LETT-NeXt: A Lightweight RECIST-Guided Model for 3D CT Lesion Segmentation
Sebastian Aas, Elias Stenhede, Arian Ranjbar · 2026
Read LETT-NeXt: A Lightweight RECIST-Guided Model for 3D CT Lesion Segmentation by Sebastian Aas, Elias Stenhede, Arian Ranjbar (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/arxiv-2606-30108.txt.
Description
Sebastian Aas, Elias Stenhede, and Arian Ranjbar turn a clinician's diameter line into a full three-dimensional tumor mask. RECIST measurements summarize lesions with simple two-dimensional diameters. Those lines cannot capture a lesion's true shape or extent. Their model, LETT-NeXt, was built for a CVPR 2026 segmentation competition. It predicts volumetric masks from CT scans plus RECIST markers.
LETT-NeXt first crops a region centered on the RECIST marker. The RECIST line and its endpoints become two extra prompt channels. These are concatenated with the CT input for the network. A compact encoder-decoder then predicts the lesion mask. Prompt-aware component selection and adaptive zoom refine the output at inference. Validation scores are strong on overlap and surface measures alike. Hidden test scores land somewhat lower but remain competitive. CPU inference takes only seconds per case with modest memory.
Catalog record
| CREATOR | Sebastian Aas, Elias Stenhede, Arian Ranjbar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.30108