Abstract Background: Ki67 is a well-established proliferation marker in breast cancer. Current clinical use focuses on the proportion of Ki67-positive cells, ignoring spatial heterogeneity in expression. However intra-tumoral heterogeneity has demonstrated to be associated with worse outcome. We hypothesized that spatial Ki67 heterogeneity carries clinical information beyond conventional scoring and aimed to evaluate its added value for predicting pathological complete response (pCR) after neoadjuvant chemotherapy (NACT) and for stratifying recurrence risk using genomic expression profiling (GEP). Using digital image analysis (DIA), precise and spatial quantification of biomarker distribution is possible. Methods: We analyzed two retrospective breast cancer cohorts using an AI-assisted DIA pipeline. Tumor sections stained for ER, PR, Ki67, and HER2 were digitized and analyzed in QuPath. Using AI, individual tumor cells were recognized and four tumor regions (0.5mm x 0.5mm) with the highest tumor/stroma ratio were selected for analysis. Spatial Ki67 heterogeneity was quantified using the Morisita-Horn Index (MHI) after the Ki67-positive and Ki67-negative tumor cells were mapped using XY-coordinates and square tessellation (100×100 µm tiles) was applied. The MHI was used to compare the similarity in cell composition between all pairs of tiles within a region. MHI values range from 0 to 1, with higher values indicating a more uneven distribution of Ki67+ cells. We used logistic regression and model comparison with Akaike Information Criterion (AIC), likelihood ratio test (LRT) or Vuong test, to evaluate the predictive value of Ki67 heterogeneity. In the first cohort (n=45), spatial heterogeneity was assessed on pretreatment biopsies from patients treated with NACT. In the second cohort (n=79), heterogeneity was evaluated in HR+/HER2- breast cancer patients stratified as high or low risk of recurrence based on GEP. Results: In the GEP cohort, both a higher proportion of Ki67- positive cells and greater Ki67 heterogeneity were significantly associated with high genomic risk. The median MHI was 0.24 (0.03–0.35) in the high-risk group compared to 0.14 (0.01–0.43) in the low-risk group (P = 0.008). This higher MHI indicates more heterogeneous regionally clustered Ki67 expression, suggesting biologically distinct proliferative zones. In multivariate models, Ki67 heterogeneity remained a significant predictor of high-risk classification (OR 0.22, P = 0.036). Furthermore, in nested model comparison using LRT, addition of Ki67 heterogeneity significantly improved the model for predicting genomic risk (P = 0.034). These findings were consistent across biopsy and resection specimens, highlighting the robustness of heterogeneity measures. In the NACT cohort, Ki67 heterogeneity was higher in patients who achieved pCR (median MHI 0.26 0.17–0.35) compared to those who did not (median MHI 0.22 0.12–0.40, P = 0.023). In multivariate modeling, Ki67 heterogeneity emerged as an independent predictor of pCR (OR 23.5, P = 0.038), outperforming Ki67 density and improving model fit (AIC 31.6 vs. 36.1; P = 0.038). Finally, in both cohorts, DIA-derived Ki67 models slightly outperformed traditional pathologist scoring, although Vuong tests did not show a statistically significant difference. Conclusion: Spatial Ki67 heterogeneity provides additional prognostic and predictive value beyond conventional Ki67 scoring. This heterogeneity indicates distinct areas of higher proliferation, clinically relevant biological variation, not captured by simple percentage positivity. Although validation in larger, prospective cohorts is necessary before clinical implementation, DIA provides a more objective and reproducible alternative to manual scoring, particularly when incorporating spatial heterogeneity. Citation Format: C. Van Berckelaer, K. Zwaenepoel, L. Cox, D. Charlotte, D. Julie, E. Louise, H. Fleur, L. Evy, G. R. Devi, A. Ramadhan, S. Koljenovic, P. Van Dam. Ki67 Spatial Heterogeneity as a Predictive and Prognostic Marker in Breast Cancer: A Spatial Image Analysis Approach abstract. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS2-08-19.
Berckelaer et al. (Tue,) studied this question.