Comparison between Ki67 labeling index determined using image analysis software with virtual slide system and that determined visually in breast cancer

In recent papers, Ki67 labeling index (LI) has been used to classify breast cancer patients into the low and high Ki67LI groups for comparison studies, which showed significant differences in many prognostic factors. It has not been clarified whether image analysis software can be used for calculating LI in breast cancer. In our study, we examined whether Ki67LI in breast cancer calculated using image analysis software correlates with that measured on the basis of visual.

Fifty patients were randomly selected among breast cancer patients who underwent surgical operation from March, 2010 to May, 2010 in our hospital without preoperative chemotherapy. In this study, for the virtual slide system (VSS: VS120-L100, Olympus, Tokyo, Japan), the high-resolution VSs of all the 50 patients were prepared as samples. The image analysis software use for calculating LI was Tissuemorph Digital Pathology (Tissuemorph DP: Visiopharm, Hoersholm, Denmark). The calculated LI was extracted from 3 to 5 views containing hot spots. The LI calculated using Tissuemorph DP was designed as LI/image/T. The digital image of 3 to 5 LI/image/T views was printed out, and on the digital photograph, we counted visually the number of Ki67-immunopositive cells in exactly the same area, and the percentage of Ki67-immunopositive cells was designed as LI/direct. Moreover, a pathologist’s assistant (PA) determined the tumor area in the same specimen using VSS and calculated LI using Tissuemorph DP, which was designed as LI/image/PA. The chief pathologist (CP) similarly calculated LI which was designed as LI/image/CP. We evaluated the degree of agreement between different data sets “LI/image/T and LI/direct” and “LI/image/T, LI/image/CP, and LI/image/PA” by using interclass correlation coefficient (ICC).

The average counts of cells were as follows: LI/direct, 3209.7 ± 1970.4 (SD); LI/image/T, 2601.6 ± 1697.1; LI/image/PA, 2886.5 ± 2027.5; LI/image/CP, 18805.5 ± 22293.4. The values of LI/direct and LI/image/T showed almost perfect agreement as showed by an ICC of 0.885 (95 % CI, 0.806–0.933; p < 0.001). The agreement among three investigators was almost perfect. The obtained ICC was 0.825 (95 % CI, 0.739–0.890; p < 0.001) among the data of LI/image/T, LI/image/CP and LI/image/PA. There were five cases that immunopositivity for Ki67 showed a more than 10 % disagreement between LI/direct and LI/image/T.

The merits of calculating Ki67 LI using Tissuemorph DP are as follows. First, the staining intensity of the cells to be counted can be adjusted. Second, the portion of a tumor including “hot spots” for counting can be chosen. Third, many cancer cells can be counted more rapidly using Tissuemorph DP than by visual observation. However, it is important that pathologist should check and carry out the final decision of the data, when Ki67 LI using Tissuemorph DP is calculated.