摘要
磁共振成像(magnetic resonance imaging,MRI)扫描时间长是其临床应用的主要瓶颈。传统压缩感知技术虽可加速采集,但在高加速因子下易产生伪影,且对复杂解剖结构的重建效果有限。人工智能辅助压缩感知(artificial intelligence-assisted compressed sensing,ACS)技术通过将深度学习(deep learning,DL)[如卷积神经网络(convolutional neural network,CNN)、生成对抗网络(generative adversarial networks,GANs)]与压缩感知原理在端到端框架中融合,实现了显著加速(通常>2倍)并致力于保留诊断特征,为突破上述瓶颈提供了有前景的解决方案。然而,ACS技术当前面临关键挑战,包括缺乏标准化的加速因子、算法对不同解剖部位和病变异质性的泛化能力不足,以及对细微病变(如微小转移淋巴结)诊断效能的验证尚不充分。此外,既往综述多聚焦于单一系统或纯技术层面,缺乏对ACS技术在全身多器官临床应用效果的系统评价。本综述旨在系统梳理ACS技术的原理演进与MRI临床研究进展,重点评述其在头颈、骨关节、心胸、腹部及盆腔等多系统成像中的优势、局限性与现存挑战,以期为优化ACS的临床应用提供依据,并为未来研究方向提供指引,推动MRI向更精准、高效、智能化的方向发展。
Prolonged scan times remain a major bottleneck for the clinical utility of magnetic resonance imaging(MRI).While conventional compressed sensing techniques accelerate acquisition,they often introduce artifacts and exhibit limited efficacy in reconstructing complex anatomical structures at high acceleration factors.Artificial intelligence-assisted compressed sensing(ACS)addresses these limitations by integrating deep learning(DL)architectures—such as convolutional neural networks(CNNs)and generative adversarial networks(GANs)—with compressed sensing principles within end-to-end frameworks.This synergy enables substantial acceleration(>2×)while preserving diagnostic features.However,ACS faces critical challenges:lack of standardized acceleration factors,insufficient algorithm generalizability across diverse anatomies and pathological heterogeneity,and inadequate validation of diagnostic efficacy for subtle lesions(e.g.,small metastatic lymph nodes).Furthermore,existing reviews predominantly focus on single-system applications or purely technical aspects,lacking a systematic evaluation of ACS's clinical utility across multiple body regions.This review systematically synthesizes technological advancements and MRI clinical progress in ACS,critically evaluating its strengths,limitations,and unresolved challenges in multi-system imaging(head-neck,musculoskeletal,cardiothoracic,abdominal,pelvic).We aim to provide evidence-based guidance for optimizing clinical implementation of ACS and direct future research toward advancing precision,efficiency,and intelligence in MRI.
作者
李涛
殷硕
张玄霄
张惠茅
周宏伟
LI Tao;YIN Shuo;ZHANG Xuanxiao;ZHANG Huimao;ZHOU Hongwei(Department of Radiology,the First Hospital of Jilin University,Changchun 130021,China)
出处
《磁共振成像》
北大核心
2025年第8期228-234,共7页
Chinese Journal of Magnetic Resonance Imaging
基金
吉林省科技发展计划项目(编号:YDZJ202402029CXJD)
吉林省医疗卫生人才专项(编号:JLSRCZX2025-010)。
关键词
人工智能
压缩感知
磁共振成像
脑血管疾病
骨关节疾病
冠状动脉疾病
腹部疾病
子宫相关病变
artificial intelligence
compressed sensing
magnetic resonance imaging
cerebrovascular diseases
bone and joint diseases
coronary artery diseases
abdominal diseases
uterine-related disorders
