Comparative study of different software in Ki67 assessment of breast cancer

Abstract

Abstract Aims Ki67 proliferation index is an important indicator for molecular classification, treatment monitoring and prognosis prediction of breast cancer, which is an integral part of the entire process of breast cancer diagnosis and treatment. In recent years, the accuracy of artificial intelligence (AI) assessment of Ki67 has been significantly improved, and numerous software options have become available. However, it is important to note that some software is not open source, and the issue of poor consistency between different laboratories remains unsolved. Therefore, further research is necessary to advance AI-assisted Ki67 interpretation. Methods This study aimed to provide a reference for clinicians regarding the more suitable interpretation method of Ki67 through the analysis and consistency assessment of results from two selected regions (hotspot and average) using four free pathological image analysis software (Qupath, ClinicaPath.AIM 1.0, Path920Ki67.A1.0 semi-automatic, and Path920Ki67.A2.0 automatic). Additionally, the study aimed to establish a theoretical basis for the precise treatment of breast cancer. To facilitate statistical analysis, 40 cases were categorized into two groups based on Ki67 distribution (evenly and unevenly distributed groups). Furthermore, the cases were divided into low expression (Ki67 ≤ 30%) and high expression (Ki67 > 30%) groups based on the Ki67 proliferation index. Results The four software programs analyzed in this study exhibited consistent statistical results and shared characteristics in interpreting immunohistochemical results for Ki67 in breast cancer. Each software demonstrated good consistency, regardless of whether the Ki67 distribution was even or uneven. Notably, the intraclass correlation coefficient (ICC) value of the Qupath software was similar in both groups and consistently above 0.95. In terms of Ki67 expression, the software performed better in the high expression group compared to the low expression group. When using the overall region averaging method, the ICC values for the high expression group ranked as follows: ClinicaPath.AIM 1.0, Qupath, Path920Ki67.A1.0, and Path920Ki67.A2.0. For the low expression group, the ICC values ranked as follows: Qupath, Path920Ki67.A2.0, ClinicaPath.AIM 1.0, and Path920Ki67.A1.0. When employing the hotspot area method, the ICC values for the high expression group ranked as follows: Qupath, Path920Ki67.A1.0, Path920Ki67.A2.0, and ClinicaPath.AIM 1.0. For the low expression group, the ICC values ranked as follows: Path920Ki67.A1.0, ClinicaPath.AIM 1.0, Path920Ki67.A2.0, and Qupath. Conclusion Regardless of the overall average region method or the hotspot region method, all four software exhibited consistent breast cancer Ki67 interpretation results. Notably, the Path920Ki67.A2.0 software, developed independently by our laboratory, demonstrated a high average ICC value of above 0.8 in the Ki67 low expression group, further affirming its consistency.