AIParkinScan

AIParkinScan combines two independent diagnostic signals — handwritten spiral drawings and short voice recordings — to flag early indicators of Parkinson's disease, namely micrographia (abnormally small, cramped handwriting) and vocal tremor. Audio clips are converted into spectrograms and classified with a custom convolutional neural network, while spiral drawing images are augmented and classified with a Random Forest model trained on a Kaggle spiral-drawing dataset. The two model outputs are combined into a single weighted assessment.

The project includes a Flask web front end where users can submit a drawing and a recording and receive a combined risk assessment, with all model development carried out in Jupyter notebooks. It was built during the MITxSureStart FutureMakers Create-a-Thon, with the goal of making early Parkinson's screening more accessible and affordable.

Highlights

• Dual-input pipeline combining audio spectrograms and spiral-drawing images
• Custom CNN for spectrogram-based vocal tremor detection
• Random Forest classifier with image augmentation for micrographia detection
• Weighted ensemble combining both modalities into one diagnosis score
• Flask web interface for user submissions