The authors compared INR-based and U-Net-based models for brain MRI segmentation across different parameter budgets and training conditions. They found INR models perform best in low-parameter, limited-augmentation settings, while U-Nets benefit more from larger capacity and standard augmentation. They also found segmentation-related information is spread across multiple INR layers, leading them to propose HierINRSeg, which aggregates these layers and outperforms a strong baseline by 5.6 points in-domain and 8.2 out-of-domain.
Hierarchical implicit models improve compact brain MRI segmentation across sites
By combining features from multiple network layers, HierINRSeg improved both in-domain and cross-site segmentation over a recent implicit neural representation baseline.
Top University
Ziyao Shang · Pouya Sadeghi · Letian Jiang · Alexander Wong · Sirisha Rambhatla
University of Waterloo
Research Digest··2 min read
In one line
INR-based segmentation excels at low parameter budgets and limited augmentation, but U-Nets win with larger capacity.
Before you read this
An idea this paper builds onRead it as
Shang et al. compared compact implicit neural representations, or INRs, with conventional U-Net pipelines for brain MRI segmentation across parameter budgets, augmentation settings and imaging sites.
The authors evaluated INR-based and U-Net-based semantic segmentation under controlled low-parameter regimes.
How it works
before and afterThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.