PS2-05-26: Analytical performance of an image analysis algorithm developed as an adjunctive aid for determining HER2 gene status from in situ hybridization

Abstract Background: Accurate detection of HER2 gene amplification via in situ hybridization (ISH) is critical for guiding targeted therapy use in patients with breast cancer. Traditionally, this involves manual quantification of the HER2-to-chromosome 17 (HER2/CHR17) signal ratio; however, this process is often time-consuming and may be complicated by factors such as chromosome 17 polysomy or intratumoral heterogeneity. The uPath HER2 Dual ISH image analysis algorithm is a tool for research use only, developed to assist medical laboratory scientists with determining HER2 gene status from images of formalin-fixed, paraffin-embedded (FFPE) neoplastic breast tissue. The algorithm was previously validated with samples stained using the VENTANA BenchMark ULTRA platform, but its performance across alternative staining platforms has not yet been evaluated. Methods: This study was designed to evaluate agreement between the BenchMark ULTRA and XT staining platforms, and between manual and algorithm-assisted methods for scoring across both platforms. In total, 120 advanced breast cancer FFPE samples with varying levels of HER2 amplification were selected for analysis (amplified HER2/CHR17 ratio 2.6, n = 50; non-amplified HER2/CHR17 ratio 1.4, n = 50; borderline amplified HER2/CHR17 ratio ≥ 2.0 and ≤ 2.5, n = 10; borderline non-amplified HER2/CHR17 ratio ≥ 1.5 and ≤ 1.9, n = 10). Tissue sections were prepared and stained using the VENTANA HER2 Dual ISH DNA probe cocktail using both the BenchMark ULTRA and XT platforms. Two trained medical laboratory scientists performed blinded manual quantification of HER2 ISH across all samples. These results were then checked by pathologists (n = 3). Following a washout period, these pathologists repeated the analysis using the algorithm. Results: High inter-platform agreement was observed with the algorithm across 118 evaluable samples, with an overall percent agreement (OPA) of 94.9% (95% confidence interval CI, 89.3-97.6); positive percent agreement (PPA) and negative percent agreement (NPA) were 95.1% (95% CI, 86.5-98.3) and 94.7% (95% CI, 85.6-98.2), respectively. High agreement was observed between the algorithm and manual reads irrespective of the platform used for staining ( Table 1 ). Perfect agreement was observed between manual reads across the two platforms. Conclusions: These data demonstrate that the uPath HER2 Dual ISH image analysis algorithm offers similar performance to manual scoring across samples stained using either the BenchMark ULTRA or XT platforms, and across samples with varying levels of HER2 amplification. With further validation, the algorithm may provide pathologists with a reliable adjunctive tool to support the diagnosis of patients with breast cancer. Citation Format: N. Thampy, K. Young-Zvandasara, C. M. Bacon, C. Chen, N. Harris, M. Howe, A. Long, L. Mansfield, J. Ness, K. A. Wharton Jr, S. Dance, L. J. Inge. Analytical performance of an image analysis algorithm developed as an adjunctive aid for determining HER2 gene status from in situ hybridization 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 PS2-05-26.