Experience & Education

Key Contributions

• Developing 3D diffusion models that generate brain images from structural MRI to produce counterfactuals: how a given brain would look with or without a pathology.
• Separating the factors shared by healthy and patient brains from those salient to the disease (contrastive analysis) to discover and quantify pathological patterns, supporting non-invasive diagnosis.
• Studying sparse autoencoders to make counterfactual generation interpretable, exposing the human-readable features behind each generated change.