HandTrak

Parkinson's Disease early detection, progression, symptom tracking and monitoring.

Comment

What it does

HandTrak is an AI-powered platform that analyzes both spiral drawings and free handwriting tasks to detect micrographia. It outputs:

A binary classification (micrographia present or absent)

A severity score (0–4)

In addition, we integrated a symptom tracker that records prodromal and non-motor features such as:

RBD (REM sleep behavior disorder)

Sleep quality

Fatigue

Mood fluctuations

Results are displayed in a dashboard with:

Longitudinal tracking

Medication timing overlays

Exportable reports for EMR integration

How we built it

Developed and trained a YOLOv11 CNN on handwriting datasets with data augmentation + hyperparameter tuning

Validated performance using AUC, confusion matrices, and sensitivity/specificity analyses, achieving ~90% accuracy

Built a React-based interactive canvas for handwriting/spiral input, capturing velocity, pauses, and letter height variance

Added a symptom diary module to track RBD, sleep, and motor fluctuations

Designed a results dashboard with severity graphs, medication “on/off” overlays, and longitudinal plots

Challenges we ran into

Limited dataset diversity → required augmentation and normalization

Binary vs multiclass performance gap → severity harder to model

EMR interoperability and FHIR standards for export were challenging

Balancing scope vs feasibility in a hackathon timeframe

Accomplishments that we're proud of

Built a functional AI model and interactive tool within hackathon constraints

Expanded scope beyond motor symptoms to include prodromal and non-motor features

Created a clinically interpretable dashboard neurologists could realistically use

Positioned HandTrak as a precision neurology platform, not just a research demo

What we learned

Accuracy isn’t enough — AI in neurology must prioritize interpretability + workflow integration

Prodromal features like RBD and sleep disturbances are key for early detection

The role of UX design in translating AI into usable clinical tools

Longitudinal monitoring > single-time-point screening

What's next for HandTrak

Clinical validation studies in patient cohorts for regulatory translation

Expansion into multimodal biomarkers:

Voice (hypophonia)

Facial expressivity (hypomimia)

Eye movement (saccadic pursuits)

Build a cognitive screening module inspired by MoCA (digitally adapted tasks)

Partnerships with neurology clinics, advocacy groups, payers

Long-term: evolve into a comprehensive neurodegenerative disease monitoring ecosystem

Built With

Share this project:

Updates