文献速递心血管技术与安全MR

心脏MRI影像组学与纹理分析鉴别左室肥厚病因的系统综述

Cardiac MRI radiomics and texture analysis for differentiating the causes of left ventricular hypertrophy: a systematic review.
PubMed 文献 · Eur J Radiol · 2026-09-06

本系统综述纳入11项研究(2019至2025年,91至621例,多为单中心回顾性),评估心脏MRI影像组学与纹理分析鉴别左室肥厚病因的证据。结果显示所有鉴别对比的GRADE证据确定性均为极低,无一对比支持有效合并。现有高精度结果均来自内部验证且存在高偏倚风险,尚不能支持临床可信的病因鉴别。

要点

  • 所有鉴别对比的证据确定性均为极低,不宜据现有影像组学结果直接判断左室肥厚病因。
  • 高精度结果多来自内部验证且偏倚风险高,阅片时应警惕过拟合导致的虚高准确率。
  • 心脏MRI影像组学目前仅宜作为辅助性、假设生成性工具,不能替代常规影像与临床综合判断。

核心结论

  • 现有证据尚不支持心脏MRI影像组学与纹理分析对左室肥厚病因的临床可信鉴别,所有表面高精度均来自内部验证的高偏倚风险研究,GRADE确定性为极低。

研究设计与样本

  • 本系统综述遵循PRISMA-DTA,以病因学对比为分析单位,由两名评价者使用QUADAS-2、QUADAS-AI和METRICS评估偏倚风险、影像组学质量与报告完整性,并用GRADE评定确定性,未行meta分析。
  • 共纳入11项研究(2019至2025年,91至621例,多为单中心回顾性),病因对比存在重叠:6项HCM对比高血压性心脏病,4项心脏淀粉样变性对比HCM,另有肌节基因型与主动脉瓣狭窄各1项,2项多分类研究。

诊断标准与关键数值

  • 11项研究中仅4项可重建验证集2×2表,其中仅3项来自留出集或交叉验证集。
  • 在心脏淀粉样变性与HCM的对比中,4项研究均未报告逐类别例数。
  • 11项研究中无任何研究报告外部地理验证,8项研究存在过拟合或过拟合高风险,无一对比支持有效合并。

临床应用与阅片要点

  • 阅片报告中不应将心脏MRI影像组学特征作为左室肥厚病因诊断的独立依据,应结合常规电影序列、延迟强化、心肌厚度分布与临床信息综合判断。
  • 解读此类研究时需核查是否提供外部验证与逐类别例数,仅内部验证的高精度结果存在明显过拟合风险。

局限性

  • 纳入研究多为单中心回顾性设计,8项研究存在过拟合或过拟合高风险,无任何研究报告外部地理验证。
  • 仅少数研究能重建验证集2×2表,关键对比(心脏淀粉样变性与HCM)均未报告逐类别例数,导致无法有效合并。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

BACKGROUND: Because the causes of left ventricular hypertrophy (LVH) share a thickened-wall phenotype yet demand divergent management, from transthyretin stabilization in cardiac amyloidosis (CA) to sudden-death prevention in hypertrophic cardiomyopathy (HCM), identifying the cause is clinically decisive. Cardiac MRI radiomics and texture analysis are proposed for this differential. We asked whether this evidence supports clinically credible differentiation and whether each accuracy could be verified.

METHODS: Following PRISMA-DTA, the etiological contrast was the unit of analysis. Two reviewers appraised risk of bias with QUADAS-2 and QUADAS-AI, radiomic quality with METRICS, and reporting completeness, then rated certainty with GRADE. Evidence was synthesized without meta-analysis.

RESULTS: Eleven studies (2019 to 2025; 91 to 621 participants, mostly retrospective, single-center) were included. By etiological contrast (studies overlapped): six HCM versus hypertensive heart disease (HHD), four CA versus HCM, one each for sarcomeric genotypes and aortic stenosis, two multiclass. A validation 2 × 2 table was reconstructable in 4 of 11 studies, only 3 from a held-out or cross-validated set; for CA versus HCM, none of 4 reported per-class counts. No study reported external geographic validation, overfitting or its high risk was evident in 8 of 11, and no contrast supported valid pooling. Certainty was very low throughout.

CONCLUSIONS: The evidence does not yet support clinically credible differentiation of LVH causes: every apparently high accuracy rests on internally validated, high-risk-of-bias studies of very low certainty; radiomics remains adjunctive and hypothesis-generating. Progress requires external validation, verifiable reporting, and conventional comparators; we distill existing standards into a minimum reporting set.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Martínez Safar, De La Ossa Díaz, Martínez D, et al. Cardiac MRI radiomics and texture analysis for differentiating the causes of left ventricular hypertrophy: a systematic review. Eur J Radiol. 2026. PMID: 42727224. doi:10.1016/j.ejrad.2026.113219

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。