🛏️ Overview

At Shirley Ryan AbilityLab, we were asked to address a high-stakes clinical issue: pressure ulcers. These injuries, caused by prolonged bed rest, are painful for patients and costly for hospitals. Despite having turn protocols in place, the hospital lacked a way to objectively verify whether patients were being repositioned as scheduled.

I joined a project that aimed to solve this problem using a novel combination of embedded hardware, fabric-integrated sensors, and machine learning. Our goal was to build a smart, non-intrusive bed sensor system that could classify a patient’s position in real time and log that data for long-term studies.

---

🎯 Objectives

We set out to design a sensor platform that could:

• Detect patient position accurately and non-invasively,
• Blend seamlessly into hospital beds without altering comfort or safety,
• Interface wirelessly with mobile and cloud systems for data collection, and
• Support long-term research and possible integration with hospital EMR systems.

I contributed across every level of the stack—from PCB design to firmware to cloud database integration and ML model deployment.

---

⚙️ What I Built

• Hardware Design:
  • Designed a modular system of 9 sensor pods, each combining an IMU and load cell.
  • Built soft sensor housings by learning to use a sewing machine—embedding electronics into stretchable fabric placed under the bedsheet.
  • Designed and assembled a 2-sided PCB shield with BLE antenna, power conditioning, and I/O routing to a central MicroPython microcontroller.
  • Created a custom wire harness for reliable, hospital-safe deployment.
• Firmware & BLE Communication:
  • Extended firmware to stream all sensor data wirelessly over BLE to an Android app.
  • Implemented calibration routines and data pre-processing for noise reduction and alignment.
• Cloud Infrastructure:
  • Integrated with a mobile + GCP pipeline for continuous data logging.
  • Built MySQL-backed backend for long-term storage and analysis.
  • Ensured compliance with clinical constraints around power usage, safety, and patient comfort.
• Data Collection & Machine Learning:
  • Collected sensor data from 60 diverse human subjects across six discrete bed positions.
  • Trained and evaluated multiple models (NNs, SVMs, Random Forests).
  • Deployed a Random Forest classifier to the cloud for real-time inference with >90% accuracy.
• Clinical Integration & IP:
  • Filed a formal invention disclosure through the hospital’s Office of Technology Transfer.
  • Participated in discussions about EMR integration, research expansion, and pilot deployment.

---

✅ Outcome

We developed a working prototype that classified patient bed position in real time with over 90% accuracy—across multiple positions and subjects.

The system was designed to be low-profile, comfortable, and safe for patients, while providing valuable, actionable data to clinicians. The project opened the door to a new class of “smart bed” technologies in hospital care—and remains one of the most clinically impactful systems I’ve helped build.

It also marked a turning point in my engineering journey—where embedded systems, firmware, and machine learning came together to solve a real-world problem that affects lives every day.

---