming alternative to feedforward networks since they provide relatively better results [64].

It is difficult for human listeners because some lung sound events have a frequency spectrum beyond human hearing ability. Thus, this paper proposed a system capable of detecting and classifying abnormal lung sounds, such as crackles or wheezes. CNN was used to detect and classify adventitious sounds in lung sound signals successfully. Various functions (power spectrum density [PSD], mel spectrum [MS], and MFCC) for converting lung sound signals into 2D images were presented. MS, when fed into CNN, achieves results in line with the current cutting-edge technology and is followed by PSD and MFCC [65].

Conclusion

The reliability of IAQ and health effect sensors has been demonstrated through many studies. However, in the case of PM1.0 with small particles, large equipment must be used, and real-time prediction is difficult through small and portable sensors. Although it is possible to predict PM1.0 through AI analysis of PM2.5 data with large particles, more research on precise sensors that can be directly measured is needed. In addition, the reliability of electrochemical sensors should be studied in the future by overcoming the interference gas effect. IoT-based real-time monitoring is efficient to monitor, collecting, and analyzing more accurate air quality, and in the future, AI analysis needs to improve precision and accuracy.

In addition, research on the convergence of medical and AI has recently continued. This is because a doctor’s subjective judgment increases the demand for accurate analysis and the predictive power of AI. Many AI-based studies have been conducted to predict lung disease from health effect data, and mortality from lung disease. In addition, a recent increase in demand for telemedicine has led to an increase in research on the development of remote healthcare services.

Notes

Authors’ Contributions

Conception or design of the work: Mun E, Cho J. Writing and/or reviewing the preliminary and definitive versions: Mun E. Approved the final version: all authors.

Conflicts of Interest

No potential conflict of interest relevant to this article was reported.

Funding

This work was supported Korea Environmental Industry & Technology Institute (KEITI) grant funded by the Korean government (Ministry of Environment). Project No. RE202101551, the development of IoT-based technology for collecting and managing big data on environmental hazards and health effects.

This research was also supported by an Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korean government (MIST) (No. 2019-0-00135, Implementation of 5G-based Smart Sensor Verification Platform).

This paper was supported by research funds from Jeonbuk National University.

References

1. World Health Organization. Air quality and health [Internet]. Geneva: WHO; 2022. [cited 2022 Nov 30]. Available from: https://www.who.int/teams/environment-climate-change-and-health/air-quality-and-health/health-impacts.
2. Bousquet J, Kaltaev N. Global surveillance prevention and control of chronic respiratory diseases: a comprehensive approach [Internet]. Geneva: World Health Organization; 2007. [cited 2022 Nov 30]. Available from: https://apps.who.int/iris/handle/10665/43776.
3. World Health Organization. Air pollution [Internet]. Geneva: WHO; 2018. [cited 2022 Nov 30]. Available from: https://www.who.int/health-topics/air-pollution#tab=tab_1.
4. Pratali L, Marinoni A, Cogo A, Ujka K, Gilardoni S, Bernardi E, et al. Indoor air pollution exposure effects on lung and cardiovascular health in the High Himalayas, Nepal: an observational study. Eur J Intern Med 2019;61:81–7.
5. Currie J, Zivin JG, Mullins J, Neidell M. What do we know about short- and long-term effects of early-life exposure to pollution? Annu Rev Resour Economics 2014;6:217–47.
6. Boffetta P. Human cancer from environmental pollutants: the epidemiological evidence. Mutat Res 2006;608:157–62.
7. Li XY, Yu XB, Liang WW, Yu N, Wang L, Ye XJ, et al. Meta-analysis of association between particulate matter and stroke attack. CNS Neurosci Ther 2012;18:501–8.
8. Franklin BA, Brook R, Arden Pope C. Air pollution and cardiovascular disease. Curr Probl Cardiol 2015;40:207–38.
9. Zhang X, Chen X, Zhang X. The impact of exposure to air pollution on cognitive performance. Proc Natl Acad Sci U S A 2018;115:9193–7.
10. Alamgir A, Mousa O, Shah Z. Artificial intelligence in predicting cardiac arrest: scoping review. JMIR Med Inform 2021;9e30798.
11. Cincinelli A, Martellini T. Indoor air quality and health. Int J Environ Res Public Health 2017;14:1286.
12. Parajuli I, Lee H, Shrestha KR. Indoor air quality and ventilation assessment of rural mountainous households of Nepal. Int J Sustain Built Environ 2016;5:301–11.
13. Saini J, Dutta M, Marques G. A comprehensive review on indoor air quality monitoring systems for enhanced public health. Sustain Environ Res 2020;30:1–12.
14. Kim SJ, Son YS, Gang HS, Kim JC, Lee JH, Kin GS, et al. Compensation of particulated matter measurement by light scattering method. In : Proceedings of the 49th Meeting of KOSAE 2009: Korea Air Pollution Research Association Conference; 2009 Oct 22-24; Gunsan. Seoul: Korean Society for Atmospheric Environment; 2009. 613–5.
15. Kim SJ, Kang HS, Son YS, Yoon SL, Kim JC, Kim GS, et al. Compensation of light scattering method for real-time monitoring of particulate matters in subway stations. J Korean Soc Atmos Environ 2010;26:533–42.
16. Lee S, Lee JK. Visualization of the comparison between airborne dust concentration data of indoor rooms on a building model. J Korean Hous Assoc 2015;26:55–62.
17. Kim JH, Oh J, Choi JS, Ahn JY, Yoon G, Park JS. A study on the correction factor of optic scattering PM2.5 by gravimetric method. J Korean Soc Urban Environ 2014;14:41–7.
18. Lee N, Um HU, Cho HS. Development of detection and monitoring by light scattering in real time. Fire Sci Eng 2018;32:134–9.
19. Nguyen NH, Nguyen HX, Le TTB, Vu CD. Evaluating lowcost commercially available sensors for air quality monitoring and application of sensor calibration methods for improving accuracy. Open J Air Pollut 2021;10:1–17.
20. Nandy T, Coutu RA, Ababei C. Carbon monoxide sensing technologies for next-generation cyber-physical systems. Sensors (Basel) 2018;18:3443.
21. Fine GF, Cavanagh LM, Afonja A, Binions R. Metal oxide semi-conductor gas sensors in environmental monitoring. Sensors (Basel) 2010;10:5469–502.
22. Kim HJ, Lee JH. Highly sensitive and selective gas sensors using p-type oxide semiconductors: overview. Sens Actuators B Chem 2014;192:607–27.
23. Majder-Lopatka M. Effects of interfering gases in electrochemical sensors NH3 and NO2. In MATEC Web Conf 2018;247:00023.
24. Khan MA, Rao MV, Li Q. Recent advances in electrochemical sensors for detecting toxic gases: NO2, SO2 and H2S. Sensors (Basel) 2019;19:905.
25. Kim JY, Chu CH, Shin SM. ISSAQ: An integrated sensing systems for real-time indoor air quality monitoring. IEEE Sens J 2014;14:4230–44.
26. Cheng Y, Li X, Li Z, Jiang S, Li Y, Jia J, et al. AirCloud: a cloud-based air-quality monitoring system for everyone. In : Proceedings of the 12th ACM Conference on Embedded Network Sensor Systems; 2014 Nov 3-6; Memphis. New York: Association for Computing Machinery; 2014. 251–65.
27. Saad SM, Saad ARM, Kamarudin AMY, Zalkaria A, Shakaff AY. Indoor air quality monitoring system using wireless sensor network (WSN) with web interface. In : 2013 International Conference on Electrical, Electronics and System Engineering (ICEESE); 2013 Dec 4-5; Kuala Lumpur. Piscataway: IEEE; 2013. 60–4.
28. Sivasankari B, Prabha CA, Dharini S, Haripriya R. IoT based indoor air pollution monitoring using raspberry pi. Int J Innov Eng Tech 2017;9:16–21.
29. Kang JH, Hwang KI. A comprehensive real-time indoor air-quality level indicator. Sustainability 2016;8:881.
30. Nittas V, Mutsch M, Ehrler F, Puhan MA. Electronic patient-generated health data to facilitate prevention and health promotion: a scoping review protocol. BMJ Open 2018;8e021245.
31. Cho S, Ensari I, Weng C, Kahn MG, Natarajan K. Factors affecting the quality of person-generated wearable device data and associated challenges: rapid systematic review. JMIR Mhealth Uhealth 2021;9e20738.
32. Strath SJ, Rowley TW. Wearables for promoting physical activity. Clin Chem 2018;64:53–63.
33. Bauer UE, Briss PA, Goodman RA, Bowman BA. Prevention of chronic disease in the 21st century: elimination of the leading preventable causes of premature death and disability in the USA. Lancet 2014;384:45–52.
34. Ohashi C, Akiguchi S, Ohira M. Development of a remote health monitoring system to prevent frailty in elderly home-care patients with COPD. Sensors (Basel) 2022;22:2670.
35. de Laurentiis G, Maniscalco M, Cianciulli F, Stanziola A, Marsico S, Lundberg JO, et al. Exhaled nitric oxide monitoring in COPD using a portable analyzer. Pulm Pharmacol Ther 2008;21:689–93.
36. Li X, Dunn J, Salins D, Zhou G, Zhou W, Schussler-Fiorenza Rose SM, et al. Digital health: tracking physiomes and activity using wearable biosensors reveals useful health-related information. PLoS Biol 2017;15e2001402.
37. Kim HK, Moon SJ. Design and implementation of an active risk situation estimation system in smart healthcare using bio and environmental sensors. J Korean Inst Commun Inf Sci 2020;45:914–25.
38. Pani D, Achilli A, Spanu A, Bonfiglio A, Gazzoni M, Botter A. Validation of polymer-based screen-printed textile electrodes for surface EMG detection. IEEE Trans Neural Syst Rehabil Eng 2019;27:1370–7.
39. Azhari A, Yoshimoto S, Nezu T, Iida H, Ota H, Noda Y, et al. A patch-type wireless forehead pulse oximeter for SpO2 measurement. In : 2017 IEEE Biomedical Circuits and Systems Conference (BioCAS); 2017 Oct 19-21; Turin. Piscataway: IEEE; 2017. 1–4.
40. Jang E, Cho G. The classification and investigation of smart textile sensors for wearable vital signs monitoring. Fash Text Res J 2019;21:697–707.
41. Sumbul H, Yuzer AH. Development of diagnostic device for COPD: a MEMS based approach. Int J Comput Sci Netw Secur 2017;17:196–203.
42. Sayem A, Haider J. An overview on the development of natural renewable materials for textile applications. In : Hashmi S, Choudhury IA, eds. Encyclopedia of renewable and sustainable materials Amsterdam: Elsevier; 2020. p. 822–38.
43. Muhammad Sayem AS, Hon Teay S, Shahariar H, Fink PL, Albarbar A. Review on smart electro-clothing systems (SeCSs). Sensors (Basel) 2020;20:587.
44. Fernandez-Carames TM, Fraga-Lamas P. Towards the Internet of smart clothing: a review on IoT wearables and garments for creating intelligent connected e-textiles. Electronics 2017;7:405.
45. Abro GEM, Shaikh SA, Soomro S, Abid G, Kumar K, Ahmed F. Prototyping IOT based smart wearable jacket design for securing the life of coal miners. In : 2018 International Conference on Computing, Electronics & Communications Engineering (iCCECE); 2018 Aug 16-17; Southend. Piscataway: IEEE; 2018. 134–7.
46. Rothmaier M, Selm B, Spichtig S, Haensse D, Wolf M. Photonic textiles for pulse oximetry. Opt Express 2008;16:12973–86.
47. Delavar MR, Gholami A, Shiran GR, Rashidi Y, Nakhaeizadeh GR, Fedra K, et al. A novel method for improving air pollution prediction based on machine learning approaches: a case study applied to the capital city of Tehran. ISPRS Int J Geoinf 2019;8:99.
48. Wang A, Xu J, Tu R, Saleh M, Hatzopoulou M. Potential of machine learning for prediction of traffic related air pollution. Transp Res D Transp Environ 2020;88:102599.
49. Chaloulakou A, Grivas G, Spyrellis N. Neural network and multiple regression models for PM10 prediction in Athens: a comparative assessment. J Air Waste Manag Assoc 2003;53:1183–90.
50. Son S, Kim J. Evaluation and predicting PM10 concentration using multiple linear regression and machine learning. Korean J Remote Sens 2020;36:1711–20.
51. Joun S, Choi J, Bae J. Performance comparison of algorithms for the prediction of fine dust concentration. In : 2017 Proceedings of Korea Computer Congress; 2017 Jun 18- 20; Jeju. Seoul: The Korean Institute of Information Scientists and Engineers; 2017. 775–7.
52. Hussain A, Draz U, Ali T, Tariq S, Irfan M, Glowacz A, et al. Waste management and prediction of air pollutants using IoT and machine learning approach. Energies 2020;13:3930.
53. Cha J, Kim J. Development of data mining algorithm for implementation of fine dust numerical prediction model. J Korea Inst Inf Commun Eng 2018;22:595–601.
54. Cho KW, Jung Y, Lee JS, Oh CH. Separation prediction model by concentration based on deep neural network for improving PM10 forecast accuracy. J Korea Inst Inf Commun Eng 2020;24:8–14.
55. Cho KW, Jung Y, Kang CG, Oh CH. Conformity assessment of machine learning algorithm for particulate matter prediction. J Korea Inst Inf Commun Eng 2019;23:20–6.
56. Munir S, Habeebullah TM, Seroji AR, Morsy EA, Mohammed AM, Saud WA, et al. Modeling particulate matter concentrations in Makkah, applying a statistical modeling approach. Aerosol Air Qual Res 2013;13:901–10.
57. Zang L, Mao F, Guo J, Wang W, Pan Z, Shen H, et al. Estimation of spatiotemporal PM1.0 distributions in China by combining PM2.5 observations with satellite aerosol optical depth. Sci Total Environ 2019;658:1256–64.
58. Ahn J, Shin D, Kim K, Yang J. Indoor air quality analysis using deep learning with sensor data. Sensors (Basel) 2017;17:2476.
59. Park JY, Hsu TC, Hu JR, Chen CY, Hsu WT, Lee M, et al. Predicting sepsis mortality in a population-based national database: machine learning approach. J Med Internet Res 2022;24e29982.
60. Goldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J 2017;38:1805–14.
61. Noh MJ, Park SC. A prediction model of asthma diseases in teenagers using artificial intelligence models. J Inf Technol Appl Manag 2020;27:171–80.
62. Fouad H, Hassanein AS, Soliman AM, Al-Feel H. Analyzing patient health information based on IoT sensor with AI for improving patient assistance in the future direction. Measurement 2020;159:107757.
63. Chang KC, Huang JW, Wu YF. Design of e-health system for heart rate and lung sound monitoring with AI-based analysis. In : 2021 IEEE International Conference on Consumer Electronics-Taiwan (ICCE-TW); 2021 Sep 15-17; Penghu. Piscataway: IEEE; 2021. 1–2.
64. Semmad A, Bahoura M. Long short term memory based recurrent neural network for wheezing detection in pulmonary sounds. In : 2021 IEEE International Midwest Symposium on Circuits and Systems (MWSCAS); 2021 Aug 9-11; Lansing. Piscataway: IEEE; 2021. 412–5.
65. Faustino P, Oliveira J, Coimbra M. Crackle and wheeze detection in lung sound signals using convolutional neural networks. Annu Int Conf IEEE Eng Med Biol Soc 2021;2021:345–8.

Article information Continued