Chronological Diagnostic Algorithm for Parkinsonism

Chronognosis is a machine learning-based diagnostic tool for predicting neuropathology in patients with parkinsonism using chronological clinical presentations.

Key Features

• Neuropathology-Confirmed: Trained on 949 patients from Mayo Clinic Brain Bank with confirmed pathology
• High Accuracy: Achieves 0.83 AUROC for predicting 9 diagnostic categories at 3 years post-onset
• Chronological Data: Incorporates symptom onset timing for enhanced diagnostic accuracy
• Interpretable Results: Uses SHAP (Shapley Additive Explanations) to show feature contributions
• Robust to Missing Data: Requires only 56 parameters for reliable prediction (AUROC > 0.80)

Installation

Option 1: Clone Repository

git clone https://github.com/onnonuro/chronognosis.git
cd chronognosis
pip install -e .

Option 2: Google Colab (Recommended for Clinicians)

1. Download tutorial.ipynb from this repository
2. Upload to your Google Drive
3. Open with Google Colab
4. Follow the tutorial instructions

Usage

import chronognosis

# Specify paths
root = '.'  # Path to chronognosis directory
input_path = './data/PSP.xlsx'  # Path to patient data
yr_th = 3  # Years threshold for prediction

# Run prediction
chronognosis.main(root, input_path, yr_th)

Input Data Format

Patient data should be in Excel format with:

• index column: Clinical parameter names
• onset column: Years from disease onset (or 100 if absent, 50 if unevaluated)

Required parameters include:

• Age at onset
• Sex (M=1, F=0)
• Family history (Yes=1, No=0)
• 197 clinical symptoms with onset timing

Output

The algorithm provides:

• Probability scores for each diagnostic category
• SHAP force plots showing feature contributions
• HTML reports saved to results directory
• Interactive visualizations for model interpretation

Dependencies

• Python 3.9.15
• pandas 1.5.3
• catboost 1.1.1
• shap 0.46.0

Disclaimer

For Research Use Only: This diagnostic algorithm is intended for research purposes and is not approved for clinical or diagnostic use. Results should not be used as the sole basis for medical decisions. Always consult qualified healthcare professionals for clinical interpretation.

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.