1000例阅片者研究:放射科医生在计算机辅助检测系统辅助下解读自动乳腺容积扫描仪图像的表现
该回顾性观察者研究纳入1000例自动乳腺容积扫描(ABVS)病例(206例恶性、486例良性、308例正常),由9名不同经验放射科医生在有/无CADe辅助下阅片。全体阅片者AUC在有CADe时为0.784、无CADe时为0.747,差异有统计学意义(p<0.05);平均阅片时间有CADe时快9.3%(p<0.05)。CADe对ABVS经验不足者获益更大。
要点
- CADe辅助使全体阅片者AUC从0.747升至0.784,差异有统计学意义(p<0.05)。
- CADe对ABVS经验不足者AUC提升更明显:B组0.696→0.757,C组0.718→0.759。
- 有CADe时平均阅片时间快9.3%(p<0.05),CADe结果与ABVS图像同时呈现供阅片者同步阅读。
核心结论
- CADe可提高放射科医生使用ABVS检测乳腺癌的准确性和阅片时间,对ABVS经验不足者获益更大。
研究设计与样本
- 回顾性观察者研究,多阅片者多病例设计,纳入本机构2012年ABVS检查中选取的1000例(206例恶性、486例良性、308例正常),恶性与良性经活检或手术证实、正常经2年随访证实,9名放射科医生参与。
诊断标准与关键数值
- 全体阅片者AUC有CADe为0.784、无CADe为0.747;A组有CADe 0.833、无CADe 0.829;B组有CADe 0.757、无CADe 0.696;C组有CADe 0.759、无CADe 0.718;除A组外各AUC差异均有统计学意义(p<0.05)。
临床应用与阅片要点
- CADe结果与ABVS图像同时呈现,阅片者同步阅读;A组(超声经验>8年且ABVS经验>4年)、B组(超声经验>5年、无ABVS经验)、C组(超声经验>1年、无ABVS经验)中B组和C组从CADe获益更明显。
局限性
- 原文未报告局限性
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
The objective of our study was to assess, in a reader study, radiologists' performance in interpretation of automated breast volume scanner (ABVS) images with the aid of a computer-aided detection (CADe) system. Our study is a retrospective observer study with the purpose of investigating the effectiveness of using a CADe system as an aid for radiologists in interpretation of ABVS images. The multiple-reader, multiple-case study was designed to compare the diagnostic performance of radiologists with and without CADe. The study included 1000 cases selected from ABVS examinations in our institution in 2012. Among those cases were 206 malignant, 486 benign and 308 normal cases. The cancer cases were consecutive; the benign and normal cases were randomly selected. All malignant and benign cases were confirmed by biopsy or surgery, and normal cases were confirmed by 2-y follow-up. Reader performance was compared in terms of area under the receiver operating characteristic curve, sensitivity and specificity. Additionally, the reading time per case for each reader was recorded. Nine radiologists from our institution participated in the study. Three had more than 8 y of ultrasound experience and more than 4 y of ABVS experience (group A); 3 had more than 5 y of ultrasound experience (group B), and 3 had more than 1 y of ultrasound experience (group C). Both group B and group C had no ABVS experience. The CADe system used was the QVCAD System (QView Medical, Inc., Los Altos, CA, USA). It is designed to aid radiologists in searching for suspicious areas in ABVS images. CADe results are presented to the reader simultaneously with the ABVS images; that is, the radiologists read the ABVS images concurrently with the CADe results. The cases were randomly assigned for each reader into two equal-size groups, 1 and 2. Initially the readers read their group 1 cases with the aid of CADe and their group 2 cases without CADe. After a 1-mo washout period, they re-read their group 1 cases without CADe and their group 2 cases with CADe. The areas under the receiver operating characteristic curves of all readers were 0.784 for reading with CADe and 0.747 without CADe. Areas under the curves with and without CADe were 0.833 and 0.829 for group A, 0.757 and 0.696 for group B and 0.759 and 0.718 for group C. All differences in areas under the curve were statistically significant (p <0.05), except that for group A. The average reading time was 9.3% (p < < 0.05) faster with CADe for all readers. In summary, CADe improves radiologist performance with respect to both accuracy and reading time for the detection of breast cancer using the ABVS, with the greater benefit for those inexperienced with ABVS.
引用
- Xu X, Bao L, Tan Y, et al. 1000-Case Reader Study of Radiologists' Performance in Interpretation of Automated Breast Volume Scanner Images with a Computer-Aided Detection System. Ultrasound Med Biol. 2018. PMID: 29853222. doi:10.1016/j.ultrasmedbio.2018.04.020
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。