Artificial Intelligence-Driven Quantitative HER2 Scoring and Spatial Bystander Effect Modeling for ADC Response Stratification in HER2-Low Breast Cancer.

Abstract

The rapid development of antibody-drug conjugates (ADCs), particularly trastuzumab deruxtecan (T-DXd), has renewed interest in more refined assessment of low-end HER2 expression in breast cancer. However, substantial inter-observer variability in manual immunohistochemistry, especially around the IHC 0 versus 1+ boundary, remains a major challenge for consistent patient stratification. In this narrative review, we summarize emerging advances in computational pathology, including weakly supervised whole-slide image modeling and quantitative HER2 scoring approaches, and discuss their potential to improve interpretive consistency in this threshold-adjacent setting. We further examine the spatial rationale of the ADC bystander effect as a conceptual basis for integrating quantitative HER2 burden, heterogeneity, and spatial organization into response assessment. Rather than presenting these approaches as clinically established predictive tools, we argue that they should currently be viewed as exploratory and hypothesis-generating frameworks that warrant rigorous validation in outcome-linked, cross-platform, and multi-center studies. Overall, AI-assisted quantitative and spatial pathology may help refine biomarker development for ADC therapy in HER2-low breast cancer, but its clinical utility remains to be established through prospective and analytically robust validation.