Skip to content
 
 

Latest commit

 

History

87 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NeuroSketch

NeuroSketch is a clinical-grade neurological screening platform designed to detect early indicators of Parkinson’s Disease through hand-drawn motor analysis. By utilizing validated screening methods—specifically the Archimedes spiral and wave tests—the platform provides real-time results directly in the browser with zero data collection.

Technical Overview

The project is architected as a high-performance web application, leveraging a deep learning ensemble to analyze motor tremors and drawing patterns.

Frontend

The frontend is built with Next.js 14 and TypeScript, focusing on an immersive 3D user experience and technical transparency.

  • UI/UX: Implemented using Tailwind CSS and Shadcn UI for a clean, professional aesthetic.
  • 3D Engine: Features a custom 3D neural mesh with a first-person navigable camera, allowing users to explore the model's internal nodes.
  • Animations: Utilizes Framer Motion for cinematic UI transitions, including proximity-based fact reveals and a state-driven cinematic mode.
  • Privacy: Architected for zero-retention; data is processed and results are delivered without permanent storage.

Backend

The backend serves as the inference engine, built with FastAPI for high-concurrency performance and TensorFlow for model execution.

  • Architecture: Utilizes a MobileNetV2 backbone pre-trained on ImageNet for efficient feature extraction, with custom dense layers optimized for multi-class classification.
  • Ensemble Logic: Implements a weighted voting ensemble (40% Spiral / 60% Wave) to maximize diagnostic sensitivity.
  • Inference: Processes image blobs and [x, y, t] coordinate data to evaluate tremor frequency and drawing accuracy.

Model Performance

The current iteration of the system demonstrates high reliability across both primary screening modalities:

Test Type Accuracy AUC
Spiral CNN 86.7% 0.9511
Wave CNN 91.3% 0.9627

Research and Data

The models were trained on the Parkinson's Drawings dataset via Kaggle. The methodology is informed by clinical research regarding the efficacy of hand-drawn Archimedes spirals as digital biomarkers for neurodegenerative conditions.


About

Project for York's Hackathon

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages