IoT + Embedded AI + Healthcare

Smart Pressure
Ulcer Prevention
System

A closed-loop sensor mat with edge AI inference and autonomous micro-pneumatic actuation that prevents bedsores before they form. No nurse intervention required.

FPGA Edge AI STM32H7 ESP32-S3 Federated Learning MIT License
Live pressure heatmap simulation -- 8x8 zone grid
Safe
Critical
Real-time monitoring

Zone risk indicators

Each metric is sampled at 10 Hz, windowed over 5 minutes, and fed into the on-device CNN inference engine running on the FPGA.

System design

Architecture layers

Four hardware and software layers work together: sensors capture raw data, the FPGA fuses and infers, cloud aggregates and trains, and the dashboard alerts.

Sensor mat
32x32 capacitive array MLX90640 thermopile MAX30102 PPG x8
SPI / I2C at 10 Hz
Edge AI controller
Lattice CrossLink-NX FPGA STM32H7 + FreeRTOS Pneumatic actuator driver
MQTT over Wi-Fi
Cloud analytics
FastAPI + InfluxDB Federated learning (Flower) CDSCO compliance audit
WebSocket
Nurse dashboard
Live heatmap Risk alerts Actuation log
Hardware and software

Components and rationale

Every component was selected for a specific reason, not availability. Medical-grade sensing, deterministic real-time control, and privacy-first cloud.

Component Role Why this part
32x32 cap. array Pressure mapping 1024-cell resolution gives per-zone granularity impossible with commercial FSR mats. Detects localized pressure spikes before clinical thresholds.
MLX90640 Skin temperature IR Non-contact 32x24 IR array via I2C. Detects skin temperature differential caused by reduced perfusion before visual symptoms appear.
MAX30102 (x8) Tissue oxygenation Reflectance PPG nodes at high-risk anatomical zones. SpO2 drop at contact point is the earliest measurable biological signal of ischemia.
Lattice CrossLink-NX CNN inference engine Low-power FPGA with deterministic sub-millisecond inference latency. Runs quantized 8-bit CNN with no OS jitter. Parallel sensor stream processing.
STM32H7 System controller 480 MHz Cortex-M7 with dual-core, hardware FPU, 1 MB RAM. Runs FreeRTOS with PID control, safety interlocks, and telemetry serialization.
ESP32-S3 Wireless gateway Dual-mode Wi-Fi + BLE 5.2. Handles MQTT and BLE simultaneously without burdening the real-time controller. Proven hospital Wi-Fi compatibility.
Flower (flwr) Federated learning Privacy-preserving model training. Gradient aggregation across hospital beds without sharing raw patient sensor data. HIPAA and DPDP Act aligned.
FastAPI + InfluxDB Cloud backend Async Python backend with native WebSocket support. InfluxDB handles 10 Hz time-series telemetry from hundreds of beds efficiently.
Value proposition

Why this system matters

Compared to manual repositioning protocols and existing commercial smart mattresses.

20-40
Minutes of advance warning
The CNN predicts ischemia onset before tissue damage occurs, giving time for automated or manual intervention.
3s
Autonomous repositioning
Micro-pneumatic bladders redistribute pressure in under 3 seconds without waking the patient or requiring nurse presence.
8x
Lower cost than alternatives
Bill of materials under Rs. 35,000 per mat versus Rs. 2.5 lakh for imported clinical-grade smart mattresses.
0
Raw patient data leaves the bed
Federated learning shares only model gradients, not sensor readings. Compliant with DPDP Act 2023 and pre-aligned with CDSCO medical device regulations.
72h
Offline operation capability
Edge AI runs entirely on-device. Cloud loss does not disable sensing, inference, or actuation. Designed for rural and power-unstable hospital environments.
+
Model improves over time
Each mat contributes to a shared model that becomes more accurate for specific patient profiles: diabetic, post-surgical, elderly, and paediatric.
Stakeholders

Who this is built for

Four distinct user groups interact with different layers of the system.

Primary user
ICU and ward nurses
See the live heatmap dashboard, receive risk alerts, override actuation, and review repositioning logs per shift.
Technical user
Biomedical engineers
Deploy and maintain the hardware, configure thresholds, integrate with existing nurse call systems and hospital HIS.
Research user
Clinical researchers
Access the InfluxDB time-series data for retrospective studies on pressure ulcer incidence, repositioning efficacy, and patient outcomes.
Developer
Embedded and ML engineers
Build on the open firmware, adapt the CNN architecture, contribute new sensor fusion strategies, or port to different FPGA families.
Roadmap

Future improvements

The current system is a research prototype. These are the next engineering priorities.

01
EMG muscle activity sensing
Detect patient self-repositioning via surface EMG electrodes and pause autonomous actuation to avoid counteracting the patient's own movement.
02
EHR integration via HL7 FHIR
Pull patient risk factors (age, BMI, diabetes status, Braden score) from the hospital information system to adjust model thresholds per patient automatically.
03
Custom ASIC for unit cost reduction
Replace the FPGA with a purpose-built ASIC to bring the per-mat cost below Rs. 8,000, enabling district hospital deployment at scale.
04
Computer vision wound monitoring
Add a bedside camera with an edge CV model to track wound progression in patients who already have stage-1 ulcers, providing objective staging data to clinicians.
05
Multi-ward triage dashboard
Aggregate risk scores across an entire floor or ward into a single priority-sorted view so charge nurses can allocate attention to the highest-risk patients first.
06
Larger clinical dataset and validation study
Partner with a teaching hospital for a prospective clinical trial to validate prediction accuracy and obtain data required for CDSCO Class B device approval.