Artificial Intelligence-Driven Wireless Sensing for Health Management

Merih Deniz Toruner, Victoria Shi, John Sollee, Wen-Chi Hsu, Guangdi Yu, Yu-Wei Dai, Christian Merlo, Karthik Suresh, Zhicheng Jiao, Xuyu Wang, Shiwen Mao, Harrison BaiBioengineering 2025 · OpenReview

TL;DR

This review synthesizes recent advancements in AI-driven wireless sensing for health management, highlighting its potential for non-invasive, real-time monitoring and early disease detection. The authors identify key sensing technologies and their applications across various medical domains, while also pointing out critical limitations and future research directions for broader clinical adoption.

What Problem Does This Paper Address?

The paper addresses the growing need for continuous, proactive health monitoring, especially for chronic diseases, which currently account for a significant portion of healthcare costs. Traditional healthcare systems struggle to provide 24/7 oversight, particularly outside clinical settings. This gap creates an opportunity for wireless sensing technologies, enhanced by Artificial Intelligence (AI), to revolutionize healthcare by enabling personalized care, early intervention, and improved patient outcomes. The authors aim to provide a focused review on AI-integrated wireless sensing specifically for clinical applications, moving beyond general vital sign monitoring.

Methods

The authors conducted a comprehensive literature search across PubMed, IEEE Xplore, Embase, and Scopus, focusing on studies published within the last ten years (2014-2024). Their inclusion criteria were strict: studies had to involve AI for signal processing or data analysis, be designed for diagnosing, prognosing, or monitoring specific medical conditions, and have been tested or validated in real-world clinical scenarios with patients. They specifically excluded general applications and studies requiring active patient input (e.g., smartphone apps). The review then synthesized the findings from 34 selected articles, analyzing their strengths, limitations, and contributions to precision medicine.

Datasets

The paper does not introduce new datasets. Instead, it analyzes studies that utilized various existing datasets for training and validation. Some studies leveraged pre-existing hospital voice recordings, combined datasets of healthy and pathological voice samples, or specific datasets like the Device-Free Human Activity Recognition and Monitoring System (DARMS) dataset.

Key Insights and Ideas

  • Beyond Vital Signs: The field is moving beyond basic vital sign monitoring to more complex applications like disease diagnosis, prognosis, and treatment response assessment.
  • AI as the Enabler: AI is crucial for processing the vast amounts of data generated by wireless sensors, enabling predictive modeling and personalized care.
  • Diverse Sensing Modalities: Radar, WiFi CSI, RFID, and acoustic signals are the leading wireless sensing techniques, each with unique strengths and weaknesses for different medical applications.
  • Clinical Validation is Key: While promising, many studies suffer from a lack of extensive clinical validation, often relying on small sample sizes or simulated environments.
  • Multimodal Monitoring is the Future: Integrating data from multiple sensing modalities (e.g., cardiopulmonary, sleep, movement) holds significant potential for earlier and more accurate detection of complex health conditions.
  • Personalization Enhances Accuracy: AI models that adapt to individual patient baselines can improve accuracy, reduce false alarms, and increase patient trust and compliance.
  • Ethical Considerations are Paramount: Patient privacy, data security, and informed consent are critical challenges that need robust solutions, potentially involving blockchain technology.

Key Results

The review identified 10,691 articles, ultimately selecting 34 that met the stringent inclusion criteria. These studies spanned seven distinct medical areas: cardiopulmonary (8 articles), neurology/psychology (5 articles), sleep medicine (11 articles), fall detection (7 articles), endocrinology (1 article), dermatology (1 article), and nephrology (1 article).

  • Methodologies: 15 studies employed deep learning, 16 used machine learning, and 3 utilized both.
  • Sensing Techniques: Acoustic sensing was the most frequently used (35.3%), followed by CSI (29.4%) and radar (23.5%).
  • Participant Numbers: A significant limitation was the small number of participants in most studies, with 22 studies (65%) having fewer than 50 participants.
  • Performance Metrics: While specific quantitative results varied greatly by application, many studies reported high accuracies (e.g., >90% for some fall detection or sleep stage classification tasks), sensitivities, and specificities, often comparable to or exceeding traditional methods. For instance, one study reported a 97.3% accuracy for CAP diagnosis using a subject-dependent setting.

Limitations and Future Work

Limitations: * Small Sample Sizes: A recurring limitation across many studies, hindering generalizability. * Limited Clinical Validation: Many studies lacked extensive validation against gold-standard assessments. * Environmental Factors: Data recording can be influenced by environmental noise, movement, and furniture, affecting accuracy. * Patient Privacy and Security: Concerns regarding data ownership, privacy, and potential security breaches. * Device Lifespan and Sustainability: Battery life and device durability remain challenges, especially in low-resource settings. * Interoperability: Lack of integration with existing healthcare systems and EHRs.

Future Work: * Multimodal Monitoring: Integrating data from various sensors for a more holistic patient view. * Personalized Algorithms: Developing AI models that adapt to individual patient data for improved accuracy and trust. * Decentralized Clinical Trials (DCTs): Utilizing telemedicine and digital tools for faster, more inclusive validation across diverse environments. * Healthcare System Integration: Developing interoperable systems that seamlessly integrate with EHRs for better decision-making. * Sustainable Technologies: Exploring energy harvesting for longer device lifespans. * Secure Data Sharing: Investigating blockchain-based solutions for enhanced patient data security and control. * Next-Generation Technologies: Exploring quantum sensors and 6G networks for unprecedented resolution and real-time capabilities.

Summary updated

BibTeX
@inproceedings{toruner2025artificial,
  author = {Merih Deniz Toruner and Victoria Shi and John Sollee and Wen-Chi Hsu and Guangdi Yu and Yu-Wei Dai and Christian Merlo and Karthik Suresh and Zhicheng Jiao and Xuyu Wang and Shiwen Mao and Harrison Bai},
  title = {{Artificial Intelligence-Driven Wireless Sensing for Health Management}},
  booktitle = {Bioengineering},
  year = {2025},
  url = {https://openreview.net/forum?id=8aPYOgqX0D},
}
Figure