2024 / Parkinson's Mobility Platform
Parkinson's Mobility Platform
Using machine learning and wearable technology to help Parkinson's patients move more confidently
DNN + BNN ModelsIEEE PublicationiOS + Android Apps
My grandfather played a huge role in raising me.
As his Parkinson's disease progressed, I watched firsthand how something as simple as walking became increasingly difficult. That experience sparked a question that stayed with me for years:
Could technology help detect or prevent these mobility challenges before they occur?
I began exploring the problem in high school through machine learning research focused on Freezing of Gait (FOG), a debilitating Parkinson's symptom that can suddenly prevent patients from moving and often leads to dangerous falls.
Attached research paper
Machine Learning for Neural Decoding Using EEG Signals to Detect Freezing of Gait in Parkinson's Patients
The research side of the project studied whether EEG signals could be used to detect Freezing of Gait episodes with personalized machine learning models.
IEEE Proceedings / 5-page PDFNeural decoding for Freezing of Gait detection
DNNBNNEEGFOG
Personalized neural models analyzed patient EEG signals to identify freezing episodes.
Working with a team of researchers, I analyzed EEG brainwave data from Parkinson's patients and built multiple machine learning models, including dense neural networks, Bayesian neural networks, polynomial regression models, and autoencoders, to study whether neural signals could detect freezing episodes.
Our work showed that personalized models could identify freezing episodes with high accuracy and was ultimately published through IEEE research proceedings.
Most people would have stopped there.
When I arrived at Duke, I became interested in what an actual patient-facing solution could look like.
Prototype app interface
A barebones mobility assistant built around the algorithm.
This mockup shows the patient-facing logic: sensor data becomes step estimates, the model watches for gait interruption, and the wearable provides a simple haptic cue.
Mobility AssistLive
FOG riskLowStep length stable across the last 8 seconds.
Algorithmstep_length = f(accel, gyro, cadence)fog_score = DNN(EEG) + gait_anomaly
01Read sensorsEEG features, gyroscope motion, and accelerometer force stream into the app.
02Estimate gaitStep length and cadence are calculated from filtered motion windows.
03Detect riskModel output is combined with gait abnormalities to flag possible FOG events.
04Cue feedbackThe wearable sends a haptic rhythm to help the user re-initiate movement.
Building on the ideas from the research, I engineered a wearable haptic feedback device designed to improve gait through real-time step length monitoring. I developed novel algorithms that used gyroscope and accelerometer data to calculate step length and identify gait abnormalities without requiring expensive clinical equipment.
To make the system accessible outside of research environments, I also contributed to the development of companion iOS and Android applications, creating a portable platform that could eventually function as a standalone mobility assistance tool.
What started as a personal desire to help my grandfather became my introduction to biomedical engineering, machine learning, mobile development, and healthcare innovation.
Looking back, this project taught me an important lesson: the most meaningful technology often begins with a deeply personal problem.
Key takeaways
From research to patient-facing product
- Co-authored an IEEE-published Parkinson's disease research paper
- Built neural decoding models to detect Freezing of Gait from EEG signals
- Tested dense neural networks, Bayesian neural networks, polynomial regression, and autoencoders
- Engineered a wearable haptic feedback device for gait assistance
- Designed step-length estimation algorithms using accelerometer and gyroscope data
- Contributed to iOS and Android applications for real-world deployment
- First exposure to healthcare technology, machine learning, and product development