Phase Relation-Aware Semantic Prototype Learning for Incomplete Multi-View Liver Tumor Diagnosis in Consumer-Edge Healthcare
Siyuan Li, Xinde Li, Chuanfei Hu
Southeast University Northeastern University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Reliable liver tumor analysis in consumer-edge healthcare becomes challenging when multi-phase contrast-enhanced computed tomography (CECT) records are assembled through distributed telemedicine networks. Asynchronous uploads, bandwidth limitations, and heterogeneous acquisition or governance policies can make individual phases unavailable at the aggregation node, turning affected patient records into incomplete multi-view samples. Because existing incomplete multi-view clustering methods largely rely on direct feature reconstruction or cross-view alignment, their performance may be impaired by substantial enhancement discrepancies across CECT phases. To address this problem, we develop Phase Relation-Aware Imputation with Semantic Prototypes (PRISM), which completes missing information in phase-specific semantic prototype spaces. Its first stage learns prototype assignments, derives sample reliability from assignment entropy, and estimates bidirectional relations using reliable paired observations. The second stage propagates observed assignments through these relations and synthesizes missing representations from target-phase prototypes. The framework targets diagnosis-oriented unsupervised patient stratification to assist clinical decision-making. Evaluation on two liver CECT datasets uses external M0/M1 references at missing rates of 10%–70%. The reported peak ACC is 85:1%, supporting cohort-level patient stratification from incomplete phase features. The framework is designed as a feature-level aggregation module for distributed healthcare workflows.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
生物医学Radiomics and Machine Learning in Medical Imaging
Advanced Clustering Algorithms Research · Medical Image Segmentation Techniques