文献速递神经CTMR

深度学习CT转MRI萎缩分析鉴别bvFTD与误诊

Differentiating behavioural variant frontotemporal dementia from misdiagnoses by deriving atrophy through deep-learning-based CT to MRI synthesis: A proof-of-concept study.
PubMed 文献 · Eur J Radiol · 2026-07-31

本研究为回顾性概念验证研究,纳入30例行为变异型额颞叶痴呆(bvFTD)和30例精神性误诊患者,利用SynthSR将连续CT合成MRI并经FastSurfer纵向分割提取萎缩指标。结果显示全额颞叶及前部亚区年化萎缩率在两组间显著差异(q均<0.001),全额颞叶指标AUC为0.79(95%CI 0.66-0.89),敏感度80%,特异度70%。

要点

  • 连续CT经深度学习合成MRI后可提取纵向萎缩轨迹,用于辅助鉴别bvFTD与精神性误诊
  • 全额颞叶年化萎缩率(Δ(wFT/eTIV))在Youden最佳截断值下敏感度80%,特异度70%,准确率75%
  • 较高分辨率的CT采集与合成MRI质量指标改善相关,阅片时需关注CT层厚与纵向一致性

核心结论

  • 纵向分析常规CT数据可捕获疾病相关的萎缩轨迹,在图像质量和纵向一致性充分的条件下,可作为区分bvFTD与精神性误诊的实用辅助手段
  • 该方法可重新利用常规采集的CT数据,有潜力整合入痴呆诊疗路径以改善bvFTD诊断

研究设计与样本

  • 回顾性概念验证研究,筛查112例来自三级认知障碍门诊疑诊bvFTD且行为综合征符合bvFTD并有连续CT的患者,图像质控后纳入30例bvFTD和30例精神性误诊
  • 使用SynthSR进行CT到MRI合成,FastSurfer纵向流进行分割,提取纵向区域体积统计量

关键影像表现/征象

  • 年化全额颞叶萎缩率归一化至颅内体积(Δ(wFT/eTIV))及其前部亚区对应指标(Δ(aFT/eTIV))在bvFTD组与误诊组间均显著不同,FDR校正后q均<0.001
  • Δ(wFT/eTIV)效应量大(Hedges g=-1.14),乐观校正后AUC为0.79(95%CI 0.66-0.89)
  • 较高分辨率CT采集与合成MRI质量指标的改善相关

诊断标准与关键数值

  • Youden最佳截断值下,Δ(wFT/eTIV)用于区分bvFTD与精神性误诊的敏感度为80%,特异度为70%,准确率为75%
  • bvFTD组与误诊组比较,Δ(wFT/eTIV)和Δ(aFT/eTIV)经FDR校正后q值均<0.001
  • Δ(wFT/eTIV)效应量Hedges g=-1.14,乐观校正后AUC为0.79(95%CI 0.66-0.89)

临床应用与阅片要点

  • 连续CT的纵向萎缩分析可作为痴呆诊疗路径中区分bvFTD与精神性误诊的实用辅助工具,尤其适用于早期表现不典型病例
  • 日常阅片需注意CT采集分辨率和纵向扫描一致性,图像质量与合成MRI质量直接相关
  • 定量萎缩分析可补充传统视觉判读,帮助识别额颞叶进行性萎缩轨迹

局限性

  • 结果来自单一三级认知障碍门诊,可能影响外部推广性
  • 依赖CT纵向一致性和图像质量,分辨率不足可能限制合成MRI与分割准确性

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

INTRODUCTION: Differentiating behavioural variant frontotemporal dementia (bvFTD) from psychiatric misdiagnoses of the disease (often referred to as 'phenocopies') remains a major diagnostic challenge, particularly at early presentation. Although Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) is often used in such cases, concomitantly acquired serial computed tomography (CT) scans are typically not analysed quantitatively for atrophy. We investigated whether longitudinal CT data could be used to distinguish bvFTD from psychiatric misdiagnoses of bvFTD, using an automated image-processing pipeline.

METHODS: In this retrospective proof-of-concept study, 112 patients from a tertiary cognitive disorders clinic who were referred with suspected bvFTD; had behavioural syndromes compatible with bvFTD; and had serial CT imaging available were screened. After image quality control, 30 patients with bvFTD and 30 psychiatric misdiagnoses of bvFTD were included. Longitudinal regional volume statistics were extracted using SynthSR for CT to Magnetic Resonance Imaging (MRI) synthesis and FastSurfer (longitudinal stream) for segmentation.

RESULTS: Annualised whole frontotemporal atrophy normalised to intracranial volume (Δ(wFT/eTIV)) and its anterior subregion counterpart (Δ(aFT/eTIV)) both significantly differed between groups after FDR correction (all q < 0.001). Δ(wFT/eTIV) showed a large effect size (Hedges' g = -1.14) and yielded an optimism-corrected area under the curve of 0.79 (95 % CI 0.66-0.89), with 80 % sensitivity, 70 % specificity, and 75 % accuracy at the Youden-optimal cutoff. Higher-resolution CT acquisitions were associated with improved synthetic MRI quality metrics.

CONCLUSION: In this proof-of-concept study, the findings show that longitudinal analysis of routinely acquired CT scans can capture disease-relevant atrophy trajectories and may provide a practical adjunct for differentiating bvFTD from psychiatric misdiagnoses of bvFTD, provided image quality and longitudinal consistency are sufficient. This approach can re-purpose routinely acquired CT and could potentially be incorporated in dementia pathways to improve bvFTD diagnosis.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Flavell, Shaw, Logan, et al. Differentiating behavioural variant frontotemporal dementia from misdiagnoses by deriving atrophy through deep-learning-based CT to MRI synthesis: A proof-of-concept study. Eur J Radiol. 2026. PMID: 42566955. doi:10.1016/j.ejrad.2026.113128

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。