论文揭示医学图像分割基准存在范围泄漏：提示词得分多来自标注信息
Counterfactual Analysis of Extent Leakage from Annotation-Derived Prompts in Vision-Language Medical Image Segmentation
如果你在做医学分割或用过这类文本提示模型，这篇论文会让你重新审视基准分数：提示词里的"约 5mm"其实是标注算出来的，删掉它分数就塌了 52% 到 77%。评估协议可以直接借鉴。
一项新研究发现，文本引导的医学图像分割模型在公开基准上的分数提升，主要来自提示词中的"范围信息"泄漏。在结肠镜和胸片数据中，提示词里的大小、数量和位置描述是直接由用于评分的标注掩膜计算出来的，仅保留尺寸词或肺区短语就能恢复 84% 到 98% 的相对空提示的分数提升，删除这些词则损失 52% 到 77%。
在乳腺超声上情况不同：近半数提示词还包含诊断名称，这部分贡献了额外的分数收益。作者在三个数据集上重新训练后发现，模型会从训练提示词中学到这种依赖习惯，并且这种习惯会削弱模型仅凭图像进行分割的能力。研究将这一现象命名为 extent leakage（范围泄漏）。
论文还提出了一种无需训练即可测量泄漏的方法：仅凭提示词文本就能预测病灶面积，在七个测试集中的六个上决定系数达到 0.54 到 0.77。在完全不使用测试掩膜的情况下，从图像估计范围的最好方法在结肠镜上取得 74.59 Dice，与已发表的 77.23 相近。作者据此提出一套评估协议，用于区分提示词描述的图像内容与其从标注中复制的信息。
Counterfactual Analysis of Extent Leakage from Annotation-Derived Prompts in Vision-Language Medical Image Segmentation
Text-guided segmentation models receive an image and a short caption about the lesion, and on public benchmarks the caption improves accuracy considerably. On colonoscopy and breast ultrasound, however, the size, count and location words of each caption were computed from the mask that later scores the prediction, and the chest X-ray captions were written by benchmark annotators. We tested what two published models take from them. On colonoscopy and chest X-ray, it is mostly the lesion's extent. Keeping only the size word or the lung-zone phrase recovers 84 to 98 percent of the improvement over an empty caption, and removing it loses 52 to 77 percent. The rest matters little, and a polyp called a rock is segmented just as well. We call this extent leakage. On breast ultrasound, nearly half the captions also name the diagnosis, which carries part of the benefit. Retraining on three datasets shows that models pick up this habit from their training captions, though only partly on breast ultrasound, and that it weakens their segmentation from the image alone. The leak can be measured without training a model, since caption words alone predict lesion area (coefficient of determination 0.54 to 0.77 on six of seven test sets). Without the test mask, the best method we tested, which estimates extent from the image, reaches 74.59 Dice on colonoscopy against the published 77.23. We propose an evaluation protocol that keeps what a caption says about the image apart from what it copies from the label.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include 医学图像分割, 视觉语言模型, 基准评测, 数据集泄漏.
Source attribution: curated from arXiv cs.LG (@Niran Nataraj, Ankita Das, C. Krishna Mohan, Kenji Kawashima).