Abstract
Voxel-wise annotation for medical image segmentation is timeconsuming and requires substantial expert effort. Active learning (AL) reduces annotation costs by iteratively selecting informative samples from an unlabeled pool for annotation and model training. Existing batch-selection strategies typically rely on automated acquisition criteria that balance uncertainty, diversity, and representativeness. However, their effectiveness varies across datasets, annotation budgets, model architectures, training procedures, and AL rounds, making it difficult to design a generalizable acquisition criterion. Moreover, clinical priorities and expert-defined segmentation goals are often not explicitly incorporated, while fully automated strategies provide limited transparency to the annotator. We propose IBS - Interactive Batch Selection, an expertguided AL framework that enables interactive adaptation of the acquisition strategy. IBS constructs a gradient-based similarity matrix between samples and allows experts to weight complementary informativeness criteria, including uncertainty, affinity to positive reference samples, repulsion from negative reference samples, and subgroup stratification. The similarity matrix and expert-weighted informativeness scores are combined in a fixed-size determinantal point process to select a diverse and informative annotation batch. Across three medical image segmentation tasks, IBS achieves competitive or improved labeling efficiency compared with conventional AL baselines while enabling transparent, expert-controllable batch selection.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HAIC26_020.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=5Fw9dGD8wR
BibTex
@InProceedings{FölBer_IBS_MICCAISAT2026,
author = { Föllmer, Bernhard AND Schulze, Kenrick AND Dewey, Marc AND Delingette, Hervé},
title = { { IBS: Interactive Batch Selection for Medical Image 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}
}