Comparison of FDA-Approved vs. Mayo Clinic’s Ki-67 Immunostaining Protocols for Breast Cancer Specimens: A Quantitative Analysis Using Artificial Intelligence (AI) Based Image Analysis Algorithms

Abstract

Abstract Introduction/Objective Ki-67, a nuclear protein present during the active phases of the cell cycle, is an important prognostic and predictive marker for many tumor types, including breast cancer. This study compared the results of breast cancer specimens immunohistochemically (IHC) stained with the FDA-approved vs. Mayo Clinic’s (MC) validated Ki-67 protocol using AI-based image analysis algorithms for quantitative examination at an individual cell level. Methods/Case Report Sixty-two breast cancer specimens were selected with a rough equal representation of Ki-67 percentage in low (<10%), intermediate (10-20%) and high (>20%) categories. Two adjacent sections of each specimen were stained with the FDA and MC protocols, respectively, and scanned to obtain whole slide digital images. Two AI algorithms, one with tumor cell identification and the other without, were used to quantify positive and negative cells and calculate Ki-67 percentages. Results (if a Case Study enter NA) Both algorithms show high concordance between the two staining protocols, with r-squared scores of 0.989 and 0.992, respectively, analyzing with or without tumor identification. AI identified tumor better on MC stained slides vs. FDA stained slides. The results demonstrate a high-performance concordance between the MC and the FDA IHC staining protocols. Conclusion The data also illustrates the accuracy of using AI image analysis algorithms to compare different IHC protocols at an individual cell level. The AI tumor identification difference in slides stained with two IHC protocols reflects the importance of training an algorithm with a specific staining protocol to achieve optimal tumor recognition.