SLPT: Selective Labeling Meets Prompt Tuning on Label-Limited Lesion Segmentation

TL;DR

This paper introduces SLPT, a novel framework combining selective labeling and prompt tuning for lesion segmentation with limited labeled data, achieving high performance with minimal annotation effort.

Contribution

The paper proposes a new method that integrates selective labeling with prompt tuning, including a feature-aware prompt updater and TESLA strategy, to improve lesion segmentation with limited labels.

Findings

Achieves state-of-the-art performance on liver tumor segmentation

Outperforms traditional fine-tuning with only 6% of parameters

Reaches 94% of full-data performance by labeling 5% of data

Abstract

Medical image analysis using deep learning is often challenged by limited labeled data and high annotation costs. Fine-tuning the entire network in label-limited scenarios can lead to overfitting and suboptimal performance. Recently, prompt tuning has emerged as a more promising technique that introduces a few additional tunable parameters as prompts to a task-agnostic pre-trained model, and updates only these parameters using supervision from limited labeled data while keeping the pre-trained model unchanged. However, previous work has overlooked the importance of selective labeling in downstream tasks, which aims to select the most valuable downstream samples for annotation to achieve the best performance with minimum annotation cost. To address this, we propose a framework that combines selective labeling with prompt tuning (SLPT) to boost performance in limited labels. Specifically, we introduce a feature-aware prompt updater to guide prompt tuning and a TandEm Selective LAbeling (TESLA) strategy. TESLA includes unsupervised diversity selection and supervised selection using prompt-based uncertainty. In addition, we propose a diversified visual prompt tuning strategy to provide multi-prompt-based discrepant predictions for TESLA. We evaluate our method on liver tumor segmentation and achieve state-of-the-art performance, outperforming traditional fine-tuning with only 6% of tunable parameters, also achieving 94% of full-data performance by labeling only 5% of the data.