Theoretical Bounds for AI-based Adaptive Breast Cancer Screening with MRI

Overview

This project calculates and visualizes theoretical bounds for an AI-based adaptive breast cancer screening protocol using MRI. It analyzes how different parameters affect:

1. Recall rates: The proportion of patients who need additional full protocol scans
2. Protocol duration: The average time required per patient

The tool helps evaluate the potential benefits and trade-offs of implementing an AI-based adaptive screening approach compared to standard abbreviated or full protocols.

Quick Start

Using Bazel (Recommended)

From the root:

bazelisk run //:main -- --save_dir=<output_directory>

Using Python directly

From the root:

1. Install dependencies:

   pip install -r requirements.txt

2. Run the script:

   python main.py --save_dir=<output_directory>

Configuration

The analysis parameters can be customized by editing the TOML files in the configs directory:

• performance_parameters.toml: Base performance metrics
  • Sensitivity and specificity for radiologists and AI
  • Protocol durations
  • Cancer prevalence
• changing_parameters.toml: Parameter ranges for analysis
• plot_parameters.toml: Visualization settings

Output

The script generates figures showing:

• Recall rates for different protocols (Adaptive, Abbreviated, Full)
• Average protocol times (Adaptive, Abbreviated, Full)
• Best and worst-case scenarios for the adaptive protocol

All figures are saved to the specified output directory.

License

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

Support

For questions or issues, please open an issue here.