CancerGUIDE: Cancer Guideline Understanding via Internal Disagreement Estimation

This project aims to generate treatment predictions and clinical trajectories for patients with non-small cell lung cancer (NSCLC) using large language models (LLMs) applied to real-world clinical notes.

📁 Project Structure

1. benchmark_generation

• Contains source code for generating consistency bench + synthetic benchmarks

2. error_analysis

• Contains source code for generating analysis regarding zero-shot model failure mode predictions

3. main_analysis

• Contains source code for generating model performance on cross-model consistency benchmark, general evaluation scripts, and ROC curve generation files.

4. plot_generation

• Includes scripts for heatmap, bar chart, ROC curves, and confusion matrix analyses from main paper.

5. bash

• Contains ordered bash scripts for executing code to validate paper performance. Ordered by analyses present in paper.