Comparison of computerized image analysis with traditional semiquantitative scoring of Perls' Prussian Blue stained...

Image analysis is now routinely employed as a tool in toxicologic pathology to help quantitate end points of efficacy and safety. It is regarded as a proficient and a sensitive technique to generate numerical data that can be easily interrogated for statistical evaluation. Traditional semiquantitative pathology scoring on the other hand is sometimes regarded as less accurate due to the limitations of the scoring systems employed and the day-to-day variations often noted between pathologists. We therefore decided to generate an optimized histochemical staining and image analysis protocol to compare the accuracy of semiquantitative scoring with computerized image analysis. In order to achieve this, we describe a standardized protocol for staining and image analysis that eliminates or minimizes as many sources of error as possible. The results of this experiment demonstrate that despite consistent variations in scoring between two independent pathologists, correlation with image analysis data of 0.91 to 0.95 (Spearman's Rho test) was achieved. These data indicate that either image analysis or traditional semiquantitative scoring can generate accurate data. As a result of this, it appears that it is equally safe to employ either method dependent upon the complexity and the practicality of the task at hand provided that the experimental conditions are rigorously optimized and rigidly adhered to.