Original title: LoFi RADIO: A Distilled In-Domain Backbone Applied for Artifact-Severity Grading of Ultra-Low-Field Neonatal Brain MR
Jonathan B. Martin, Yashwant Kurmi, Charlotte R. Sappo · September 2026
Portable ultra-low-field brain scans help doctors image newborn babies in low-resource clinics. However, these machines lack special shielding, take a long time to scan, and capture weak signals. This causes image flaws called artifactsDistortions or errors in MRI images caused by the scanning process, such as streaks or noise.. To fix this, researchers built automated computer tools to grade how severe seven different artifacts are. They found that standard vision models can pair with classifiers, but no single model handles every artifact best. The team tested two fixes: routing images to different large teacher models using a specialized gate, and distilling those teachers into one smaller student model called LoFi RADIO. Both methods improved the overall score. The single student model matched or beat the gated setup without needing multiple large models running at inference.
Imagine you need to dress for a day where the weather might be rainy, snowy, or sunny, but you can only bring one outfit. Packing separate clothes for each scenario is ideal but impractical, while a single all-weather outfit might not be perfect. In this paper, the different weather types are like different image artifacts, and the single outfit is the distilled model that learns to handle all artifacts efficiently by mimicking specialized models.
For newborns in hospitals without advanced equipment, this automated system checks if brain scans are clear enough to use. It could help doctors in remote areas trust low-cost MRI machines and avoid missing problems due to poor image quality.
The abstract does not specify whether this model works on other types of MRI scans beyond ultra-low-field neonatal brain imaging. It also does not indicate if the system can fully replace human experts in quality control, only that it improves automated grading.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
LoFi RADIO: A Distilled In-Domain Backbone Applied for Artifact-Severity Grading of Ultra-Low-Field Neonatal Brain MR — Jonathan B. Martin, Yashwant Kurmi, Charlotte R. Sappo
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.