About RouterSense

RouterSense is a collection of software- and hardware-based tools that capture the Internet traffic of digital devices and provide clinicians/participants with health insights. For more information, see our homepage.

Publications

Papers & Posters

1. Digital Phenotyping via Passive Network Traffic Monitoring: Feasibility and Acceptability in University Students
   Under peer-review at Journal of Medical Internet Research (JMIR). 2025. See preprint or read summary.

2. RouterSense: turning existing home routers into scalable, low-cost, long-term, passive health sensors (Poster)
   Alzheimer's Association International Conference (AAIC). 2025. See poster.

3. RouterSense: A Passive, Network-Based Health Monitoring System for In-Home Patients (Poster)
   AAAI Fall Symposium Series 2024 AI for Aging in Place. 2024. See poster abstract.

Talks

1. NYU Urban Research Seminar. December 4, 2025.
   See slides.

2. NYU Urban AI Symposium. Oct 28, 2025.
   See slides.

3. C2SHIP (Center to Stream Healthcare in Place) Weekly Think Tank webinar. Jan 31, 2025.
   See slides and video.

Who are we?

Core team

• Rameen Mahmood (New York University, PhD Student) - Lead PhD Student
• Andrew Quijano (New York University, PhD Student)
• Chenxi Qiu, PhD (Harvard Medical School / Beth Israel Deaconess Medical Center, Neurology)
• Danny Y. Huang, PhD (New York University, Center for Cyber Security) - Principle Investigator

Collaborators

• Jeffrey Kaye, MD and Zachary Beattie, PhD (Oregon Health and Science University, Neurology)
• Chao-Yi Wu, PhD, OT (Harvard Medical School / Mass General Research Institute, Neurology)
• JR Rizzo, MD (NYU Langone, Ophthalmology)
• Catherine Hartley, PhD (New York University, Psychology)
• Andrew Kiselica, PhD (University of Georgia, Neurology)
• Ching Hui Sia, MD (National University Heart Centre Singapore, Cardiology)

Questions? Contact us.

RouterSense: A Passive, Scalable System for Continuous Monitoring of Functional Behavior

Introduction

RouterSense is a software- and hardware‑based system designed to enable continuous, passive, low‑cost monitoring of individuals’ functional behaviors using network traffic. The system addresses a central limitation in current approaches to monitoring Alzheimer’s disease and related dementias (ADRD): clinical and cognitive assessments occur infrequently and provide noisy, momentary snapshots rather than stable, longitudinal behavioral trajectories.

Traditional assessments such as the Montreal Cognitive Assessment (MoCA), Mini-Mental State Examination (MMSE), or biomarker-based tests (e.g., PET scans for amyloid burden) are important for understanding cognitive status. However, they do not capture information about daily-functioning changes—for example, whether an individual can manage online banking, navigate digital interfaces, follow directions, or effectively communicate with family and caregivers. Additionally, these assessments require travel to clinical settings, which can limit access and create disparities in care. Day-to-day variability, mood, and contextual factors also introduce noise. As a result, individuals may receive inconsistent classifications (e.g., fluctuating between “MCI (mild cognitive impairment)” and “non-MCI”) across assessments.

To address these challenges, RouterSense provides a complementary behavioral layer: continuous functional monitoring derived from passively captured internet traffic. This approach enables high-frequency assessment of real-world behavior without requiring proximity to clinics, self-reporting, active engagement, or additional hardware worn on the body.

Limitations of Existing Continuous Monitoring Approaches

A wide range of continuous and near-continuous monitoring strategies already exist, but each presents major limitations.

Ecological Momentary Assessments (EMAs)

EMAs prompt individuals with daily or weekly surveys, often delivered via smartphone apps, interactive voice systems, or automated calls. Although they collect data more frequently than clinic visits, they remain burdensome, disruptive, and reliant on subjective self-report.

Caregiver Reports

Reports from caregivers can provide valuable context, but caregiver fatigue and subjective interpretation introduce variability and inaccuracy. These approaches also shift burden to caregivers rather than reducing it.

Wearables and In‑Home Sensors

Wearables such as smartwatches or fitness trackers provide high-resolution data on mobility, sleep, and physiology. However:

• They require individuals to charge, wear, and maintain the devices.
• They can be expensive at scale.
• Adherence and compliance decay rapidly over time.
• Digital adoption varies substantially across older adults.

Similarly, in-home environmental sensors (motion, sleep, respiration, appliance sensors) require purchased equipment and professional installation, limiting their scalability and equity.

Phone-Based App or Sensor Tracking

Mobile OS-level tracking tools (e.g., accelerometer, GPS, or screen-state monitoring) offer rich behavioral data but suffer from well-known issues:

• High battery consumption due to constant sensor polling.
• Negative user experience (e.g., keyboard-based monitoring used in some cognitive apps created significant discomfort for users in prior deployments).
• Varied support across device models and OS versions.
• Privacy concerns related to invasive sensor access.

Across all these categories, the common challenges include intrusiveness, high cost, poor compliance, and dependence on self-report or active participation.

Key Insight: Internet Traffic as an Unobtrusive Behavioral Signal

Most adults—including older adults and individuals with mild cognitive impairment (MCI)—interact with a wide range of internet-connected devices daily: smartphones, tablets, laptops, smart TVs, Amazon Echo and Google Home devices, security cameras, printers, and other IoT devices. These routine interactions generate encrypted but highly structured network traffic patterns.

Prior research from our group and others demonstrates that such traffic contains latent behavioral signals:

• Daily routines: wake-up and bedtime windows.
• Screen time and digital engagement: frequency, duration, and type of device use.
• Social behavior: communication and speech patterns, visitor presence inferred from device activity.
• Mobility: time in and out of the home, derived from device presence and network connectivity.
• Task-oriented behavior: patterns associated with maps, online banking, email, or multimedia use.

These features reflect functional behavior, which many clinical tests do not capture. RouterSense leverages these passive signals to estimate behavioral trajectories, detect changes between clinical snapshots, and support early detection of concerning shifts.

Deployment Approaches

RouterSense is explicitly designed to require no new infrastructure beyond what participants already have. We support two deployment modalities:

1. Smartphone-Based VPN Client (Software-Only)

Participants install an off-the-shelf VPN application on their smartphone. The VPN automatically and continuously routes encrypted network metadata to RouterSense servers for analysis. The app runs in the background with no required interaction.

This enables:

• Continuous measurement of smartphone-based digital behaviors.
• Zero hardware cost.
• Easy onboarding for large cohorts.

2. Home-Router Monitoring via Raspberry Pi (Hardware-Assisted)

We deploy a low-cost Raspberry Pi device that connects to the participant’s home router. The device requires no configuration by the participant. The network does not need to be reconfigured, either.

This enables:

• Monitoring of all devices in the home, not just the phone.
• Inclusion of IoT device patterns, smart TVs, voice assistants, and smart appliances.
• Household-level activity understanding.

Both methods capture only the encrypted metadata that an internet service provider (ISP) or mobile carrier already sees. RouterSense does not collect content or personally identifiable information.

Privacy Model

RouterSense provides strict privacy guarantees:

• We do not know which participant generated which specific behavior. Data is anonymized at collection.
• We cannot see the content of communications due to end-to-end encryption.
• We cannot see the specific videos, messages, or websites accessed.
• We observe only metadata that ISPs already observe.

In practice, this means we may know that "someone in the cohort watched YouTube at time X," but we do not know who or what video.

Pilot Studies and Early Results

RouterSense has undergone extensive feasibility testing:

• Large-sample student study: Over 200 NYU students deployed RouterSense, generating high-volume, high-variance behavioral data. Participants reported high acceptability; many noted they "forgot" a monitoring system was running because it was entirely passive.
• Older adult clinical study: In collaboration with Harvard Medical School and Oregon Health & Science University (OHSU), RouterSense is being actively deployed among older adults with MCI to evaluate feasibility, acceptability, and behavioral richness in real-world home environments.

Across these deployments, RouterSense has captured:

• Clear circadian rhythms and sleep-wake patterns.
• Screen time fluctuations correlated with academic workload (in the student cohort).
• Rich household interaction signals associated with smart home device use.

These studies demonstrate that RouterSense is technically feasible, unobtrusive, and capable of generating data at population scale.

Capabilities and Clinical Potential

RouterSense is designed to support multiple clinical and research applications:

1. Longitudinal Functional Monitoring

RouterSense captures continuous measures of digital and household functioning, revealing how functional abilities change between clinical assessments.

2. Early Detection of Cognitive or Functional Decline

Changes in digital activity patterns may precede measurable changes in clinic-based cognitive scores. RouterSense aims to detect subtle shifts earlier and more reliably than snapshot testing.

3. Monitoring Treatment Response

Interventions, medications, or cognitive therapies may affect daily function. RouterSense can measure these effects continuously rather than waiting months for clinic-based reassessment.

4. Large-Scale, Low-Cost Behavioral Phenotyping

RouterSense dramatically reduces the cost and complexity of collecting rich behavioral data, enabling:

• Multi-site studies.
• Population-level analyses.
• Equity-focused research in underserved or rural communities.

5. Integration with Clinical Trials

RouterSense provides high-frequency outcome measures that complement existing cognitive or biomarker endpoints.

Conclusion

RouterSense represents a new paradigm for continuous, unobtrusive monitoring of functional behavior. By leveraging routine interactions with internet-connected devices, it addresses the limitations of snapshot clinical tests, self-report measures, and wearable sensors. Its low cost, scalability, and strong privacy protections position it as a transformative tool for studying aging, detecting early signs of cognitive decline, and evaluating treatment effects.

Through ongoing collaborations with NYU, Harvard Medical School, and OHSU, RouterSense is poised to advance our understanding of digital behavior as a window into real-world functioning and long-term cognitive health.