Quantitative image analysis in the assessment of diffuse large B-cell lymphoma.

Proliferation rates in diffuse large B-cell lymphoma have been associated with conflicting outcomes in the literature, more often with high proliferation associated with poor prognosis. In most studies, the proliferation rate was estimated by a pathologist using an immunohistochemical stain for the monoclonal antibody Ki-67. We hypothesized that a quantitative image analysis algorithm would give a more accurate estimate of the proliferation rate, leading to better associations with survival. In all, 84 cases of diffuse large B-cell lymphoma were selected according to the World Health Organization criteria. Ki-67 percentage positivity estimated by the pathologist was recorded from the original report. The same slides were then scanned using an Aperio ImageScope, and Ki-67 percentage positivity was calculated using a computer-based quantitative immunohistochemistry nuclear algorithm. In addition, chart review was performed and survival time was recorded. The Ki-67 percentage estimated by the pathologist from the original report versus quantitative image analysis was significantly correlated (P<0.001), but pathologist Ki-67 percentages were significantly higher than quantitative image analysis (P=0.021). There was less agreement at lower Ki-67 percentages. Comparison of Ki-67 percentage positivity versus survival did not show significant association either with pathologist estimate or quantitative image analysis. However, although not significant, there was a trend of worse survival at higher proliferation rates detected by the pathologist but not by quantitative image analysis. Interestingly, our data suggest that the Ki-67 percentage positivity as assessed by the pathologist may be more closely associated with survival outcome than that identified by quantitative image analysis. This may indicate that pathologists are better at selecting appropriate areas of the slide. More cases are needed to assess whether this finding would be statistically significant. Due to the good correlation between pathologist estimate and quantitative image analysis, there is no substantial benefit to using quantitative image analysis at this point of time.