Biomedical imaging / Scientific machine learning
Deep Learning for Biomedical Imaging
Research at Harvard University on inferring brain fiber pathways from diffusion MRI.
Problem statement
Infer brain structure when noisy measurements and crossing fibers make image interpretation difficult.
My contribution
Research on Bayesian deep learning, convolutional models, and transfer learning for diffusion MRI at Harvard University.
Technical challenges
Domain-level considerations, without implementation-specific details.
- Inferring structure from noisy and incomplete imaging measurements.
- Representing uncertainty where multiple fiber orientations may be plausible.
- Evaluating learned representations in a scientific setting.
Methodology
Use Bayesian deep learning, convolutional models, and transfer learning for probabilistic fiber inference and voxel classification.
- Convolutional models for voxel classification in regions with crossing fibers.
- Transfer learning for likelihood-free posterior estimation and tract mapping.
Conceptual architecture
MRI measurements
Learned representation
Probabilistic inference
Scientific interpretation
Technologies and methods
- Diffusion MRI
- Bayesian Deep Learning
- CNN