PaperlayerView exact revision on arXiv

Source-linked AI summary

Smart Healthcare in the Age of AI: Recent Advances, Challenges, and Future Prospects

Mahmoud Nasr, MD. Milon Islam, Shady Shehata, Fakhri Karray, Yuri Quintana

arXiv:2107.03924v1cs.CYcs.AIcs.HCcs.MAcs.NI

TL;DR

Smart healthcare research spans monitoring, diagnosis, ambient assisted living, and software integration, while the reviewed systems face privacy, security, and energy challenges. This review compares approaches across those areas and finds COVID-19 diagnostic accuracy varies across reviewed systems, while identifying smart healthcare as a potentially lower-cost care option.

  • Problem

    Reviewed systems face patient-data security and privacy challenges, and wearable and smartphone monitoring also raises power-consumption and energy-efficiency concerns.

    Source: DIRECTIONS · p. 18

  • Method

    The paper systematically reviews smart medical devices, machine learning for disease prediction, AAL, and software architectures, comparing approaches, advantages, and challenges.

    Source: §I · p. 2

  • Results

    Across reviewed domains, the paper synthesizes monitoring, diagnosis, AAL support, and integration architectures; COVID-19 diagnostic accuracy ranges from 66.67% to 99.06%, and almost all reviewed systems are not practically used in the target environment.

    Source: A. OPEN DISCUSSIONS · pp. 16–17

  • Takeaways & Limitations

    The review presents smart healthcare as an easily deployable, lower-cost alternative to hospital or assisted-living care that can provide quality healthcare services.

    Source: §VII · p. 20

    Wearable and smartphone systems must address patient-record security and privacy as well as low power consumption and energy efficiency for long-term monitoring.

    Source: DIRECTIONS · p. 18

Abstract

from arXiv · show

The significant increase in the number of individuals with chronic ailments (including the elderly and disabled) has dictated an urgent need for an innovative model for healthcare systems. The evolved model will be more personalized and less reliant on traditional brick-and-mortar healthcare institutions such as hospitals, nursing homes, and long-term healthcare centers. The smart healthcare system is a topic of recently growing interest and has become increasingly required due to major developments in modern technologies, especially in artificial intelligence (AI) and machine learning (ML). This paper is aimed to discuss the current state-of-the-art smart healthcare systems highlighting major areas like wearable and smartphone devices for health monitoring, machine learning for disease diagnosis, and the assistive frameworks, including social robots developed for the ambient assisted living environment. Additionally, the paper demonstrates software integration architectures that are very significant to create smart healthcare systems, integrating seamlessly the benefit of data analytics and other tools of AI. The explained developed systems focus on several facets: the contribution of each developed framework, the detailed working procedure, the performance as outcomes, and the comparative merits and limitations. The current research challenges with potential future directions are addressed to highlight the drawbacks of existing systems and the possible methods to introduce novel frameworks, respectively. This review aims at providing comprehensive insights into the recent developments of smart healthcare systems to equip experts to contribute to the field.

I. INTRODUCTION

The review surveys smart healthcare research spanning health monitoring, disease diagnosis, ambient assisted living, and software integration architectures. It compares approaches and their advantages and challenges, and identifies recommendations for future work.

I. INTRODUCTION

  • Rising chronic disease, healthcare demand, workforce pressure, and costs motivate more affordable and inclusive care using sensors, communication technologies, and AI.The introduction cites a projection that 22% of the population will be aged 60 or older by 2050.Source: §I · p. 1
  • The review covers wearable and smartphone monitoring, machine-learning disease prediction, assistive frameworks including social robots, and software architectures, comparing approaches and outlining improvement recommendations.Its overview organizes monitoring, diagnosis, and AAL support as major areas, with architectures discussed as an additional component.Source: §I · p. 2

II. HEALTHCARE MONITORING DEVICES IN IOT

  • The review describes IoT health monitoring in which sensors send patient data to cloud processing and physicians or family monitor status through web or mobile interfaces.Wi-Fi or a LoRa gateway may transfer data, and emergency services may handle critical situations.Source: §II · p. 2

A. WEARABLE DEVICES FOR HEALTH MONITORING

  • Wearable systems monitor vital signs remotely, but reviewed prototypes vary in readiness, emergency response, security, analytics, and sensor coverage.Examples include systems not fabricated for real-time testing, lacking emergency handling, or lacking security measures,.Source: A. WEARABLE DEVICES FOR HEALTH… · p. 4
  • The reviewed wearable systems measure parameters such as heart rate, temperature, ECG, oxygen saturation, and blood pressure, using varied edge devices, communication channels, and visualization tools.Table 1 compares sensors, edge devices, communication channels, visualization tools, and system comments.Source: A. WEARABLE DEVICES FOR HEALTH… · p. 4

B. SMARTPHONE SOLUTIONS FOR HEALTH MONITORING

Smartphone sensors and applications support portable monitoring of several vital signs, but reviewed systems vary in measured signals and report practical limitations. The review spans glucose, blood-pressure, respiratory, and heart-rate estimation.

B. SMARTPHONE SOLUTIONS FOR HEALTH MONITORING

  • Smartphone monitoring can use built-in sensors and applications to measure heart rate, blood pressure, oxygen levels, and respiratory rate while offering portability and lower additional cost.The review also notes that a limited number of sensors may not capture a patient's complete status.Source: B. SMARTPHONE SOLUTIONS FOR HEALTH… · pp. 4–5
  • Zhang et al. reported 81.49% accuracy for smartphone-video blood-glucose estimation, which the review deemed lower than needed for reliable monitoring.The system processed phone-collected data in the cloud and classified estimated glucose into normal, warning, and borderline ranges.Source: B. SMARTPHONE SOLUTIONS FOR HEALTH… · p. 5
  • 97.8% average accuracy and 2.2% average error were reported for smartphone-based respiratory-rate estimation, although temperature, skin colour, and condition effects were not considered.The method used smartphone imaging and Fourier-transform techniques with plethysmography data.Source: B. SMARTPHONE SOLUTIONS FOR HEALTH… · p. 5
  • Tabei et al. [25] reported blood-pressure mean absolute error, standard deviation, and correlation of 2.10, 1.96, and 0.90, respectively; Dey et al. reported precision below the standard value.The review also notes that Seismo was not able to monitor blood pressure continuously.Source: B. SMARTPHONE SOLUTIONS FOR HEALTH… · p. 6
  • 99.7% accuracy and 0.04–0.3 beats/min absolute error were reported for a smartphone-camera heart-rate method, while another system estimated heart rate in real time with less than 5% error.The latter system's reported limitation was that it ignored differences among smartphone camera modules.Source: B. SMARTPHONE SOLUTIONS FOR HEALTH… · p. 6
  • The reviewed smartphone studies varied in monitored signs, sensors, phone model, participant count, video length, sampling rate, and reported comments.One reviewed system reported mean absolute percentage error of 2.965 for vital-sign monitoring, but did not show a real-time prototype.Source: B. SMARTPHONE SOLUTIONS FOR HEALTH… · p. 6

LEARNING IN IOHT

The Internet of Health Things combines connected devices with machine learning for healthcare data analysis. The described diagnosis workflow moves from benchmark or sensor data through preprocessing and feature extraction to a decision for the concerned person.

LEARNING IN IOHT

  • The Internet of Health Things (IoHT) comprises interlinked devices that share and handle data to enhance patient health.The passage also refers to the Internet of Medical Things (IoMT) as an alternative name.Source: LEARNING IN IOHT · p. 6
  • Machine learning is described as enabling IoT devices to perform information inference, data analytics, and intelligence across contexts from cloud computing to smart sensors.Source: LEARNING IN IOHT · p. 6
  • The diagnosis workflow uses benchmark or real-time sensor data sent to the cloud, then preprocesses data, extracts features, applies machine-learning techniques, and transfers decisions to concerned people.Source: LEARNING IN IOHT · p. 6

A. NOVEL CORONAVIRUS (COVID-19)

The review surveys machine-learning frameworks for COVID-19 diagnosis in IoT environments, motivated by the disease's communicable nature. Reported results vary across systems, datasets, and approaches.

A. NOVEL CORONAVIRUS (COVID-19)

  • COVID-19 is presented as a public-health crisis because of its communicable nature, and the review focuses on machine-learning diagnosis frameworks in IoT environments.Source: A. NOVEL CORONAVIRUS (COVID-19) · p. 6
  • 99.06% accuracy for the multiclass scenario was the highest reported COVID-19 result, while 66.67% was the lowest among the reviewed systems.The values are reported for Le et al. and Ramallo-González et al., respectively.Source: A. NOVEL CORONAVIRUS (COVID-19) · p. 7
  • ElRashidy et al. reported 97.95% accuracy and 98.85% specificity, while Karmore et al. reported 97.95% average accuracy.ElRashidy et al. used wearable sensors, a smartphone app, fog networking, and CNN transfer learning; Karmore et al. used a humanoid robot and pretrained architectures.Source: A. NOVEL CORONAVIRUS (COVID-19) · p. 8
  • Cacovean et al. [28] reported 73% accuracy, while Kumar et al. reported 87.23% for recovery and 86% for prediction using SVM.Source: A. NOVEL CORONAVIRUS (COVID-19) · p. 8

B. HEART DISEASE

The section reviews IoT-based machine-learning approaches to heart-disease diagnosis, including sensor and benchmark-data systems. Reported outcomes vary by dataset and evaluation setting, and several studies note deployment or performance constraints.

B. HEART DISEASE

  • The modified DLMNN system had comparatively low performance with small data size, while the LogitBoost system used traditional feature-selection, reduction, and classification techniques.Source: B. HEART DISEASE · pp. 8–9
  • 100% accuracy, sensitivity, and specificity were reported on the PASCAL dataset, while real-time participants yielded 96.03% accuracy for healthy heart sounds, 91.91% for extrasystole, and 90.11% for murmur.These results are from Deperlioglu et al..Source: B. HEART DISEASE · p. 9
  • Ali et al. reported 98.5% accuracy using ensemble learning and feature fusion for heart-disease prediction.The system merged sensor and patient-history features and used a semantic web rule language to recommend activities.Source: B. HEART DISEASE · p. 9
  • The MSSO/ANFIS system [31] reported 99.45% accuracy, 99% AUC, and 96.54% precision, but presented no results for real-time data.Source: B. HEART DISEASE · p. 9
  • The MDCNN system reported 98.2% accuracy, while HealthFog achieved 89% accuracy and incurred comparatively high time consumption from training at each fog worker node.Source: B. HEART DISEASE · pp. 9–10
  • Nguyen et al. reported 98.03% accuracy; the J48 classifier in reported 91.48% accuracy, 91.50% precision, 91.50% recall, and 91.50% F1-Score.The review notes no physician-notification system for and no real-time study for.Source: B. HEART DISEASE · p. 10

C. DIABETES

The diabetes studies combine IoT sensing and machine-learning classifiers for monitoring and prediction, with reported accuracy varying across datasets and systems. Several reports also identify practical constraints, including latency, sensor compatibility, limited real-time performance, and absent clinical testing.

C. DIABETES

  • The section reviews diabetes frameworks using IoT and machine learning against a backdrop of projected growth in diabetes prevalence and the need for early diagnosis.It gives estimates of 463 million individuals with diabetes in 2019, rising to 578 million by 2030 and 700 million by 2045.Source: C. DIABETES · p. 10
  • 96.43% accuracy and a 5.37% error rate were reported for a decision-tree diabetes prediction framework, though its data-collection procedure was not detailed.A separate wearable-sensor system reported 84.29% accuracy on data from 25 diabetes patients and was described as relatively low-performing for practical use.Source: C. DIABETES · p. 10
  • 99.66% accuracy was reported for SMO on diabetes records collected through an IoT framework monitoring activity, glucose, and temperature.The system sends doctors a text notification when it detects an abnormality; the authors also report comparatively high latency.Source: C. DIABETES · p. 10
  • 97.87% accuracy and 0.03 seconds training time were reported for Random Tree, while another IoT framework reported 96.05% accuracy for Random Forest.The first system's glucose sensor and Arduino could not operate simultaneously; the second used five features and 12,612 records.Source: C. DIABETES · p. 10
  • 80.51% accuracy, 76% precision, 65% recall, and 70% F1-Score were reported for SVM on the positive diabetes class; performance was described as relatively low for real-time implementation.The framework used wearable data from different scenarios and a Kaggle diabetes dataset.Source: C. DIABETES · pp. 10–11
  • SVM achieved 90% accuracy and 89% F1-Score on benchmark data, while a decision-tree and neural-network ensemble reported 94.5% accuracy, 79.5% sensitivity, and 83.12% specificity.The first study did not use real-time cases, and the second had no real-life clinical tests.Source: C. DIABETES · p. 11

IV. AMBIENT ASSISTED LIVING

Ambient assisted living combines IoT sensors and actuators to support older adults in care homes and private homes. The reviewed needs studies emphasize natural interaction, practical assistance, monitoring, and social companionship.

IV. AMBIENT ASSISTED LIVING

  • AAL combines sensors and actuators in an IoT environment to support enhanced lifestyle and human-independent care for older adults in senior care homes and private homes.The passage reports that 87% of older adults prefer living at home over senior homes.Source: §IV · p. 11

A. USER NEEDS STUDIES

  • Older-adult requirements span physical assistance, emotional and social support, and reminders, with studies highlighting fetch-and-carry, entertainment, video chat, and natural speech interaction.The ACCOMPANY study involved 11 formal caregivers, seven informal caregivers, and ten older adults; HomeMate defined five beneficial scenarios,.Source: A. USER NEEDS STUDIES · pp. 11–12
  • The summarized requirements include low-learning natural interaction, reminders, infotainment, real-time vital monitoring and emergency detection, and robotic help paired with social companionship.Studies also cite monitoring movement and falls, and report that 50% of older adults requested efficient speech interaction,.Source: A. USER NEEDS STUDIES · p. 12

B. SMART HOMES

The reviewed AAL systems range from smart-home sensing and cloud platforms to social robots and robot-integrated environments. Across these studies, real-world evaluation, user interaction, autonomy, and integration with health sensors remain recurring concerns.

B. SMART HOMES

  • BLE beacons and RSSI were used for senior activity and room tracking, but the reported localization systems achieved only room-level accuracy.One proposed low-cost approach applies a Random Forest classifier to detect activity and early signs of frailty.Source: B. SMART HOMES · p. 12
  • A radar system reported 95% heart-rate accuracy, 91% respiration-rate accuracy, and fall-detection sensitivity of 97% with 90% specificity, but measures only a few vital signs.Other smart-home work monitored environmental conditions, including gases that may reach explosive levels,.Source: B. SMART HOMES · p. 12
  • The IN LIFE cloud platform uses sensor layers for data collection, gateway transmission, and cloud storage and analysis, while a fog-based AAL system is reported to reduce response delay and energy use.The section notes that proposed AAL solutions still need elaborate real-world testing and user feedback.Source: B. SMART HOMES · p. 13

C. SOCIAL ROBOTS

  • SocialRobot and HomeMate offer different interaction and assistance functions, but users requested more effective or natural speech, and HomeMate lacked links to health and environmental smart devices.SocialRobot was tested for a week in an elderly care home; HomeMate includes reminders, entertainment, and video chat,,.Source: C. SOCIAL ROBOTS · p. 13
  • The reviewed social-robot studies had not integrated robots with smart-home environmental and health-monitoring sensors; the proposed heterogeneous-agent platform still needed real-world testing and a functional mobile robot.The platform links robots, virtual assistants, phones, smart sensors, and wearables through MQTT and REST API.Source: C. SOCIAL ROBOTS · pp. 13–14
  • RiSH integrated a robot with body and home sensors and recognized 37 activities and falls with accuracy up to 88% and 80%, respectively, in experiments with 12 older adults.Its reported human-trajectory accuracy was 0.2 meter.Source: C. SOCIAL ROBOTS · p. 14
  • A Pepper robot added ground-plane and object detection, autonomous navigation, and fall detection to AAL danger detection, improving detection across residential, clinical, and office scenarios.The system would benefit from integration with sensors that monitor users’ health vitals.Source: C. SOCIAL ROBOTS · p. 15

V. SOFTWARE INTEGRATION ARCHITECTURES

The reviewed IoMT architectures organize devices, local processing, and cloud services to connect heterogeneous sensors, agents, and analytics. Their trade-offs include cloud-transfer delays, varying modularity, and gaps in validation or user interaction.

V. SOFTWARE INTEGRATION ARCHITECTURES

  • The section surveys modular, extensible IoMT architectures while noting that each architecture is tailored to a specific use case.The aim is to support information sharing across software, tasks, and applications.Source: §V · p. 15
  • A COTS-based proposal combines off-the-shelf devices, an interactivity component for data retrieval and preprocessing, and a cloud component for synchronization and analytics, but it was not validated.The authors present it as a faster, lower-cost solution.Source: §V · p. 15
  • A related multi-layer medical platform integrates sensor and environmental data and supports fall detection, whereas its similar mHealth extension adds a caregiver interface but not one for patients.The reviewed platform itself lacks a means for users to interact,.Source: §V · p. 15
  • Cloud-heavy architectures offer storage and computing capacity but incur transfer delays, motivating local-network fog processing for faster responses to dangerous situations.Fog devices are described as local-network devices that can lower latency, increase bandwidth, and reduce energy consumption and data overload compared with cloud computing.Source: §V · p. 15
  • The robot-focused architecture decouples agents across ROS and web-service servers, but requires ROS network integration for each new agent, limiting usefulness for non-robotic agents.Its demonstrated system used different sensors, a robotic platform, and a hologram for interaction.Source: §V · pp. 15–16
  • A three-layer framework separates devices, fog preprocessing, and cloud storage and analytics, with defined communication protocols intended to support flexible, personalized use.Its two prototypes still require real-world user testing and feedback.Source: §V · p. 16
  • One three-layer service architecture handles device coordination and time-sensitive reactions before presenting data through web interfaces, but it was neither prototyped nor tested in a real-world environment.It uses BLE for device communication and JSON and RESTful API for the web-service layer.Source: §V · p. 16

DIRECTIONS

The review synthesizes wearable and smartphone monitoring, machine-learning diagnosis, ambient assisted living, social robots, and software architectures. It reports implementation gaps alongside approaches and challenges across these areas.

DIRECTIONS

  • The review covers wearable and smartphone monitoring, machine-learning diagnosis, IoMT-supported assisted living, social robots, and software integration architectures.It compares approaches, advantages, and challenges and identifies future research directions.Source: A. OPEN DISCUSSIONS · p. 16, DIRECTIONS · p. 16

A. OPEN DISCUSSIONS

  • Wearable systems commonly measure heartbeat and body temperature, while most smartphone systems monitor a single sign; most lack adequate clinical-use readiness, and security concerns are largely unaddressed.NodeMCU and WiFi are common in wearable systems; smartphone studies used 5–205 participants and nearly all sampled at 30 Hz.Source: A. OPEN DISCUSSIONS · pp. 16–17
  • COVID-19 accuracy ranged from 99.06% to 66.67%; heart-disease accuracy ranged from 99.45% to 89%, and diabetes from 99.66% to 80.51%.The review also reports that most systems were not practically used in target environments and that clinical trials were absent.Source: A. OPEN DISCUSSIONS · p. 17
  • Ambient-assisted-living studies use home sensors to track movement and activities, while social robots add companionship and daily-living assistance; many systems remain standalone and lack real-world validation.Integrated approaches connect commercial robots to IoT environments, commonly using ROS with REST or MQTT.Source: A. OPEN DISCUSSIONS · p. 17
  • Smart-healthcare architectures are modular and multilayered, combining device or perception layers with local or cloud processing; some add fog for faster responses to dangerous situations.The review identifies protocol standardization, commercial off-the-shelf sensor compatibility, security, and privacy as design concerns.Source: A. OPEN DISCUSSIONS · pp. 17–18

B. CHALLENGES AND FURTHER RESEARCH

The review identifies technical, clinical, and user-centered challenges across smart healthcare. It recommends research on integration, privacy, energy use, data scarcity, and practical testing.

B. CHALLENGES AND FURTHER RESEARCH

  • Heterogeneous wearable sensors produce signals that must be converted into meaningful formats; data fusion and hybrid body-sensor networks are proposed for future investigation.The proposed motivation includes streamlined signals, improved reliability, and reduced cloud communication bandwidth.Source: DIRECTIONS · p. 18
  • Wearable monitoring must address privacy and security, long-term energy use, and sensor failure; proposed directions include protective protocols, low-power equipment, long-life batteries, energy harvesting, and sensor sleep/wake cycles.Blockchain is cited as a possible security approach.Source: DIRECTIONS · p. 18
  • Smartphone monitoring faces noise in camera images and videos; future prototypes should handle noisy environments or use noise-free approaches while remaining low-cost, easy to use, and compatible.The review also calls for assistive devices designed around user needs.Source: DIRECTIONS · p. 18
  • Remote monitoring must handle missing or inconsistent data from power loss, disasters, or device failure, and route requests when care teams cannot respond promptly.The review points to methods for dealing with missing data.Source: DIRECTIONS · p. 18
  • Disease-diagnosis research faces limited data, unnecessary features, and prototypes that often lack clinical tests or practical validation; proposed directions include optimized learning, synthetic data, feature selection, and lightweight models.The review also recommends long-lived sensors for diabetes monitoring and detection.Source: DIRECTIONS · pp. 18–19
  • AAL systems need feedback from older adults and redesign through real-life studies, with attention to independent living, personalization, and intuitive interfaces.The review recommends user-based testing and improvement with older adults.Source: DIRECTIONS · p. 19
  • Smart-healthcare architectures need standardized protocols, interfaces for varied users and applications, fog-cloud integration, and security and privacy designed in early.The review identifies Blockchain technologies as one possible direction for protecting access to data.Source: DIRECTIONS · pp. 19–20

VII. CONCLUSION

The review surveys smart-healthcare monitoring, diagnosis, assisted-living frameworks, and software integration, identifying shortcomings and future research directions. It concludes that technology cannot replace the medical system but may reduce experts’ burden through collaboration between researchers and medical experts.

VII. CONCLUSION

  • The review covers wearable and smartphone monitoring, machine-learning diagnosis of COVID-19, heart disease, and diabetes, assisted-living frameworks, and software integration.It also reviews system advantages, shortcomings, challenges, and potential future directions.Source: §VII · p. 20
  • The authors state that technology cannot replace the whole medical system, but novel architectures may reduce the burden on medical experts.They describe development of assistive systems as feasible when medical experts and researchers work jointly.Source: §VII · p. 20
  • The review characterizes smart healthcare as a means of providing quality services at a fraction of current hospital or assisted-living-center costs.Source: §VII · p. 20
Loading 2107.03924v1…