CELL INTELLIGENCE · 肿瘤细胞治疗研究

PURPOSE: Despite recent advances in anti-human epidermal growth factor receptor 2 (HER2) treatments for HER2-positive biliary tract cancer (BTC), current guidelines lack clear thresholds for defining HER2 positivity in BTC. This study investigated the use of artificial intelligence (AI) to analyze HER2 expression and immune phenotypes (IP) in patients with HER2-positive BTC treated with anti-HER2 therapy. MATERIALS AND METHODS: We conducted a post hoc analysis of a phase II trial (KCSG HB19-14) of trastuzumab plus folinic acid, fluorouracil, and oxaliplatin (FOLFOX) for HER2-positive BTC. AI-powered HER2 quantification and IP analyses were performed on whole-slide images of pretreatment samples. Clinical outcomes were analyzed on the basis of HER2 positivity using a continuous AI-based HER2 immunohistochemistry scoring system. Additionally, we evaluated the spatial distribution of tumor-infiltrating lymphocytes using AI-based IP analysis. RESULTS: Among 29 patients, the overall concordance rate between pathologists and the HER2-AI analyzer was 79.1%. AI-defined HER2-positivity status, characterized by a ≥30% H3 tumor cell proportion threshold, significantly predicted improved outcomes with trastuzumab plus FOLFOX (progression-free survival: 6.7 v 4.9 months, P = .039; overall survival: not reached v 8.4 months, P = .018). By contrast, traditional pathologist-based scoring did not stratify outcomes. AI-powered immune profiling revealed that HER2 3+ tumors predominantly exhibited immune-desert phenotypes, whereas HER2 2+ tumors displayed more inflamed phenotypes, potentially limiting the efficacy of current immunotherapy regimens for HER2 3+ BTC. CONCLUSION: AI-powered HER2 quantification provides a refined biomarker for predicting the response to HER2-targeted therapies in BTC, proposing a ≥30% HER2 3+ tumor cell proportion threshold. Our findings highlight the potential of combining anti-HER2 therapy with immune checkpoint inhibitors on the basis of IP profiles.