AI4Health - Framework for Transparent AI
Framework for transparency and fairness of AI algorithms in healthcare
Overview
Framework for transparency and fairness of AI algorithms in healthcare
AI4Health - Algorithmic Transparency
Addressing the problem of opacity and bias in "black box" algorithms, AI4Health aims to define an indicator framework for transparency and fairness of AI models in healthcare.
Working with controlled data, the team develops solutions that prevent discrimination based on gender or ethnicity, ensuring more equitable and reliable care.
Importance of Transparency
Opaque algorithms can perpetuate biases present in training data, leading to unequal diagnoses and treatments. Transparency is fundamental for trust and clinical adoption.
Video
Technical Specs
Indicator Framework
- Explainability Metrics: Model interpretability measures
- Fairness Indicators: Equity across demographic groups
- Bias Detection: Identification of gender/ethnicity bias
- Performance Parity: Equivalent performance across subgroups
Methodology
Testing on controlled datasets with performance analysis disaggregated by gender, ethnicity, and other sensitive variables.
Project Partners
Università di Genova
Legacoop LiguriaTimeline
Timeline TBD
Gallery
No media available.