Search papers, labs, and topics across Lattice.
This paper introduces SUGFW+, an innovative framework that enhances Cold Start Active Learning (CSAL) for medical image segmentation by leveraging the Segment Anything Model (SAM) to optimize sample selection with a low annotation budget. By integrating uncertainty-guided feature weighting and a novel Patch-based Global Distinct Representation (PGDR) module, SUGFW+ effectively combines diversity and uncertainty in selecting samples for annotation. Experimental results across four public datasets show that SUGFW+ outperforms existing CSAL methods, demonstrating its potential for improving model adaptation in resource-constrained settings.
SUGFW+ achieves state-of-the-art performance in medical image segmentation by seamlessly integrating sample selection and model training through uncertainty-guided feature weighting.
Cold Start Active Learning (CSAL) is important in improving the performance of a medical image segmentation model with low annotation budget by querying a small subset for annotation from an unlabeled training set. Existing CSAL methods typically rely on inefficient dataset-specific Self-Supervised Learning (SSL) to map the unlabeled images into a feature space for sample selection. Recently, the advent of foundation models such as the Segment Anything Model (SAM) offer a promising alternative as the pre-trained model can provide strong generalizable feature embeddings, and allow high performance in downstream tasks after fine-tuning (adaptation). However, how to systematically exploit SAM's inherent embeddings for cold-start sample selection during adaptation with low annotation budget remains underexplored. To address this, we propose an extended SAM-based Uncertainty-guided Feature Weighting (SUGFW+) framework for CSAL and adaptation of SAM. Specifically, it leverages the SAM for Patch-level Feature and Uncertainty Calculation (PFUC), and introduces a Patch-based Global Distinct Representation (PGDR) module that aggregates patch-level embeddings into highly discriminative, uncertainty-aware image-level features. These features are then utilized by a Greedy Selection with Cluster and Uncertainty (GSCU) strategy to combine diversity and uncertainty during sample selection. Unlike prior CSAL methods that decouple sample selection from model training, SUGFW+ tightly integrates these two stages via an Uncertainty-Prompted Fine-Tuning (UPFT) process of SAM in model training. Extensive experiments on four public datasets demonstrate that SUGFW+ achieves state-of-the-art performance against existing CSAL methods. Code is available at https://github.com/HiLab-git/SUGFW-plus.
