Masked autoencoder pretraining for peripheral nerve segmentation in ultrasound images
Matthew Webster, Ko Eun Kim, Jaewon Lim, Joonnyong Lee, Beom Suk Kim · 2026
Read Masked autoencoder pretraining for peripheral nerve segmentation in ultrasound images by Matthew Webster, Ko Eun Kim, Jaewon Lim, Joonnyong Lee, Beom Suk Kim (2026) 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/PMC13424653.txt.
Description
Matthew Webster and four colleagues borrow a self-teaching trick from image AI, masked autoencoder pretraining, to outline peripheral nerves in ultrasound scans, a task starved of labelled examples. Most earlier work covered only the median nerve, so the team assembled 10,603 images spanning eight nerves and more than 1,500 patients with normal nerve tests.
A segmentation model first studied unlabelled scans by hiding and reconstructing patches, then fine-tuned on labelled nerves. Pretraining lifted accuracy for every nerve. Gains were dramatic where labels were scarcest: the sural nerve's overlap score rose from 14% to 40%, and another small nerve's from 28% to 55%. Well-supplied nerves like the median barely moved, confirming the method's value lies in low-data corners.
Catalog record
| CREATOR | Matthew Webster, Ko Eun Kim, Jaewon Lim, Joonnyong Lee, Beom Suk Kim |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42760 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |