Training the ACRIN 6666 Investigators and effects of feedback on breast ultrasound interpretive performance and agreement in BI-RADS ultrasound feature analysis.
Training the ACRIN 6666 Investigators and effects of feedback on breast ultrasound interpretive performance and agreement in BI-RADS ultrasound feature analysis.
复制标题
DOI：
10.2214/ajr.11.7324
复制
发表时间：
2012-07
期刊：
影响因子：
--
通讯作者：
Mendelson EB
中科院分区：
文献类型：
--
作者：
Berg WA;Blume JD;Cormack JB;Mendelson EB
为美国放射学会成像网络（ACRIN）6666研究者开发了乳房X线摄影和乳腺超声的资格鉴定任务。我们试图评估反馈对乳腺超声判读性能的影响，并在这些经验丰富的观察者的一个子集中进行BI-RADS特征分析。在1小时的BI-RADS：超声教学课程后，向100名经验丰富的乳腺成像观察员展示了70组正交乳腺超声图像（包括25例（36%）恶性肿瘤）的解释技能测验集。35名观察员对测验集进行了两次审查：第一次没有，然后立即反馈共识特征分析，管理建议和病理真相。在无反馈和有反馈的情况下计算观察员性能（灵敏度、特异性、曲线下面积[AUC]）。确定Kappas，以获得特征分析和评估的一致性。对于35名没有反馈的观察者，平均灵敏度为89%（范围，68-100%）;特异性为62%（范围，42-82%）; AUC为82%（范围，73-89%）。在反馈的情况下，平均灵敏度为93%，（范围，80-100%;平均增加，4%;增加范围，0-12%; p < 0.0001），平均特异性为61%（范围，45-73%;平均降低，1%;变化范围，-18%至11%; p = 0.19），平均AUC为84%（范围，78-90%;平均增加，2%;变化范围，-3%至9%; p < 0.0001）。初始性能最低四分位数的三个乳腺成像仪显示灵敏度的最大改善，AUC无变化或改善。特征分析的kappa值没有变化，但最终评估的一致性有所改善，kappa值从无反馈的0.53（SE，0.02）增加到有反馈的0.59（SE，0.02）（p < 0.0001）。大多数有经验的乳腺成像显示出出色的乳腺超声判读技能。共识BI-RADS的即时反馈：超声特征和组织病理学结果改善了所有经验变量的超声判读性能。
Qualification tasks in mammography and breast ultrasound were developed for the American College of Radiology Imaging Network (ACRIN) 6666 Investigators. We sought to assess the effects of feedback on breast ultrasound interpretive performance and agreement in BI-RADS feature analysis among a subset of these experienced observers. After a 1-hour didactic session on BI-RADS: Ultrasound, an interpretive skills quiz set of 70 orthogonal sets of breast ultrasound images including 25 (36%) malignancies was presented to 100 experienced breast imaging observers. Thirty-five observers reviewed the quiz set twice: first without and then with immediate feedback of consensus feature analysis, management recommendations, and pathologic truth. Observer performance (sensitivity, specificity, area under the curve [AUC]) was calculated without feedback and with feedback. Kappas were determined for agreement on feature analysis and assessments. For 35 observers without feedback, the mean sensitivity was 89% (range, 68–100%); specificity, 62% (range, 42–82%); and AUC, 82% (range, 73–89%). With feedback, the mean sensitivity was 93% (range, 80–100%; mean increase, 4%; range of increase, 0–12%; p < 0.0001), the mean specificity was 61% (range, 45–73%; mean decrease, 1%; range of change, −18% to 11%; p = 0.19), and the mean AUC was 84% (range, 78–90%; mean increase, 2%; range of change, −3% to 9%; p < 0.0001). Three breast imagers in the lowest quartile of initial performance showed the greatest improvement in sensitivity with no change or improvement in AUC. The kappa values for feature analysis did not change, but there was improved agreement about final assessments, with the kappa value increasing from 0.53 (SE, 0.02) without feedback to 0.59 (SE, 0.02) with feedback (p < 0.0001). Most experienced breast imagers showed excellent breast ultrasound interpretive skills. Immediate feedback of consensus BI-RADS: Ultrasound features and histopathologic results improved performance in ultrasound interpretation across all experience variables.