Codebase for the FastDiag I paper:

Is Less More? Comparing the Performance of Pauci to Multiparameter Prediction of the Neurotrauma Patient Pathway – a Comparative Observational Study (FASTDIAG I)

---

Summary

This is the codebase for the development of 49 prediction models for seven outcomes and seven feature sets across the neurotrauma patient pathway.

Setup

• Use python version 3.9+
• run pip install -r requirements.txt

Run the model computations to get all the scores

python3 compute_models.py

Structure of the project

• compute_models.py is the main file which processes the data, runs hyperparameter search and returns metrics with confidence intervals for each model.
• SHAP_plots/ contains SHAPley values explanation plots for four test inference example (TP, TN, FP, FN) per model.
• results_summary/ contains the output csv files with performance metrics and 95% confidence intervals.
• data/ contains the unprocessed anonymized database with all clinical variables and CT scan segmentation volumes for each patient as well as the outcomes.
• old drafts/ contains old exploratory studies in notebook format.
• utility/ contains tool scripts.