Scientific Literature

Comparison of Ki67 counted by digital pathology image analysis software application and scores determined by pathologists


N. Niikura, N. Kumaki, S. Masuda, T. Xiaoyan, M. Miyazawa, T. Iwamoto, T. Okamura, Y. Saito and Y. Tokuda , Cancer Research , DOI: 10.1158/0008-5472.SABCS13-P3-05-11

Background: Immunohistochemical assessment of the Ki67 has been described as a prognostic and predictive marker for breast cancer. The international panel of experts recommends the counting of as many as 500 or 1000 cancer cells including those in hot spots, which in a routine clinical setting, will be time consuming for a pathologist. Recently, a digital pathology image analysis software application (DPIA) for measuring biomarkers within pre-defined regions of interest was developed. However, the clinical efficacy of Ki67 counted by a DPIA has not been evaluated in a clinical setting. This study was aimed at comparing Ki67 counted by DPIA without a pathologist and Ki67 scores determined by a pathologist. In addition, the clinical outcomes predicted by the Ki67 counted by the DPIA and scores determined by pathologists were compared.
Patients and methods: Between January 1, 2005, and December 31, 2010, we retrospectively identified all patients listed in the pathology database of the Tokai University for whom IHC Ki67 data were available. Ki67 levels were classified by pathologists into score categories such as <10, 10, 20, 30, 40, 50, 60, 70, 80, 90, and >90 by observation under the high power magnification. For the estimation of Ki67 positive and negative cells, 3–5 areas with an average degree of Ki67 positive cell distribution were selected by observation under low power magnification. Immunohistochemically stained slides were scanned and converted to whole slide image data using Nano Zoomer 2.0-HT (Hamamatsu Photonics, Hamamatsu, Japan). We used Tissue Studio 3 (Definiens AG, Munich, Germany) as the DPIA, and Ki67-positive and Ki67-negative cells on each virtual slide, whole slide, and the most strongly positive area were counted. Ki67 levels were categorized as low and high if the Ki67 scores were ≤13.25% and >13.25%, respectively. To assess the relationship between Ki67 and survival outcomes, survival curves were calculated by the Kaplan–Meier method and were compared using the log-rank test.
Results: We identified 1902 slides with data that have virtual slide and counted Ki67 by DPIA and reported ki67 score by pathologist. Of the 1902 data, we identified 1003 slide with survival data. The 2 × 2 analysis revealed that Ki67 counted by DPIA and ki67 scored by the pathologists were moderately correlated for all cases (κ: 0.41 [95% CI: 0.360–0.448]; sensitivity: 0.573 [95% CI: 0.552–0.590]; specificity: 0.878 [95% CI: 0.845–0.906]). Among estrogen receptor (ER)-positive and stage I or II cases (n = 645), when Ki67 scores determined by pathologists were analyzed, patients with high Ki67 scores had poorer relapse-free survival (RFS) than those with low Ki67 scores (p < 0.001). When Ki67 counted by DPIA were analyzed, patients with high Ki67 had poorer RFS than those with low Ki67(p = 0.031).
Conclusions: The study results demonstrate that Ki67 counted by the DPIA determined without a pathologist and Ki67 scores determined by pathologists were moderately correlated and associated with survival in ER-positive cases. This indicates that Ki67 scores by the DPIA determined without a pathologist can be used for measuring Ki67 values. However, Ki67 scores determined by pathologists were better for predicting survival.