M2 Internship : Segmentation and Registration of Cerebral Vascular Networks for Whole-Brain Flow Simulation using PINNs

Context

Whole-brain blood flow simulation remains limited due to the high computational cost of classical Computational Fluid Dynamics (CFD) methods. Physics-Informed Neural Networks (PINNs) offer a promising alternative. To train PINNs, two critical types of information must be extracted from medical images:

  • Cerebral vascular geometry: obtained via Time-of-Flight (TOF) MRI segmentation. The extracted networks must accurately reflect the patient-specific vascular structure, be post-processed to reduce noise and partial volume effects, and registered to account for anatomical variability.
  • Velocity profiles: at the inlets and outlets of the segmented vessels, derived from perfusion MRI.

These segmentations and velocity profiles will form the basis for PINN training in a collaborative project with Professor Makoto Ohta's team at Tohoku University, Japan, aiming for patient-specific whole-brain flow simulations. Ultimately, these simulations could inform clinical decision-making in vascular pathologies such as ischemic stroke and aneurysms.

Internship Objectives

  • Apply a nnU-Net deep learning model to segment TOF MRI images.
  • Develop a quality control pipeline for the segmentations.
  • Post-process segmentations using in-house algorithms from a previous PhD student to provide flow simulation-ready geometries.
  • Register multiple segmentations in a common space.
  • Extract inlet and outlet velocity profiles from perfusion MRI for integration into PINN training.

Research Environment

This internship is part of a collaborative project between CREATIS and Tohoku University, Japan. The intern will be integrated into an interdisciplinary team comprising researchers from both institutions, working at the intersection of medical imaging, computational fluid dynamics, and machine learning.

This internship may lead to a PhD project in co-supervision with Tohoku University, Japan, offering an international research experience and exposure to cutting-edge computational and clinical applications.

Candidate Profile

The ideal candidate should demonstrate:

  • Strong expertise in medical image analysis and deep learning, preferably with experience in nnU-Net or similar frameworks.
  • Proficiency in Python and deep learning libraries such as PyTorch or TensorFlow.
  • Interest in computational fluid dynamics, physics-informed machine learning, and biomedical applications.
  • Autonomy, rigor, and motivation for interdisciplinary research bridging computer science and neuroscience.

Internship Information