Putting the Segment Anything Model to the Test with 3D Knee MRI - A Comparison with State-of-the-Art Performance

Oliver Mills, Nishant Ravikumar, Philip Conaghan, Samuel D. Relton · 2025

Read Putting the Segment Anything Model to the Test with 3D Knee MRI - A Comparison with State-of-the-Art Performance by Oliver Mills, Nishant Ravikumar, Philip Conaghan, Samuel D. Relton (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-13340.txt.

Description

Oliver Mills and co-authors present a 2025 arXiv paper testing a general image segmentation model on 3D knee MRI scans. The target is the meniscus, cartilage whose damage feeds knee osteoarthritis. The team adapted the Segment Anything Model for full automation and trained a standard 3D U-Net as a baseline for comparison.

Tuning only the model's decoder gave a Dice overlap score of 0.81, behind the U-Net at 0.87. Tuning the whole model end to end matched the U-Net at 0.87, near a past challenge winning score. Both model versions trailed the U-Net on shape distance. The authors doubt the model's fit for fine 3D anatomy.

Catalog record

CREATOROliver Mills, Nishant Ravikumar, Philip Conaghan, Samuel D. Relton
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-13340

Browse the full library · About the Tower of Babel