多模态医学影像在肝细胞癌精准诊断与预后预测中的融合策略进展

Recent Advances in Multimodal Medical Imaging Fusion Strategies for Precision Diagnosis and Prognostic Prediction of Hepatocellular Carcinoma

摘要: 肝细胞癌(HCC)是我国发病率与致死率均居高的重大恶性肿瘤，传统单模态影像在早期诊断与复发预测中受限于信息维度与主观差异。近年来，融合CT、MRI、CEUS及PET等多模态影像的智能分析策略，结合影像组学与深度学习技术，显著提升了病灶检出率、病理分型及术前风险分层的准确性。多模态模型通过在数据、特征与决策层实现信息互补，已广泛应用于微血管侵犯预测、免疫治疗反应评估及个体化预后管理。尽管在多中心异质性、隐私保护与可解释性等方面仍存挑战，基于联邦学习和自监督预训练的多模态框架，正为构建可落地的HCC智能决策支持体系奠定基础。未来，规范化数据采集、临床验证及与指南流程的深度融合，将是多模态影像实现临床转化的关键路径。

Abstract: Hepatocellular carcinoma (HCC) is a major malignant tumor with high incidence and mortality rates in China. Traditional single-modality imaging is limited by information dimensions and subjective variations in early diagnosis and recurrence prediction. In recent years, intelligent analysis strategies integrating multi-modality imaging such as CT, MRI, CEUS, and PET, combined with radiomics and deep learning technologies, have significantly improved lesion detection rates, pathological classification accuracy, and preoperative risk stratification. Multimodal models, achieving complementary information across data, feature, and decision layers, have been widely applied in predicting microvascular invasion, evaluating immunotherapy response, and personalized prognosis management. Despite ongoing challenges in multicenter heterogeneity, privacy protection, and interpretability, multimodal frameworks based on federated learning and self-supervised pretraining are laying the foundation for establishing a practical intelligent decision support system for HCC. Moving forward, standardized data collection, clinical validation, and deep integration with guideline workflows will be critical pathways for translating multimodal imaging into clinical practice.