Abstract
BACKGROUND: HER2-targeted therapies have improved survival in HER2-positive breast cancer, and recent data suggest potential benefits for patients with HER2-low tumors (defined as immunohistochemistry (IHC) 1 + or 2 + and, in situ hybridization (ISH)-negative). HER2-low tumors are heterogenous, spanning the hormone receptor-positive and triple-negative subtypes. Assessing HER2-low and HER2-ultralow status remains challenging, especially across specimen types. AIMS: This study aims to (1) compare HER2 assessment using conventional microscopy, digital pathology, and an artificial intelligence (AI) model, and (2) investigate changes in HER2-low status between core biopsies, surgical specimens, and metastases. MATERIALS AND METHODS: IHC slides from 47 HER2-low advanced breast carcinomas were analyzed using conventional microscopy, digital pathology, and an AI model developed on Aiforia® Create. HER2 statuses were categorized as low, ultralow (score 1 + in 1-10%), and null (score 0 or 1 + in < 1% with difficult-to-interpret minimal membranous-like staining). Changes in HER2 expression across specimen types were evaluated using agreement measures and visualization tools. RESULTS: The AI model identified more HER2-low and HER2-ultralow cases compared to conventional methods, improving detection accuracy. HER2 expression differed between specimen types, with metastases exhibiting increased HER2 expression compared to surgical specimens and core biopsies. Digital pathology also showed stronger membranous staining and identified more HER2 expressor tumor cells with any kind of membranous staining than microscopy. CONCLUSIONS: AI evaluation is a more sensitive method for HER2-low assessment and reveals expression changes across disease progression. These findings emphasize the need for standardized HER2 assessment to ensure accurate therapy eligibility, particularly for novel treatments like Trastuzumab-Deruxtecan.