Use case · Idea to test
MRI reconstruction with fewer model steps
Study when further reconstruction stops helping.
- Acquired scan measurements
- Reconstruction checkpoints
- Image + quality checks
An iterative reconstruction model may refine an MRI image through several stages. A research experiment could ask whether some inputs need fewer model stages while retaining important detail.
Where the small model fits
A quality head checks intermediate reconstructions, alongside consistency with the acquired measurements. It decides whether to continue computing. This is a different output task from an ECG label or a preference classifier.
What would need to work
Compare with a full reconstruction model at matched image quality, including the cost of the stopping checks. Evaluate errors and clinically relevant detail with expert review. Fewer neural-network steps do not by themselves demonstrate a shorter MRI acquisition or safe scanner protocol.
Research proposal only. SkinDeep has no clinical model, patient-use demo or validated medical performance for this task.
A dataset to start with
The NYU fastMRI project provides data for studying MRI reconstruction. Dataset access and use follow the project's terms. NYU fastMRI research project.