Weakly-Supervised Lung Nodule Segmentation via Training-Free Guidance of 3D Rectified Flow

2026-04-09

AI Summary

This work addresses the challenge of accurately segmenting pulmonary nodules in 3D medical images under weak supervision, where only image-level labels are available and dense pixel-wise annotations—typically expensive to acquire—are absent, particularly hindering the delineation of small structures. To this end, the authors propose a novel training-free weakly supervised segmentation method that leverages a pre-trained 3D rectified flow generative model coupled with a lightweight predictor. By employing a guidance mechanism that requires no retraining of the generative model and fine-tuning the predictor solely with image-level labels, the approach achieves high-quality segmentation. The resulting plug-and-play framework demonstrates superior performance over existing baselines on the LUNA16 dataset, exhibiting robust accuracy across nodules of varying sizes and morphologies.

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 propose a weakly-supervised segmentation method for lung nodules by combining pretrained state-of-the-art rectified flow and predictor models in a plug-and-play manner. Our approach uses training-free guidance of a 3D rectified flow model, requiring only fine-tuning of the predictor using image-level labels and no retraining of the generative model. The proposed method produces improved-quality segmentations for two separate predictors, consistently detecting lung nodules of varying size and shapes. Experiments on LUNA16 demonstrate improvements over baseline methods, highlighting the potential of generative foundation models as tools for weakly supervised 3D medical image segmentation.

Problem

• weakly-supervised segmentation
• lung nodule
• 3D medical image
• dense annotation
• small structure detection

Innovation

• weakly-supervised segmentation
• 3D rectified flow
• training-free guidance
• lung nodule
• generative foundation model