Dean Judy C, Ilvento Christina C
AJR Am J Roentgenol. 2006 Jul;187(1):20-8. doi: 10.2214/AJR.05.0111.
This study prospectively evaluated a computer-aided detection (CAD) device used with diagnostic and screening mammography by assessing cancers detected; tumor sizes, histology, and stage; positive predictive value (PPV) of biopsy recommendation; and recall rates before and after CAD introduction.
Interpretations of 9,520 consecutive mammograms were recorded without and then with CAD for a 28-month period. Cancer detections based on initial radiologist review and additional detections based on CAD findings were noted. Recall rates, tumor size and histology, and PPV of biopsy recommendation before and after the introduction of CAD were compared.
Cancers detected only with CAD assistance were 9.6% of all cancers (10 of 104); screening-detected cancers increased 13.3% with CAD assistance (four in addition to 30 screening-detected cancers). The 95% one-sided confidence boundary using binomial distribution is consistent with at least 5.3% for all cancers and 5.1% for nonpalpable cancers. The greatest impact was on ductal carcinoma in situ, for which CAD increased cancer detection by 14.2% (three added to 21). Similar percentages of cancers were detected only with CAD assistance in both screening (11.4%; 4 of 35) and diagnostic (9.5%; six of 63) studies. Additional cancers were detected using CAD in patients with implants and previous lumpectomy. The additional cancers detected with CAD were smaller (p = 0.01 for all cancers, p = 0.03 for nonpalpable invasive cancer). The screening recall rate increased from 6.2% to 7.8% after CAD, with a decrease in the biopsy rate and a nonsignificant increase in the biopsy PPV from 21.9% to 26.3%.
CAD resulted in detection of more cancers in screening and diagnostic patients, with an increased recall rate but no deterioration in PPV of biopsy. Additional cancers detected were significantly smaller.
本研究通过评估检测出的癌症、肿瘤大小、组织学类型和分期、活检建议的阳性预测值(PPV)以及引入计算机辅助检测(CAD)前后的召回率,对用于诊断性和筛查性乳腺钼靶检查的CAD设备进行前瞻性评估。
在28个月期间,记录了9520例连续乳腺钼靶检查的解读结果,先是在无CAD辅助的情况下,然后是在有CAD辅助的情况下。记录基于最初放射科医生阅片检测出的癌症以及基于CAD结果检测出的额外癌症。比较引入CAD前后的召回率、肿瘤大小和组织学类型以及活检建议的PPV。
仅在CAD辅助下检测出的癌症占所有癌症的9.6%(104例中的10例);在CAD辅助下筛查检测出的癌症增加了13.3%(除30例筛查检测出的癌症外,又增加了4例)。使用二项分布的95%单侧置信区间对于所有癌症至少为5.3%,对于不可触及的癌症至少为5.1%。最大的影响在于导管原位癌,CAD使导管原位癌的检测率提高了14.2%(在21例的基础上又增加了3例)。在筛查(11.4%;35例中的4例)和诊断(9.5%;63例中的6例)研究中,仅在CAD辅助下检测出的癌症比例相似。在有植入物和既往接受过肿块切除术的患者中,使用CAD检测出了额外的癌症。CAD检测出的额外癌症较小(所有癌症p = 0.01,不可触及的浸润性癌症p = 0.03)。引入CAD后,筛查召回率从6.2%提高到7.8%,活检率下降,活检PPV从21.9%增至26.3%,但差异无统计学意义。
CAD在筛查和诊断患者中检测出了更多癌症,召回率提高,但活检PPV没有恶化。检测出的额外癌症明显较小。