Dataset
Parkinson's Drawings — sourced from Kaggle. Contains spiral and wave drawings from healthy individuals and patients diagnosed with Parkinson's disease, captured via digitizing tablets with high-resolution coordinate tracking.
healthy & parkinsonModel Architecture
Backbone
MobileNetV2 pre-trained on ImageNet. Feature extraction layers are frozen; only the custom fully-connected classification head is fine-tuned on the Parkinson's drawing dataset.
Classification Head
Global Average Pooling → Dense(128, ReLU) → Dropout(0.3) → Dense(2, Softmax)
Ensemble Strategy
Weighted ensemble: 40% Spiral / 60% Wave — weights derived from validation AUC performance.
Performance Metrics
Spiral CNN
Wave CNN
Tech Stack
Backend
High-performance Python web framework
Deep learning inference engine
Model definition & training API
Frontend
React framework with App Router
Type-safe JavaScript
Utility-first styling
Interactive drawing capture