Reproducibility of Ki67 Haralick entropy as a prognostic marker in estrogen receptor-positive HER2-negative breast cancer.

Abstract

Objective: Intratumoral heterogeneity (ITH) of Ki67 expression reflects the proliferative diversity of breast cancer (BC) cells and has been associated with disease progression. Quantification of Ki67 ITH using Haralick entropy metric from digital image analysis (DIA) has been reported as an independent predictor of breast cancer-specific survival (BCSS); however, its reproducibility across DIA platforms and dependence on tumor tissue sampling have not been investigated.

Methods: Whole-slide images of Ki67-stained tumor sections from 254 patients with ER+/HER2- BC were analyzed independently using HALO and Aiforia DIA platforms. The DIA outputs were subsampled using hexagonal grids to compute Ki67 Haralick entropy. Reproducibility was tested across DIA platforms and under simulated surgical excision and core biopsy scenarios. Lastly, the impact on prognostic modeling for BCSS was assessed.

Results: Haralick entropy demonstrated strong Ki67 ITH cross-platform reproducibility. For prognosis, it provided stronger model performance than conventional Ki67% metrics and independently predicted worse BCSS alongside lymph node involvement. Its prognostic value remained consistent across simulated sampling scenarios.

Conclusions: Ki67 Haralick entropy is a reproducible and robust image-derived ITH metric in ER+/HER2- BC. It demonstrated improved prognostic modeling performance compared to conventional Ki67% across 2 different DIA platforms and sampling conditions, supporting its potential for clinical implementation.