arXiv 2606.068672026-06-05cs.CV

Multi-FRuGaL：面向癌症诊断与预后的多模态灵活冗余感知分解门控学习

Multi-FRuGaL: Multimodal Flexible Redundancy-aware Decomposed Gated Learning for Cancer Diagnosis and Prognosis

Division of Computational Pathology, Department of Pathology and Laboratory Medicine, Indiana University School of Medicine（计算病理学部，病理学与实验室医学部，印第安纳大学医学院）；IU Melvin and Bren Simon Comprehensive Cancer Center（印第安纳大学Melvin和Bren Simon综合癌症中心）；Departments of Biostatistics and Health Data Science（生物统计学与健康数据科学部）；Radiology and Imaging Sciences（放射学与影像科学部）；Neurological Surgery（神经外科）；Indiana University School of Medicine（印第安纳大学医学院）；Department of Computer Science, Luddy School of Informatics, Computing, and Engineering（计算机科学部，Luddy信息、计算与工程学院）

Sanket Kachole, Siddhesh Thakur, Shubham Innani, Sanyukta Adap, Suhang You, Carla Pitarch-Abaigar, Spyridon Bakas

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

提出Multi-FRuGaL框架，通过分解感知自适应门控中间融合，在缺失模态下学习模态级表示，分离冗余与互补信号，提升癌症诊断与预后性能。

摘要：

现代医学依赖于涵盖放射学、病理学、文本报告和结构化临床信息的异构数据源。然而，真实世界的患者数据常常不完整，存在缺失或稀疏获取的模态，限制了标准多模态融合方法的有效性。为此，我们提出了多模态灵活冗余感知分解门控学习（Multi-FRuGaL）框架，这是一种分解感知的自适应门控中间融合框架，可在数据缺失下执行模态级表示学习。Multi-FRuGaL 集成了每个模态的编码器、信号分解层、输入条件门控网络和信息感知融合目标，以将冗余信号与模态特异性互补信号分离，选择性地提升信息丰富的模态并抑制冗余或噪声输入，即使在多个模态缺失时也能保持良好定义。我们在两个多模态头颈癌队列上评估了 Multi-FRuGaL：HANCOCK 挑战数据集（N = 763），包含五种模态和两个预后终点（5年生存率和2年复发率）；以及 HECKTOR 挑战数据集（N = 588），包含三种模态用于人乳头瘤病毒（HPV）状态分类。Multi-FRuGaL 在多个任务上始终比评估的基线方法获得更高的平均性能，将生存预测的 AUC 从 0.601 提高到 0.8496，复发预测的 AUC 从 0.672 提高到 0.8102，并在 HECKTOR 上实现 HPV 预测的 AUC 为 0.975。对于生存分析，它在 HANCOCK 上进一步实现了总生存期的 C-index 为 0.6814，无复发生存期为 0.7421，无进展生存期为 0.7143，在 HECKTOR 上无复发生存期为 0.7203。定性分析进一步表明，即使在严重缺失模态条件下，Multi-FRuGaL 也能学习到判别性和鲁棒的多模态表示。

英文摘要：

Modern medicine relies on heterogeneous data sources spanning radiology, pathology, text reports, and structured clinical information. However, real-world patient data are frequently incomplete, with missing or sparsely acquired modalities, limiting the effectiveness of standard multimodal fusion approaches. To this end, we propose the Multimodal Flexible Redundancy-aware decomposed GAted Learning (Multi-FRuGaL) framework, a decomposition-aware, adaptive gated intermediate-fusion framework that performs modality-level representation learning under missing data. Multi-FRuGaL integrates per-modality encoders with a signal decomposition layer, an input-conditioned gating network, and an information-aware fusion objective to separate redundant from modality-specific complementary signals, selectively upweighting informative modalities and suppressing redundant or noisy inputs, and remaining well-defined even when multiple modalities are absent. We evaluate Multi-FRuGaL on two multimodal head and neck cancer cohorts: the HANCOCK challenge dataset (N = 763) comprising five modalities and two prognostic endpoints (5-year survival and 2-year recurrence), and the HECKTOR challenge dataset (N = 588) comprising three modalities for human papillomavirus (HPV) status classification. Multi-FRuGaL consistently achieves higher mean performance than the evaluated baselines across multiple tasks, improving AUC from 0.601 to 0.8496 for survival, from 0.672 to 0.8102 for recurrence, and achieving 0.975 AUC for HPV prediction on HECKTOR. For survival analysis, it further achieves a concordance index of 0.6814 for overall survival, 0.7421 for recurrence-free survival, and 0.7143 for progression-free survival on HANCOCK, and 0.7203 for recurrence-free survival on HECKTOR. Qualitative analyses further show that Multi-FRuGaL learns discriminative and robust multimodal representations, even under severe missing-modality conditions.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。