Our group is actively collecting MRI data from human volunteers. simultaneously measuring arterial, venous, and cerebrospinal fluid (CSF) flow at multiple locations across the intracranial system using phase-contrast MRI (PC-MRI). The scientific question is compelling. The engineering challenge is equally so: our scanner is a 1.5T MRI system, which introduces image noise, resolution limitations, and systematic phase errors that must be corrected before the physiological signal can be reliably extracted. This is where you come in.
What You Will Do
You will build and validate a complete AI-based post-acquisition image processing pipeline that takes raw 2D cine PC-MRI data from our 1.5T scanner and produces clean, high-fidelity blood and CSF flow waveforms ready for analysis. The pipeline combines three complementary methods:
- Background phase offset correction — automated removal of eddy-current-induced phase errors using an established algorithmic approach (MSAC), which systematically bias flow measurements if left uncorrected.
- AI-driven in-plane super-resolution — fine-tuning a state-of-the-art pre-trained Enhanced Super-Resolution GAN (ESRGAN), called CRISPFlow (Harvard/BWH, 2025), for the intracranial and CSF imaging territories in our protocol. The pre-trained model already exists and has been validated in cardiovascular MRI; your work is to adapt and validate it for our specific application — intracranial arteries, dural venous sinuses, and the cerebral aqueduct — using our own 1.5T data.
- Partial-volume correction — model-based correction of systematic underestimation of flow in small vessels (particularly the ~1–2 mm cerebral aqueduct) caused by boundary voxel signal mixing.
Scanning is already underway, you will have real patient and healthy-subject data available from day one.
What You Will Learn
- Hands-on experience with deep learning model fine-tuning applied to real clinical imaging data
- End-to-end medical image processing pipeline development, from DICOM raw data to validated flow metrics
- Quantitative validation methodology (Bland-Altman, ICC, test-retest) for medical imaging AI
- Working knowledge of MRI physics, phase-contrast flow imaging, and cerebrovascular physiology
- Scientific writing toward a thesis and a co-authored journal manuscript
What We Offer
- A project with immediate clinical relevance — your pipeline will directly enable ongoing brain physiology research
- Access to a real 1.5T MRI dataset collected in our lab
- Daily supervision and integration into an active vascular neurology research group
- Co-authorship on a peer-reviewed journal manuscript upon successful project completion
- A project at the intersection of AI, medical imaging, and neuroscience— a highly visible and employable skill set