SleepSentinel

*🧠 Inspiration Hospital stays often come with uncomfortable sleep — something I’ve personally experienced. But the issue is far more serious than just discomfort:

76% of hospitalized patients report poor sleep quality

Hospitalized adults sleep 1.3–3.2 hours less than the healthy average

Over 40% of ICU patients are at risk of pressure ulcers

$26.8 billion is spent annually in the U.S. on pressure ulcer treatment

When I discovered a dataset tracking sleep through pressure sensors, I knew I wanted to help solve this problem by improving sleep care in hospitals.

💡 What It Does SleepSentinel is an intelligent patient monitoring system designed to:

🛌 Classify patient sleep posture in real time using a CNN model

📊 Track pressure accumulation across body regions over time

🧾 Log posture as FHIR Observations, enabling direct integration with hospital EHRs

🔥 Visualize pressure heatmaps and posture change patterns

📱 Send SMS alerts to nurses for high-risk patients

📝 Record all patient events and nurse actions for full audit trails and compliance

🛠️ How We Built SleepSentinel Component Description Data Input Simulated 64×32 pressure mat data streamed in real time ML Model PyTorch CNN classifying posture (supine, left, right) Backend Flask server offering real-time API endpoints FHIR Integration Posts observations to a local HAPI FHIR server Frontend Interactive dashboard with pressure heatmaps, tooltips, and posture change charts Alert System SMS notifications using Textbelt for critical pressure alerts Audit Logging Secure audit_logs table to record all nurse-patient actions Deployment Fully Dockerized, includes a separate app to view logs

⚔️ Challenges We Faced Building accurate sleep posture classification with limited labeled data

Converting high-resolution sensor frames into actionable pressure zones

Designing a real-time, low-latency, nurse-friendly frontend

Understanding and implementing the FHIR Observation structure correctly

Securing nurse access and managing role-based authentication

🏆 Accomplishments ✅ End-to-end real-time patient monitoring system

🔗 Seamless FHIR integration with a self-hosted HAPI FHIR JPA server

🧠 Real-time posture recognition and ulcer risk visualization

📊 Clean, intuitive nurse-facing dashboard

📝 Robust audit logging for HIPAA compliance

🚨 Implemented automated SMS-based alerting for critical patient care

📚 What We Learned How to work with FHIR healthcare standards and REST APIs

Training and deploying CNNs for pressure-based image classification

Designing secure and trackable audit logs for sensitive environments

Building real-time UX that balances clarity, speed, and usability in healthcare

🚀 What’s Next for SleepSentinel Implement LSTM/Transformer models for better temporal pressure prediction

Add data from wearable devices to expand the sleep monitoring suite

Integrate with real-world pressure mat hardware for deployment pilots

Develop an AI-assisted repositioning assistant for nursing staff

📎 References Machine Learning Model GitHub Repo: [link]

Pressure Map Dataset: [link]

Sleep & Ulcer Statistics:

“76% of hospitalized patients report poor sleep quality”: [link]

“Hospitalized adults sleep 1.3–3.2 h less than recommended”: [link]

“40% of ICU patients at risk for pressure ulcers”: [link]

“$26.8B spent annually on pressure ulcer treatment”: [link]

Built With

• ai
• chatjs
• machine-learning
• python