超声医师自报的绝经后正常卵巢经阴道超声可视化不可靠:英国卵巢癌筛查合作试验存档图像的专家复核结果
该研究从英国卵巢癌筛查合作试验(UKCTOCS)2008至2011年间68931次经阴道超声检查中随机抽取1000次报告卵巢「可见且正常」的检查,由一位专家复核存档图像,并开发软件自动测量可视化率。专家确认双侧卵巢可视化仅50.2%(502/1000),提示超声医师可能将其他结构误认为正常卵巢,基于自报的可视化质量控制指标不可靠。
要点
- 专家复核确认双侧卵巢可视化率为50.2%(502/1000),远低于超声医师自报的「可见且正常」。
- 分类器在测试数据(133次检查)中检测专家确认的双侧卵巢可视化灵敏度为90.5%(95%CI 80.9-95.8%),特异度为47.5%(95%CI 34.5-60.8%)。
- 作者认为基于自报正常卵巢可视化的质量控制指标不可靠。
核心结论
- 作者认为,相当大比例的经阴道超声年度筛查中,超声医师可能将其他结构误认为正常卵巢。
- 作者认为基于自报正常卵巢可视化的质量控制指标不可靠。
- 作者指出尚不确定这是否影响UKCTOCS超声组的卵巢癌检出灵敏度及检出时分期。
- 作者认为分类器在解决该问题上显示一定潜力,但需要进一步研究。
研究设计与样本
- 单一位专家复核从UKCTOCS 2008至2011年间68931次经阴道超声检查中随机抽取的1000次报告卵巢「可见且正常」的存档图像。
- 开发软件通过测量图像中卡尺尺寸并与超声医师记录值匹配,以识别用于测量卵巢的精确图像。
- 使用专家报告的卵巢尺寸和可视化数据训练并验证逻辑回归分类器以判定可视化。
主要发现
- 专家复核确认双侧卵巢可视化(VR-Both)的检查占50.2%(502/1000)。
- 软件在534次检查中识别出测量图像,按2:1:1分为训练、验证和测试数据。
- 分类器在验证数据上的平均准确率为70.9%(95%CI 70.0-71.8%)。
- 在测试数据(133次检查)中,检测专家确认的双侧卵巢可视化灵敏度为90.5%(95%CI 80.9-95.8%),特异度为47.5%(95%CI 34.5-60.8%)。
临床怎么用
- 原文未据此给出推荐。
局限性
- 作者指出尚不确定超声医师误认结构是否影响UKCTOCS超声组卵巢癌检出灵敏度及检出时分期。
- 作者指出分类器显示一定潜力但需要进一步研究。
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)。
原文
OBJECTIVE: In the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS), self-reported visualization rate (VR) of the ovaries by the sonographer on annual transvaginal sonographic (TVS) examinations was a key quality control (QC) metric. The objective of this study was to assess self-reported VR using expert review of a random sample of archived images of TVS examinations from UKCTOCS, and then to develop software for measuring VR automatically. METHODS: A single expert reviewed images archived from 1000 TVS examinations selected randomly from 68 931 TVS scans performed in UKCTOCS between 2008 and 2011 with ovaries reported as 'seen and normal'. Software was developed to identify the exact images used by the sonographer to measure the ovaries. This was achieved by measuring caliper dimensions in the image and matching them to those recorded by the sonographer. A logistic regression classifier to determine visualization was trained and validated using ovarian dimensions and visualization data reported by the expert. RESULTS: The expert reviewer confirmed visualization of both ovaries (VR-Both) in 50.2% (502/1000) of the examinations. The software identified the measurement image in 534 exams, which were split 2:1:1 providing training, validation and test data. Classifier mean accuracy on validation data was 70.9% (95% CI, 70.0-71.8%). Analysis of test data (133 exams) provided a sensitivity of 90.5% (95% CI, 80.9-95.8%) and specificity of 47.5% (95% CI, 34.5-60.8%) in detecting expert confirmed visualization of both ovaries. CONCLUSIONS: Our results suggest that, in a significant proportion of TVS annual screens, the sonographers may have mistaken other structures for normal ovaries. It is uncertain whether or not this affected the sensitivity and stage at detection of ovarian cancer in the ultrasound arm of UKCTOCS, but we conclude that QC metrics based on self-reported visualization of normal ovaries are unreliable. The classifier shows some potential for addressing this problem, though further research is needed. © 2017 The Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of the International Society of Ultrasound in Obstetrics and Gynecology.
引用
- Stott W, Campbell S, Franchini A, et al. Sonographers' self-reported visualization of normal postmenopausal ovaries on transvaginal ultrasound is not reliable: results of expert review of archived images from UKCTOCS. Ultrasound Obstet Gynecol. 2018. PMID: 28796383. doi:10.1002/uog.18836
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。