GD-MIL: Grade-Disentangled Multiple Instance Learning for Multimodal Biochemical Recurrence Prediction in Prostate Cancer
作者: Dasari Naga Raju
分类: cs.CV
发布日期: 2026-06-08
💡 一句话要点
提出GD-MIL以解决前列腺癌生化复发预测问题
🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)
关键词: 前列腺癌 生化复发 多实例学习 H&E切片 风险预测 机器学习 临床应用
📋 核心要点
- 现有方法在前列腺癌生化复发预测中主要依赖Gleason评分,未能充分利用H&E切片图像的潜在信息。
- 本文提出GD-MIL,通过门控注意力机制和对抗训练,使得模型能够提取与Gleason评分无关的特征。
- GD-MIL在C-index上达到0.704,显著提高了预测性能,相较于临床基线提升了0.029,表明其有效性。
📝 摘要(中文)
生化复发(BCR)是前列腺癌根治性前列腺切除术后的关键终点,然而现有风险分层几乎完全依赖于Gleason评分。H&E全切片图像是否包含超越评分的预后信号,以及多实例学习(MIL)是否能够恢复这些信号仍未确定。本文在TCGA-PRAD数据集上构建了严格的基准,提出了Grade-Disentangled MIL(GD-MIL),通过引入门控注意力机制和梯度反转对抗网络,使得切片表示对Gleason评分不变。GD-MIL在C-index上达到0.704,显著优于临床基线和最佳成像模型,表明H&E形态学包含与评分互补的预后信息。
🔬 方法详解
问题定义:本文旨在解决前列腺癌生化复发预测中,现有方法过于依赖Gleason评分的问题,未能充分挖掘H&E切片图像的潜在预后信号。
核心思路:提出GD-MIL,通过门控注意力机制和梯度反转对抗网络,鼓励模型提取与Gleason评分无关的特征,从而增强模型的泛化能力。
技术框架:GD-MIL的整体架构包括特征提取模块(使用ResNet50等模型)、门控注意力机制、梯度反转对抗网络和与临床变量的融合阶段。
关键创新:GD-MIL的核心创新在于引入了梯度反转对抗网络,使得模型在提取特征时对Gleason评分保持不变,从而有效提升了预测性能。
关键设计:在模型设计中,采用了多种特征提取器(如ResNet50)和不同的MIL聚合器(如ABMIL、CLAM等),并通过严格的五折交叉验证确保模型评估的可靠性。损失函数设计上,结合了对抗训练和传统的回归损失。
🖼️ 关键图片
📊 实验亮点
GD-MIL在C-index上达到了0.704,显著优于临床基线(提升0.029,p=0.0005)和最佳成像模型(提升0.062,p=0.039),表明H&E形态学包含与Gleason评分互补的预后信息,且在生化复发无事件生存期的分离上表现出极高的统计显著性(log-rank p < 0.0001)。
🎯 应用场景
该研究的潜在应用领域包括前列腺癌的临床决策支持系统,能够帮助医生更准确地预测患者的生化复发风险,从而制定个性化的治疗方案。未来,GD-MIL方法也可能扩展到其他癌症类型的预后预测中,具有广泛的实际价值。
📄 摘要(原文)
Biochemical recurrence (BCR) after radical prostatectomy is a critical endpoint in prostate cancer, yet risk stratification relies almost entirely on variables dominated by Gleason grade. Whether H&E whole slide images (WSIs) carry prognostic signal beyond grade, and whether multiple instance learning (MIL) can recover it, remains unsettled. A key obstacle is that many pipelines select model checkpoints on the evaluation fold, artificially inflating concordance. We construct a rigorous benchmark on TCGA-PRAD (487 patients, 101 BCR events) using strict out-of-fold scoring over five-fold cross-validation repeated across five seeds. The choice of MIL aggregator (ABMIL, CLAM, TransMIL, PatchGCN) has little effect (C-index 0.61-0.64 with UNI2-h), while the feature extractor is the dominant factor (ResNet50 0.566 versus pathology foundation models up to 0.639). A clinical Cox model on grade, stage, and age reaches 0.687; no imaging-only model significantly outperforms it (p > 0.10). We introduce Grade-Disentangled MIL (GD-MIL), a gated-attention MIL encoder trained with a gradient-reversal grade adversary that encourages the slide representation to be invariant to Gleason grade before late fusion with clinical variables. GD-MIL achieves C-index 0.704, significantly outperforming both the clinical baseline (delta-c = +0.029, p = 0.0005) and the best imaging-only model (delta-c = +0.062, p = 0.039), suggesting H&E morphology contains prognostic information complementary to grade. A median risk split yields log-rank p < 0.0001 separation in BCR-free survival (~20% vs ~70% at five years).