Subgroup Performance of a Commercial Digital Breast Tomosynthesis Model for Breast Cancer Detection

Code used for the evaluation of the Lunit INSIGHT DBT model on the Emory Breast Imaging Dataset (EMBED).

Preprint available at https://arxiv.org/abs/2503.13581

Brown-Mulry, Beatrice, Rohan Satya Isaac, Sang Hyup Lee, Ambika Seth, KyungJee Min, Theo Dapamede, Frank Li et al. "Subgroup Performance of a Commercial Digital Breast Tomosynthesis Model for Breast Cancer Detection." arXiv preprint arXiv:2503.13581 (2025).

Installation

Analysis was conducted on Python 3.12 running on JupyterLab v4.0.11 on Ubuntu 22.04.5 LTS

Python packages used for the analysis are listed in requirements.txt and can be installed with pip install -r requirements.txt

Create a .env file inside the notebooks directory with the following environment variables:

• EXAM_LEVEL_PATH: Path to the exam-level data file.
• FINDING_LEVEL_PATH: Path to the finding-level data file.

Usage

Data Engineering

notebooks/screening_label_assignment.ipynb

This file contains the label assignment pipeline to assign ground truth labels to the raw EMBED dataset.

Analysis

notebooks/model_performance_analyses.ipynb

This file contains the model evaluation code which runs on the data produced by screening_label_assignment.ipynb.

notebooks/*.py

These files contain some support functions and constants used by model_performance_analyses.ipynb.