Abstract
Interactive segmentation enables clinicians to guide annotation, but existing zero-shot models like nnInteractive fail to consistently reach expert-level performance across diverse medical imaging tasks. Because annotation campaigns produce a growing stream of task-specific labelled data, online adaptation of the segmentation model is a natural complement to zero-shot inference. We demonstrate this using a continual adaptation strategy that does not introduce any new parameters or changes to the inference pipeline. We only tune a small fraction of a baseline model’s parameters (the instance normalisation parameters) on the annotation cache triggered by lightweight episode scheduling. We elevate the performance of a state-of-the-art baseline model on eight Medical Segmentation Decathlon tasks spanning diverse anatomical targets and imaging characteristics to expert-level performance, even for tasks where the baseline model previously failed, with the majority of gains realised after a single training episode. We show that the benefits of tuning also depends on task characteristics, with saturating performance gains in targets with complex geometries (e.g., hepatic vessels), ambiguous boundaries (brain tumour core), or where there is a mismatch between the spatial scale of pretrained features and the scale of the target (hippocampus). These all suggest a need for feature-representation alignment in the most challenging scenarios.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HAIC26_002.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=6ItSrs5DBt
BibTex
@InProceedings{EsmPar_Adapt_MICCAISAT2026,
author = { Esmaeili, Parhom AND Tangwiriyasakul, Chayanin AND Gibson, Eli AND Ourselin, Sébastien AND Cardoso, M. Jorge},
title = { { Adapt While You Annotate: The Missing Complement to Interactive Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17279},
month = {pending},
page = {pending}
}