Revisiting model scaling with a U-net benchmark for 3D medical image segmentation

Are larger models always better for 3D medical image segmentation? Despite the widespread adoption of 3D U-Net in various medical imaging tasks, this critical question remains underexplored. To challenge the common assumption, we systematically benchmark 18 U-Net variants-adjusting resolution stages, depth, and width-across 42 diverse public datasets. Our findings reveal that the answer is no: optimal architectures are highly task-specific, with smaller models often performing competitively. Specifically, we identify three key insights: (1) increasing resolution stages provides limited benefits for datasets with larger voxel spacing; (2) deeper networks offer limited advantages for anatomically complex shapes; and (3) wider networks provide minimal advantages for tasks with limited segmentation classes. Based on these insights, we provide practical guidelines for optimizing U-Net architectures according to dataset characteristics. Our findings highlight the limitations of the"bigger is better"paradigm while establishing a framework for balancing performance and computational efficiency in 3D medical image segmentation tasks.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2025-08-14
DOI
https://doi.org/10.1038/s41598-025-15617-1
Citations
5
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
6.80

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Revisiting model scaling with a U-net benchmark for 3D medical image segmentation

Lixu Gu, Zhongying Deng, Yanzhou Su, Junjun He et al.
5 citations
Scientific Reports
Radiomics and Machine Learning in Medical Imaging
6.80
article

Revisiting model scaling with a U-net benchmark for 3D medical image segmentation

Lixu Gu, Zhongying Deng, Yanzhou Su, Junjun He, Yun Gu, Haoyu Wang, Tianbin Li, Jie Liu, Yu Qiao, Ziyan Huang, Shaoting Zhang, Zhikai Yang
article en
5 citations

Abstract

Are larger models always better for 3D medical image segmentation? Despite the widespread adoption of 3D U-Net in various medical imaging tasks, this critical question remains underexplored. To challenge the common assumption, we systematically benchmark 18 U-Net variants-adjusting resolution stages, depth, and width-across 42 diverse public datasets. Our findings reveal that the answer is no: optimal architectures are highly task-specific, with smaller models often performing competitively. Specifically, we identify three key insights: (1) increasing resolution stages provides limited benefits for datasets with larger voxel spacing; (2) deeper networks offer limited advantages for anatomically complex shapes; and (3) wider networks provide minimal advantages for tasks with limited segmentation classes. Based on these insights, we provide practical guidelines for optimizing U-Net architectures according to dataset characteristics. Our findings highlight the limitations of the"bigger is better"paradigm while establishing a framework for balancing performance and computational efficiency in 3D medical image segmentation tasks.

Scientific ReportsVol. 15(1)
Shanghai Jiao Tong University (CN), University of Cambridge (GB), Sichuan University (CN), Beijing Academy of Artificial Intelligence (CN), Artificial Intelligence in Medicine (Canada) (CA), Shanghai Artificial Intelligence Laboratory (CN), Monash University (AU), Shanghai Institute of Technology (CN), KTH Royal Institute of Technology (SE)
Shanghai Jiao Tong University, National Science and Technology Major Project, National Key Research and Development Program of China
Openalex Percentile: Top 3%
Radiomics and Machine Learning in Medical Imaging
6.80
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.