Unstable Prompts, Unreliable Segmentations: A Challenge for Longitudinal Lesion Analysis

Niels Rocholl, Ewoud Smit, Mathias Prokop, Alessa Hering · 2025

Read Unstable Prompts, Unreliable Segmentations: A Challenge for Longitudinal Lesion Analysis by Niels Rocholl, Ewoud Smit, Mathias Prokop, Alessa Hering (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-2507-19230.txt.

Description

Niels Rocholl, Ewoud Smit, Mathias Prokop, and Alessa Hering show why a top lesion-outlining model falters when scans repeat over time. Cancer care depends on tracking lesions across visits. Models built for single scans may not hold that thread.

The team ran a public model on paired baseline and follow-up chest scans. Two linked failures appeared. First, alignment errors between the scans cut outline quality on the follow-up. Then the matching of lesions across time broke down. In a controlled test, shifting the input volume away from the true lesion center reproduced the fall. The results name prompt placement as a weak point. Tools for tracking must be built and tested for time series, not adapted by hope.

Catalog record

CREATORNiels Rocholl, Ewoud Smit, Mathias Prokop, Alessa Hering
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-19230

Browse the full library · About the Tower of Babel