[1587] Comparison of Manual Count and Image Analysis Methods in Reporting Ki-67 Index as Prognostic/Predictive Marker for Breast Cancer

Lik Hang Lee, Gilbert Bigras, Robert T Ogilvie, Hua Yang. University of Calgary, Calgary, AB, Canada; University of Alberta, Edmonton, AB, Canada

Background: Ki-67 index is a known important prognostic/predictive factor in breast cancer and is increasingly requested by oncologists. However, the accuracy of manual Ki-67 evaluation is questionable with significant inter-observer variability and intra-tumour heterogeneity. Additionally, it is time consuming in an era of cost cutting and increased pathology workload. Thus, there is interest in automated digital assessment of Ki-67 to increase precision, accuracy, and efficiency. The objective of this study is to compare automated image analysis and manual counting of Ki-67 in invasive breast carcinoma.
Design: Cases of invasive ductal carcinoma with Ki-67 index requests at our institution were identified. A breast pathologist reviewed the cases and selected the block with highest mitotic activity. A slide was made and stained for Ki-67 (Dako MIB-1 stain at 1/200 dilution, DAB chromogene) with haematoxylin counter-stain. The Ki-67 index was reported as the average independent count of two breast pathologists. The slides were then digitized using a Nikon Eclipse E600 microscope (20X objective, 0.40 aperture) with a QImaging MicroPublisher 5.0 camera (24-bit color, 2560 x 1920 pixel sensor). A priori background correction was applied. Three images targeting the most proliferative areas were analyzed with the ImmunoRatio application plugin in the ImageJ environment. The software identified carcinoma nuclei based on nuclear circularity and calculated the DAB-to-nuclear area ratio for the Ki-67 index. Manual and automated analysis results were compared with the paired samples t-test.
Results: 21 cases of were identified. There was a significant difference in the Ki-67 index between manual count (M=29.4, SD=13.2) and automated count (M= 19.0, SD=9.8; t(20)=6.22, p<0.001). The automated count was lower than the manual score (mean difference = 10.5, 95% CI=6.1 to 14.0) in all but one case, which was equal.
Conclusions: Automated assessment provides a consistently lower Ki-67 index than manual assessment. We postulate that this is due to the software algorithm used to identify neoplastic epithelial cells. Non-tumour elements, which usually do not stain for Ki67, may be mistaken as tumor, including benign epithelial cells, lymphocytes, and visual artifacts. With improved techniques or algorithms to isolate neoplastic epithelial cells, automated Ki-67 accuracy can be improved, potentially making it a viable alternative to manual Ki-67 assessment.
Category: Informatics

Monday, March 4, 2013 9:30 AM

Poster Session I Stowell-Orbison/Surgical Pathology/Autopsy Awards Poster Session # 251, Monday Morning

 

Close Window