A Deep-Learning Solution Identifies HER2 Negative Cases and Provides ER and PR Results From H&E-Stained Breast Cancer Specimens: A Blind Validation Study
Elizabeth Walsh, Salim Arslan, André Geraldes, Rebecca A. Millican-Slater, Andrew M. Hanby, Nicholas Bennett, Bejal Mistry, Julian Schmidt, Steffen Wolf, Cher Bass, Foivos Ntelemis, Narender Kumar, Pahini Pandya, Nicolas Michel Orsi · Clinical Breast Cancer · 2025
Background PANProfiler Breast is a UKCA-marked, deep-learning image analysis tool. It provides oestrogen and progesterone receptor (ER/PR) status and identifies human epidermal growth factor receptor-2 (HER2) negativity from whole slide images (WSIs) of haematoxylin and eosin (H&E)-stained breast cancer (BC) tissue. This study blindly validated PANProfiler's prediction of ER/PR status and identification of HER2 negative status. Materials and Methods Three cohorts of WSIs of H&E-stained BC specimens were used for calibration (200 cases, 344 WSIs) and blind validation (200 cases, 348 WSIs). For the blind validation, PANProfiler analysed WSIs to provide results for ER, PR (‘Positive', ‘Negative', or ‘Indeterminate') and HER2 (‘Negative' or ‘Indeterminate'). These were compared to the corresponding pathology reports. To discern PANProfiler's performance, concordance and other metrics were calculated, including test replacement rate (TRR) (cases PANProfiler produced a definitive result for) and complete test replacement rate (CTTR) (cases with definitive results for all markers). Results Following blind validation, concordance for ER and PR status across the cohorts was 89·74-93% and 86-91%, respectively. The TRR for ER was 70-84% and 55-84% for PR. For HER2 negative cases, concordance across cohorts was 91-100%, with a TRR ranging from 22 to 27%. CTTRs for the cohorts were between 18% and 20%. Conclusion PANProfiler Breast showed high concordance for ER and PR status and identified HER2 negativity from WSIs of H&E-stained BCs. For HER2 negativity, whilst the TRR was lower than that of ER and PR, the high level of concordance indicated its reliability in identifying negative cases. MicroAbstract Breast cancers undergo oestrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor-2 (HER2) testing, which adds time and cost to the diagnostic process. PANProfiler Breast is a deep-learning image analysis tool that provides results for these markers from breast cancer whole slide images (WSIs). A blind validation was carried out on this platform. It showed high concordance for ER/PR status and identified HER2 negative cases.