Automation Bias in AI-Assisted Medical Decision-Making Under Time Pressure in Computational Pathology

Overview

This repository contains the source code and materials for Automation Bias in AI-Assisted Medical Decision-Making Under Time Pressure in Computational Pathology. We quantified automation bias in the form of negative consultations, where a correct initial assessment was overturned by incorrect AI guidance, and examined the impact of time constraints in a web-based experiment. In this study, trained pathology experts (n=28) estimated tumor cell percentages.

Contents

• ExperimentApplication: The ExperimentApplication/ directory contains the Django project source code for the experiment interface, where participants rated tumor cell percentages. The static/ subdirectory includes study materials, such as image patches from different H&E-stained tissue slides (licensed under Creative Commons) and a table with additional information on these study images.
• DataAnalysis: The DataAnalysis/ directory includes the source code to assess data normality and generate the descriptive statistics reported in the paper. Additionally, it contains an Excel file with all anonymized study data, including results from the parametric tests referenced in the paper. A PDF summarizing participant demographic data is also included.

Running the Online Experiment Interface

1. Clone or download this repository, extract the files into a directory of your choice, and navigate to the ExperimentApplication/ subdirectory.

2. Ensure Python is installed, then create a new virtual environment and install the required packages using:

pip install -r requirements.txt

In Visual Studio Code, you may need to set or change the Python interpreter.

3. To start the experiment interface, run the following commands, and access it via http://127.0.0.1:8000/ or http://127.0.0.1:8000/XAI (for the AI-assisted version).

python manage.py makemigrations
python manage.py migrate
python manage.py runserver

License

The images in the ExperimentApplication/static directory were obtained with necessary rights and modified (e.g., for cell detection visualization) to meet study requirements. We publish these images for public use under the Creative Commons 4.0 BY-NC-SA License.