自动 ASPECTS 评分软件的 MRMC 验证研究
本研究采用完全交叉的多阅片者多病例研究,8 名临床医生分别独立评价 100 例大脑中动脉或颈内动脉闭塞患者的 NCCT,比较辅助 StrokeSENS ASPECTS 前后的表现。结果辅助后平衡准确率提高 5.7 个百分点,阅片者间一致性 Fleiss Kappa 从 0.323 升至 0.608。
要点
- AI 辅助可提高 NCCT 早期缺血改变的检出敏感度,尤其适合急诊卒中影像初筛。
- 辅助后阅片者间一致性显著改善,利于报告结果标准化与互认。
- 该软件主要用于 MCA/ICA 闭塞患者区域级 ASPECTS 评分,需结合全影像评估使用。
核心结论
- StrokeSENS ASPECTS 可提高临床医生在 NCCT 上检出早期缺血改变的能力并降低阅片者间差异。
研究设计与样本
- 研究类型为完全交叉的多阅片者多病例临床阅片研究。
- 8 名临床医生独立评价 100 例确诊大脑中动脉或颈内动脉闭塞患者的 NCCT。
- 参考标准为 3 名神经放射学专家共识结果。
- 读取条件包括未辅助和 StrokeSENS ASPECTS 辅助两种模式。
诊断标准与关键数值
- 比较对象为区域级早期缺血改变的 ASPECTS 评分,病变符合 MCA/ICA 闭塞背景。
- 辅助后较未辅助的平衡准确率提高 5.7 个百分点。
- 辅助后总体准确率、敏感度、特异度分别提高 2.6、9.7、1.6 个百分点,P 值均小于 0.001。
- 未辅助时阅片者间 Fleiss Kappa 为 0.323%,辅助后为 0.608%,提高 0.285。
临床应用与阅片要点
- 该软件可作为急诊 NCCT 卒中影像阅片的辅助工具,帮助提高早期缺血改变识别能力。
- 在报告 ASPECTS 评分或早期缺血改变区域描述时,可参考软件输出进行核对。
- 本研究所涉专家共识仍是参考影像,使用软件时应注意结合临床情况与全序列综合判断。
局限性
- 本研究对象仅限于确诊 MCA/ICA 闭塞的卒中患者,未覆盖其他血管分布区卒中。
- 参考标准为专家共识而非独立病理或随访金标准。
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: Rapid, accurate detection of early ischemic changes (EIC) on non-contrast computed tomography (NCCT) is critical for the triage and treatment of acute ischemic stroke (AIS) patients, yet NCCT interpretation remains challenging due to low soft-tissue contrast and inter-reader variability.
MATERIALS AND METHODS: This study validates StrokeSENS ASPECTS, a fully automated AI-based software tool designed to aid clinicians in grading region-level EIC. A fully crossed multi-reader multi-case (MRMC) clinical reader study was conducted with eight clinicians who independently scored 100 NCCT scans from patients with confirmed middle cerebral artery (MCA)/internal carotid artery (ICA) occlusion, unaided and aided by StrokeSENS ASPECTS, with a three-expert-neuroradiologist consensus serving as the reference standard. Reader performance was assessed using binary classification metrics, including balanced accuracy, overall accuracy, sensitivity, and specificity; inter-reader agreement was assessed using Fleiss's kappa.
RESULTS: The use of StrokeSENS ASPECTS improved readers' balanced accuracy by 5.7 percentage points, whereas overall accuracy, sensitivity, and specificity improved by 2.6, 9.7, and 1.6 percentage points, respectively, when compared to the unaided baseline (p < 0.001). Inter-reader agreement showed a significant increase in Fleiss's Kappa of 0.285 from 0.323% (unaided) to 0.608% (aided).
CONCLUSION: StrokeSENS ASPECTS have shown to improve clinicians' ability to detect EIC on NCCT and reduce inter-reader variability. This demonstrates StrokeSENS ASPECTS's safety and effectiveness as an aid in the evaluation of AIS.
引用
- Golan, Duarte, Duszynski, et al. Validation of an Automated ASPECTS Software via a Multi-Reader Multi-Case Clinical Reader Study. AJNR Am J Neuroradiol. 2026. PMID: 42373208. doi:10.3174/ajnr.A9504
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。