首页|期刊导航|磁共振成像|多模态影像组学与深度学习无创预测胶质瘤TERT启动子状态及辅助临床决策的研究进展

多模态影像组学与深度学习无创预测胶质瘤TERT启动子状态及辅助临床决策的研究进展OA

Advances in multimodal radiomics and deep learning for predicting TERT promoter status and assisting clinical decision-making in glioma

中文摘要英文摘要

端粒酶逆转录酶启动子突变是胶质瘤分子分型的核心分子事件,其检测结果对判断患者预后及制订个体化治疗方案具有重要指导价值.传统检测手段依赖有创组织活检,存在肿瘤时空异质性、取样偏差及无法动态监测等固有局限,难以满足精准的诊疗需求.近年来,以多参数MRI为基础的影像组学与深度学习技术快速发展,已成为术前无创预测胶质瘤端粒酶逆转录酶启动子突变状态的重要研究方法.研究显示,部分多模态融合模型在有限的外部验证中展现出一定泛化能力,但跨中心性能衰减仍是普遍挑战.同时,该技术向常规临床实践转化仍面临影像标准化、跨中心数据共享、模型可解释性不足等关键挑战.未来,通过推进大规模前瞻性临床试验、建立标准化跨中心影像数据库、融合空间多组学与液体活检等跨模态信息,并结合联邦学习、可解释人工智能等技术优化模型,有望将影像人工智能预测模型转化为可靠的临床决策支持工具,促进胶质瘤诊疗的全程智能化与个体化.本文系统梳理该领域最新研究进展,重点探讨多模态影像组学及深度学习模型在预测胶质瘤端粒酶逆转录酶启动子突变中的方法学构建、模型性能优化及临床转化价值,并分析当前研究中模型精细化、多模态融合、可解释性探索等核心发展特征,以期为胶质瘤端粒酶逆转录酶启动子突变的术前无创预测提供方法学参考.

Telomerase reverse transcriptase promoter mutation is a core molecular event in the molecular classification of gliomas.Detecting these mutations provides crucial guidance for patient prognosis assessment and the development of individualized treatment plans.Traditional detection methods rely on invasive tissue biopsy,which has inherent limitations such as tumor spatiotemporal heterogeneity,sampling bias,and the inability to perform dynamic monitoring,making it difficult to meet the clinical needs of precise diagnosis and treatment.In recent years,radiomics based on multiparametric MRI and deep learning technology have developed rapidly and have become an important research direction for non-invasive preoperative prediction of TERT promoter mutation status in gliomas.Research shows that some multimodal fusion models demonstrate certain generalization ability in limited external validation,but performance degradation across centers remains a common challenge.Meanwhile,the translation of this technology into routine clinical practice still faces key challenges,including image standardization,cross-center data sharing,and insufficient model interpretability.Looking ahead,by promoting large-scale prospective clinical trials,establishing standardized cross-center imaging databases,integrating spatial multi-omics and liquid biopsy cross-modal information,and optimizing models with technologies such as federated learning and explainable AI,it is hoped that imaging AI prediction models can be transformed into reliable clinical decision-support tools,advancing intelligent and individualized glioma diagnosis and treatment.This review systematically summarizes the latest research progress in this field,focusing on the methodological construction,model performance optimization,and clinical translation value of multi-modal radiomics and deep learning models in predicting TERT promoter mutations in gliomas.It also analyzes core development features in current research,such as model refinement,multi-modal fusion,and interpretability exploration,aiming to provide methodological reference for non-invasive preoperative prediction of TERT promoter mutations in gliomas.

杨舵;李丽鑫;延雨欣;王晓彤;蒋盈盈;白玉萍

兰州大学第二临床医学院,兰州 730030||兰州大学第二医院核磁共振科,兰州 730030兰州大学第二临床医学院,兰州 730030||兰州大学第二医院核磁共振科,兰州 730030兰州大学第二临床医学院,兰州 730030||兰州大学第二医院核磁共振科,兰州 730030兰州大学第二临床医学院,兰州 730030甘肃省中医药大学,兰州 730030兰州大学第二临床医学院,兰州 730030||兰州大学第二医院核磁共振科,兰州 730030

医药卫生

胶质瘤端粒酶逆转录酶启动子突变影像组学深度学习磁共振成像术前预测

gliomatelomerase reverse transcriptase promoter mutationradiomicsdeep learningmagnetic resonance imagingpreoperative prediction

《磁共振成像》 2026 (8)

154-160,183,8

Cuiying Scholars Research Cultivation Program Project of Lanzhou University Second Hospital(No.CXYZ2022-43,CYXZPT2026-11)Cuiying Scientific and Technological Innovation Program of Lanzhou University Second Hospital(No.CY2024-LC-B02). 兰州大学第二医院萃英学子科研培育计划项目(编号:CXYZ2022-43、CYXZPT2026-11)兰州大学第二医院萃英科技创新计划项目(编号:CY2024-LC-B02)

10.12015/issn.1674-8034.2026.08.017

评论