Deployment Strategies for Machine Learning Algorithms in Patient Healthcare Domains: An Imaging Perspective
Despite recent advancements in machine learning and artificial intelligence for applications to patient healthcare, real hospital settings have experienced few of the potential benefits and improvements to clinical medicine. To facilitate clinical adaptation of methods in machine learning, we propose DPSP: a standardized framework for step-by-step deployment that focuses on four key components: Data acquisition; Problem identification; Stakeholder alignment; and Pipeline integration. We leverage recent literature and empirical evidence in radiologic imaging applications to justify our approach, and offer discussion to help other research groups and hospital practices leverage machine learning to improve patient care.
To install and run our code, first clone the PennMedBiobankAIDeployment repository.
git clone https://github.com/michael-s-yao/PennMedBiobankAIDeployment
cd PennMedBiobankAIDeployment
Next, create a virtual environment and install the relevant dependencies.
python -m venv env
source env/bin/activate
pip install -r requirements.txt
Quantitative figures presented in our work can be easily reproduced by running the following commands:
python PMBB.py --use_sans_serif
python distribution_shift.py
Questions and comments are welcome. Suggests can be submitted through Github issues. Contact information is linked below.
When available, relevant citation information will be added in a future commit.
This repository is MIT licensed (see LICENSE).