What Matters is the Prompt: Prompt Sensitivity and Prompt Generation in Foundation Models for Lung Nodule Segmentation
Authors
Abstract
Lung nodule segmentation in computed tomography is essential for extracting clinically relevant information for lung cancer assessment and treatment planning. Foundation models have shown notable segmentation capabilities, but state-of-the-art approaches often depend on input prompts, such as points or boxes, making their performance sensitive to prompt quality and placement. Understanding the limitations and constraints of prompt-based foundation models is therefore essential for designing reliable medical image segmentation solutions. In this work, we investigate how prompt quality affects foundation models performance for lung nodule segmentation. We further propose a synthetic prompt-generation model to test if the dependence on manually provided prompts can be mitigated by generating synthetic prompts that can also improve segmentation performance. Perturbation experiments show that bounding box prompts generally outperform point prompts, while latest specialized medical imaging models achieve better performance than general purpose ones. The proposed approach obtains a Dice coefficient of 0.85, suggesting that synthetic prompt generation as a promising strategy for lung nodule segmentation with foundation models.