Abstract
Dense annotations, such as segmentation masks, are expensive and time-consuming to obtain, especially for 3D medical images where expert voxel-wise labeling is required. Weakly supervised approaches aim to address this limitation, but often rely on attribution-based methods that struggle to accurately capture small structures such as lung nodules. In this paper, we investigate a pseudo-label generation approach for lung nodule segmentation by combining a pretrained 3D rectified flow model with a weakly supervised predictor in a plug-and-play manner. Our approach requires only fine-tuning of the predictor model using image-level labels and no retraining of the generative model. The method achieves improved pseudo-label quality compared to attribution-based methods across two predictor architectures, segmenting lung nodules of varying sizes and shapes. When used for subsequent fully supervised training, the generated pseudo-labels improve segmentation performance compared to the evaluated weakly supervised baseline and narrow the gap to models trained on dense manual annotations, highlighting the potential of pretrained generative foundation models as tools for weakly supervised 3D medical image segmentation. The code is publicly available at https://github.com/Gholle-e/PlugAndPlayWSS.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DGM4MICCAI_011.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/profile?id=~Richard_Petersen1
BibTex
@InProceedings{PetRic_Weakly_MICCAISAT2026,
author = { Petersen, Richard AND Kahl, Fredrik AND Alvén, Jennifer},
title = { { Weakly Supervised Lung Nodule Segmentation via Plug-and-Play Guidance of 3D Rectified Flow } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}