Abstract—Human Epidermal Growth factor receptor 2 (HER2)-positive breast cancer is the most aggressive subtype, requiring targeted treatments for its effective management. Immunohistochemistry (IHC) is the gold standard for manual HER2 scoring that requires specialized antibodies and advanced lab equipment for accurate evaluation. Manual inspection of the IHC images requires a high level medical expertise, while having the downside of being tedious and extremely time consuming. Hematoxylin and Eosin (H&E) staining is the routine procedure for breast cancer detection making it more widely available and cost effective than IHC images. Deep learning-based methods can provide fast, reliable, and cost-efficient automated tools for HER2 scoring using histopathological images. ConvNeXt is a purely convolutional neural network that is based on ResNet and incorporates several advanced techniques inspired by vision transformers to enhance its performance. In this work, ConvNeXt is compared to three standard networks: InceptionV3, ResNet50, and MobileNetV2. The Breast Cancer Immunohistochemical (BCI) public dataset consisting of over three thousand IHC and H&E image pairs was used to evaluate the pretrained networks’ performance. For both IHC and H&E images, ConvNeXt and ResNet50 achieved the highest accuracies. For 4-class HER2 classification, ConvNeXt attained accuracies of 97.79% and 95.58% for IHC and H&E images, respectively. These results outperform state-of-the-art methods from literature by up to 10%. Both IHC and H&E stained images are shown to be reliable for HER2-scoring using deep learning-based approaches. H&E stained images, given their low cost and widespread availability, thus represent strong candidates for integration into practical AI-assisted HER2 scoring systems.
 
Keywords—breast cancer, Human Epidermal Growth factor receptor 2 (HER2) scoring, histopathological images, classification, Transfer Learning (TL), ConvNeXt


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).