Anatomical Grounding and Leakage-Aware Multimodal Contrastive Learning for Alzheimer's Disease Classification from Structural MRI

📄 arXiv: 2609.15888v1 📥 PDF

作者: Paul-Gabriel Nicolae, Irina Georgiana Mocanu

分类: cs.CV, cs.AI

发布日期: 2026-09-14

备注: 13 pages, 4 figures

💡 一句话要点

提出解剖基础与泄漏感知的多模态对比学习以解决阿尔茨海默病分类问题

🎯 匹配领域: 支柱二：RL算法与架构 (RL & Architecture) 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 阿尔茨海默病 结构性MRI 多模态对比学习 解剖基础 深度学习 YOLOv8 FastSurfer 认知评分

📋 核心要点

  1. 现有方法在阿尔茨海默病分类中常关注解剖无关区域，导致分类准确性受限。
  2. 论文提出了一种轻量级的切片编码器，并结合解剖分割与多模态对比学习以提高分类性能。
  3. 实验结果显示，使用区域体积的融合准确率为73.0%，而认知评分的融合准确率高达87.3%。

📝 摘要（中文）

本研究探讨了在结构性MRI上进行阿尔茨海默病（AD）分级时，深度网络常常关注解剖上无关的区域的问题。我们采用轻量级的切片编码器（ResNet18与一层Transformer），在1075个ADNI-1基线T1加权扫描上进行实验。通过FastSurfer分割作为解剖参考，训练YOLOv8模型以定位与阿尔茨海默相关的结构。我们还适应了CLIP风格的图像-表格对比框架，并组织ADNIMERGE变量以减少标签泄漏。最终，融合认知评分的准确率达到87.3%。

🔬 方法详解

问题定义：本论文旨在解决在阿尔茨海默病分类中，现有深度学习方法常常关注解剖上无关区域的问题，导致分类准确性不足。

核心思路：提出了一种轻量级的切片编码器，并结合解剖分割与多模态对比学习，旨在提高模型对相关解剖结构的关注，减少标签泄漏的影响。

技术框架：整体架构包括一个ResNet18编码器与一层Transformer，利用FastSurfer分割进行解剖参考，并通过YOLOv8模型定位相关结构。采用CLIP风格的对比学习框架，组织ADNIMERGE变量以减少标签泄漏。

关键创新：最重要的创新在于结合解剖信息与多模态对比学习，显著提高了分类模型对相关结构的关注度，与传统方法相比，减少了对无关区域的依赖。

关键设计：在实验中，采用了特定的损失函数和参数设置，确保模型在不同对比目标下的学习效果，特别是在认知评分与区域体积的对比中，表现出不同的分类准确率。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，融合认知评分的三分类准确率达到87.3%，而仅使用区域体积的准确率为73.0%。此外，限制输入到内侧颞叶的每个受试者裁剪，图像分类准确率从58.7%提升至65.1%。

🎯 应用场景

该研究的潜在应用领域包括医学影像分析、阿尔茨海默病早期筛查和临床决策支持。通过提高分类准确性，能够帮助医生更好地识别和管理阿尔茨海默病患者，具有重要的实际价值和未来影响。

📄 摘要（原文）

Deep networks trained on structural MRI for Alzheimer's disease (AD) staging often reach reasonable accuracy while attending to anatomically irrelevant regions, and multimodal models that add clinical tables frequently rely on variables that were used to assign the diagnostic label in the first place. We study both issues with a deliberately lightweight slice-based encoder (ResNet18 with a one-layer Transformer over slices) on 1,075 baseline T1-weighted scans from ADNI-1. First, we use FastSurfer segmentations as an anatomical reference: YOLOv8 models trained on segmentation-derived labels localize Alzheimer-relevant structures with mAP_50 above 0.96, and a Grad-CAM comparison shows that the image-only classifier frequently attends to the skull, orbits and background. Second, we adapt a CLIP-style image - tabular contrastive framework and organize ADNIMERGE variables along a label-leakage spectrum. Fusion with cognitive scores yields 87.3% three-way accuracy, which we treat as a leakage-driven upper bound rather than an imaging result; fusion with regional volumes yields 73.0%. We observe that the choice of contrastive target changes what the image encoder learns: on MCI vs. CN, the image-only head reaches 52.4% when the encoder is aligned to cognitive scores and 73.8\% when aligned to volumes, although no tabular input is used at inference. Third, restricting the input to a per-subject crop of the medial temporal lobe raises image-only three-way accuracy from 58.7% to 65.1%. All results come from single runs on a small balanced test set, and we report confidence intervals and the protocol differences that prevent direct comparison with published numbers.