# Optimization of Scanning Protocol for AI-Integrated Assessment of HER2 Dual Bright-Field In-Situ Hybridization Application in Breast Cancer

## TL;DR

This study identifies optimal scanning protocols for an AI system that automates HER2 testing in breast cancer, improving accuracy and reliability.

## Contribution

The study introduces optimized scanning protocols for AI-based Dual bright-field in situ hybridization analysis of HER2 in breast cancer.

## Key findings

- Scanning protocols A1, A2, B2, and B3 showed consistent HER2 results compared to manual FISH.
- Protocol C failed due to nuclei detection issues in six cases.
- AI performance was best at 0.12 µm/pixel and 0.17 µm/pixel with extended focus.

## Abstract

Accurately determining HER2 status is essential for breast cancer treatment. We developed an AI-integrated in-house application for automated Dual bright-field (BF) in situ hybridization (ISH) analysis on whole slide images (WSIs), although optimal scanning conditions remain unclear. We evaluated scanners and optimized scanning protocols for clinical application. Ten de-identified invasive breast carcinoma cases, with HER2 immunohistochemistry and FISH results, were analyzed using three scanners and six scanning protocols. WSIs scanned by Scanner ‘A’ have 0.12 µm/pixel with 0.95 NA (A1) and 1.2 NA (A2); Scanner ‘B’ have 0.08 µm/pixel (B1); 0.17 µm/pixel (B2); and 0.17 µm/pixel with extended focus (1.4 µm step size and three layers) (B3); Scanner ‘C’ has 0.26 µm/pixel (C1) resolution. Results showed scanning protocols A1, A2, B2, and B3 yielded HER2 gene amplification status and ASCO/CAP ISH group results consistent with manual FISH as the ground truth. However, protocol C demonstrated poor concordance due to nuclei detection failure in six cases. The AI-integrated application achieved the best performance using scanning protocols with optimized resolutions of 0.12 µm/pixel and 0.17 µm/pixel with extended focus. This study highlights the importance of scanner selection in AI-based HER2 assessment and demonstrates that optimized scanning parameters enhance the accuracy and reliability of automated Dual BF ISH analysis.