Clinical Validation of Generative Models for Medical Image Synthesis

We have developed predictive models that synthesize PET images from multi-parametric MRI, using publicly available datasets. This internship will define and apply a rigorous clinical evaluation framework for these models — including a downstream disease-staging task and a comprehensive set of clinical performance and reliability metrics — benchmark them across three clinical scenarios (MRI-only, MRI+synthetic PET, MRI+real PET), and then optimize the models to improve these metrics.

What you’ll work on:

• State of the art & data. Review existing literature and appropriate the datasets and in-house developed generative models.
• Disease-stage prediction. Build a downstream task predicting Alzheimer’s disease stage from synthetic PET using vision-language models (VLMs).
• Biomarker quantification. Compute amyloid beta, tau and FDG biomarkers (SUVR, centiloid, A/T/N classification) from synthetic PET.
• Clinical evaluation metrics. Assess fidelity, robustness, generalizability, reliability, safety, fairness & bias, explainability (XAI) and uncertainty (evidential deep learning).
• Benchmarking & optimization. Compare the three clinical scenarios against these metrics, then fine-tune the generative models to improve them.

Profile sought:

• Enrolled in a Master’s program in Biomedical Engineering, Computer Science, AI/Data Science, Physics, or a related field
• Strong programming skills in Python, with hands-on experience in deep learning frameworks (PyTorch, MONAI preferred)
• Interest in medical imaging, neuroscience, or clinical AI applications; prior exposure to MRI/PET data is a plus
• Familiarity with explainability, uncertainty quantification, or model evaluation methods is an asset
• Autonomous, rigorous, and comfortable working at the interface of AI research and clinical validation

How to apply: Send your CV and your Master transcripts by email.