AI for Stroke Outcome Prediction
Jan 13, 2025·
·
2 min read
Maëlle Moranges
Role: Postdoctoral Researcher at Inria in AIstroSight team
Collaborators : Dr. Metchouff – Physician at Hospices Civils de Lyon (HCL)
Scientific and Technological Objectives
The early inflammatory response after an acute ischemic stroke (AIS) plays a crucial role in patient recovery. This research aims to:
- Identify immune system signatures in patients experiencing loss of autonomy post-thrombectomy.
- Analyze gender-based recovery differences to improve personalized stroke care.
Research Approach
1. Statistical & Graphical Analysis for Clinical Interpretation
- Replicated key statistical from a prior study (originally focused on age) to examine gender differences in stroke recovery.
- Applied Subgroup Discovery on biomarkers to identify patient-specific risk intervals.
- Developed a manual interpretation framework with clear thresholds, similar to standard blood test markers, enabling physicians to make quick prognoses without requiring computational tools.
2. Machine Learning for Predicting Loss of Autonomy
- Expanded the dataset to include medical history, MRI report descriptions and additional features.
- Implemented supervised learning models to predict functional outcomes post-stroke.
- Assessed explainability techniques (SHAP, LIME, counterfactual) to:
- Identify gender impact in stroke recovery predictions.
- Evaluate the consistency of different explanation methods for medical decision-making.
Dissemination and Publications
Public Science Communication
🎥 Infographic from the “AI and Health” project → View on Pop’Sciences
🎤 Excerpts from the roundtable discussion on AI in health → Watch on YouTube
🎓 Digital Sciences Outreach at Lycée Simone Weil → Read more
