Foundation model embeddings capture pre-diagnostic changes on screening mammograms

📄 arXiv: 2609.26605v1 📥 PDF

作者: Kalina P. Slavkova, Eric Brattain, Aditya Gowd, Akash Pattnaik, Jean-Benoit Delbrouck, Matthew Morgan, Julie Bauml, Javid Abderezaei, Khan Siddiqui

分类: cs.CV, cs.LG

发布日期: 2026-09-22

备注: 13 pages, 5 figures, supplementary info attached

💡 一句话要点

提出基础模型嵌入以捕捉筛查乳腺X光片上的前诊断变化

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 基础模型 乳腺癌筛查 医学影像 前诊断变化 嵌入分析 生物活检 深度学习

📋 核心要点

  1. 现有方法在筛查乳腺X光片中难以有效捕捉前诊断的组织变化，通常需要特定任务的适应。
  2. 本研究提出利用基础模型嵌入，分析乳腺X光片的变化，探索其在癌症诊断中的潜在应用。
  3. 实验结果显示，恶性病例的嵌入在筛查前的移动速度显著快于健康对照组，验证了模型的有效性。

📝 摘要（中文）

本研究探讨了基础模型嵌入在筛查乳腺X光片中是否能够编码前诊断组织变化，而无需特定任务的适应。我们分析了1773名女性的生物活检数据，发现恶性病例在筛查前的嵌入速度显著快于匹配的阴性对照组，且这一现象与预训练领域相关。研究结果表明，基础模型嵌入能够有效捕捉前诊断的乳腺影像变化，具有重要的临床应用潜力。

🔬 方法详解

问题定义：本研究旨在解决现有乳腺X光筛查方法无法有效捕捉前诊断组织变化的问题，现有方法通常依赖于特定任务的适应，限制了其通用性。

核心思路：论文提出利用基础模型嵌入来分析乳腺X光片的变化，探索其在癌症早期诊断中的潜力，强调无需特定任务的适应性。

技术框架：研究中使用了四种2D模型（Mammo-CLIP、HOPPR、MedImageInsight和BiomedCLIP），对1773名女性的生物活检数据进行分析，采用相同的处理流程，量化乳腺级别的嵌入变化。

关键创新：研究的主要创新在于发现基础模型嵌入能够有效编码前诊断的乳腺影像变化，且这一能力与临床相关的预训练领域密切相关，区别于一般的生物医学预训练。

关键设计：在实验设计中，采用了混合效应分析方法，比较了恶性病例与健康对照组的嵌入速度，特别关注了不同模型在不同筛查间隔的表现。

🖼️ 关键图片

img_0
img_1
img_2

📊 实验亮点

实验结果表明，在MII嵌入空间中，恶性病例在筛查前的前两个间隔内的嵌入速度显著快于健康对照组，且在第一个间隔内，MC模型的差异在两个生物活检组中均显著，验证了基础模型嵌入的有效性。

🎯 应用场景

该研究的潜在应用领域包括乳腺癌的早期筛查和诊断，能够帮助医生更早地识别潜在的癌症病例，从而提高患者的生存率。未来，基础模型嵌入技术可能在其他医学影像分析领域也展现出类似的应用价值。

📄 摘要（原文）

Foundation model embeddings of screening mammograms may encode pre-diagnostic tissue change without task-specific adaptation. We tested whether embeddings move faster along a data-derived "cancer direction" in women later biopsied for cancer than in matched screen-negative controls, and whether this depends on pretraining domain. We studied 1,773 biopsied women (785 malignant, 988 biopsy-negative) and 1,773 matched controls, each with at least two annual screening exams before their index exam. An identical pipeline was applied to four 2D models: Mammo-CLIP (MC, out-of-distribution mammography), HOPPR (in-distribution mammography), MedImageInsight (MII, general medical imaging), and BiomedCLIP (biomedical vision-language pretraining on literature figures). Breast-level embeddings quantified longitudinal movement along the cancer direction. We compared cases and controls using a between-patient design with complementary mixed-effects analysis, and biopsied versus healthy contralateral breasts within patients. Under matched modality in MII embedding space, malignant cases drifted significantly faster than controls in the first two screening intervals preceding the index exam; biopsy-negative cases showed significance only in the first. MC differences were significant in the first interval for both biopsy groups. Within-patient comparisons showed a broadly similar pattern, with MC significance extending to the second interval in both groups and HOPPR showing significance at interval 1. BiomedCLIP showed no significant differences in either design or biopsy group. Overall, directional embedding velocity emerges as a property of clinically grounded rather than general biomedical pretraining, showing that foundation model embeddings can encode pre-diagnostic mammographic change without task-specific adaptation.