Institute of Pathology, Charité University Hospital Berlin, Charitéplatz 1, Berlin, Germany.
Institute of Pathology, Charité University Hospital Berlin, Charitéplatz 1, Berlin, Germany. VMscope GmbH, Charitéplatz 1, Berlin, Germany.
Clin Cancer Res. 2015 Aug 15;21(16):3651-7. doi: 10.1158/1078-0432.CCR-14-1283. Epub 2014 Dec 11.
Scoring proliferation through Ki67 immunohistochemistry is an important component in predicting therapy response to chemotherapy in patients with breast cancer. However, recent studies have cast doubt on the reliability of "visual" Ki67 scoring in the multicenter setting, particularly in the lower, yet clinically important, proliferation range. Therefore, an accurate and standardized Ki67 scoring is pivotal both in routine diagnostics and larger multicenter studies.
We validated a novel fully automated Ki67 scoring approach that relies on only minimal a priori knowledge on cell properties and requires no training data for calibration. We applied our approach to 1,082 breast cancer samples from the neoadjuvant GeparTrio trial and compared the performance of automated and manual Ki67 scoring.
The three groups of autoKi67 as defined by low (≤ 15%), medium (15.1%-35%), and high (>35%) automated scores showed pCR rates of 5.8%, 16.9%, and 29.5%, respectively. AutoKi67 was significantly linked to prognosis with overall and progression-free survival P values P(OS) < 0.0001 and P(PFS) < 0.0002, compared with P(OS) < 0.0005 and P(PFS) < 0.0001 for manual Ki67 scoring. Moreover, automated Ki67 scoring was an independent prognosticator in the multivariate analysis with P(OS) = 0.002, P(PFS) = 0.009 (autoKi67) versus P(OS) = 0.007, PPFS = 0.004 (manual Ki67).
The computer-assisted Ki67 scoring approach presented here offers a standardized means of tumor cell proliferation assessment in breast cancer that correlated with clinical endpoints and is deployable in routine diagnostics. It may thus help to solve recently reported reliability concerns in Ki67 diagnostics.
通过 Ki67 免疫组化对增殖进行评分是预测乳腺癌患者化疗治疗反应的一个重要组成部分。然而，最近的研究对“视觉”Ki67 评分在多中心环境中的可靠性提出了质疑，尤其是在较低但具有临床意义的增殖范围内。因此，在常规诊断和更大的多中心研究中，准确和标准化的 Ki67 评分至关重要。
我们验证了一种新的全自动 Ki67 评分方法，该方法仅依赖于细胞特性的最小先验知识，并且不需要校准的训练数据。我们将该方法应用于 neoadjuvant GeparTrio 试验的 1082 例乳腺癌样本，并比较了自动和手动 Ki67 评分的性能。
根据低（≤ 15%）、中（15.1%-35%）和高（>35%）自动评分定义的三组 autoKi67 的 pCR 率分别为 5.8%、16.9%和 29.5%。与手动 Ki67 评分相比，autoKi67 与总生存期和无进展生存期的预后显著相关，P 值分别为 P（OS）<0.0001 和 P（PFS）<0.0002。此外，与手动 Ki67 评分相比，在多变量分析中，自动化 Ki67 评分是一个独立的预后指标，P（OS）=0.002，P（PFS）=0.009（autoKi67）vs. P（OS）=0.007，P（PFS）=0.004（manual Ki67）。
这里提出的计算机辅助 Ki67 评分方法提供了一种标准化的乳腺癌肿瘤细胞增殖评估方法，与临床终点相关，并且可以在常规诊断中使用。因此，它可能有助于解决最近报道的 Ki67 诊断中的可靠性问题。
