A modular and transparent framework for evaluating medical AI systems
This repository provides a lightweight Python library implementing evaluation criteria inspired by the FUTURE-AI Initiative, a framework for building trustworthy, safe, and clinically reliable AI systems for healthcare.
The objective of this package is to offer a consistent, extensible, and easy-to-use set of metrics that developers, researchers, and clinicians can use to quantify different aspects of model performance and data quality across medical imaging tasks.
Medical AI systems must be assessed beyond simple accuracy numbers.
The FUTURE-AI framework defines six foundational principles for trustworthy AI:
- Fairness – unbiased performance across populations
- Universality – generalization across data sources
- Traceability – transparent and reproducible model behavior
- Usability – practical and interpretable outputs
- Robustness – resilience to variability and imperfections
- Explainability – clear reasoning behind predictions
git clone https://github.com/pabloiislafe/FUTURE-AI-metrics.git
cd FUTURE-AI-metrics
conda env create -f environment.yml
conda activate future-ai-metrics