NAISS
SUPR
NAISS Projects
SUPR
Active learning for super-resolution imaging
Dnr:

NAISS 2025/22-1570

Type:

NAISS Small Compute

Principal Investigator:

Hannes Waldetoft

Affiliation:

Uppsala universitet

Start Date:

2025-11-13

End Date:

2026-06-01

Primary Classification:

30118: Medical Biostatistics

Webpage:

Allocation

Abstract

4D Flow MRI enables full-field quantification of various hemodynamic parameters. While the technique faces limitations related to resolution, deep neural networks have shown promise in enhancing resolution post-scan. So far, however, networks have been predominantly data-driven, overlooking the information carried in individual training sets and instead requiring a substantial amount of training data to converge. To address this, advancements within scientific machine learning propose active learning – a concept in which a learning algorithm interactively queries prediction uncertainty during training and selects unlabeled data of high uncertainty to label for further learning. While assessed in non-medical settings in both practice and theory, the implementation for medical imaging remains unexplored. The aim of this study is to as part of PhD based work and collaborative setup to implement active learning for super-resolution MRI, assessing potential in efficient learning.