人工智能
计算机科学
模糊逻辑
模式识别(心理学)
特征(语言学)
融合
特征提取
模糊集
传感器融合
模糊控制系统
人工神经网络
数据挖掘
特征向量
图像融合
机器学习
特征选择
作者
Zhengmin Liu,Linlin Hou,Yanghui Gai,Peide Liu
标识
DOI:10.1109/tfuzz.2026.3734464
摘要
Robust multimodal survival prediction serves as a cornerstone of clinical decision-making, with the core objective of effectively integrating multisource data to accurately assess patient survival risk. Existing methods have achieved promising progress, but most of them still rely on deterministic feature construction and fusion paradigms. Such paradigms tend to discretize continuous features within each modality, leading to insufficient modeling of intramodal fuzziness, and also struggle to capture the nonexclusive and context-dependent intermodal fuzziness between genomic and pathological features. To address these challenges, we propose a fuzzy-logic-inspired multimodal survival prediction framework for intramodal fuzzy feature enhancement and intermodal fuzzy-aware fusion. We first design a Mamba-based fuzzy feature reweighting module to learn pathway-level fuzzy weights, transforming pathway contributions from hard binary decisions into continuous degrees of prognostic relevance. Meanwhile, a Mamba-based learnable fuzzy clustering module learns pathology-level fuzzy memberships by allowing each whole slide image (WSI) patch to be associated with multiple pathological prototypes with different membership degrees and thereby preserving continuous transition information in compact WSI representations. Based on these intramodal fuzzy-enhanced representations, a fuzzy-aware bidirectional gated fusion module further models intermodal feature associations and adaptively regulates the contributions of different modalities during fusion for survival prediction. Extensive experiments and comprehensive analyses on four public Cancer Genome Atlas cancer datasets demonstrate the effectiveness, robustness, and interpretability of the proposed framework.
科研通智能强力驱动
Strongly Powered by AbleSci AI