Deep learning based automated HER2 score prediction using immunohistochemistry histopathological images: a...

Area of Science:

• Computational pathology
• Digital pathology
• Artificial intelligence in oncology

Background:

• HER2 is a key prognostic biomarker in breast cancer, linked to aggressive tumor biology.
• Current HER2 immunohistochemistry (IHC) scoring is subjective and time-consuming.
• Automating HER2 IHC analysis using deep learning is an underexplored area needing external validation.

Purpose of the Study:

• To evaluate the performance of deep learning models for automated HER2 IHC scoring.
• To assess the robustness of these models across different centers and imaging protocols.

Main Methods:

• A dual-center retrospective study analyzed 135 HER2 IHC whole-slide images from 118 breast cancer patients.
• Tumor regions were annotated, patches were created, normalized, and augmented.
• Six pretrained deep learning models were trained and evaluated on an independent test set.

Main Results:

• All models demonstrated good discrimination for three-class HER2 grading (1+, 2+, 3+).
• AlexNet achieved the highest performance with a macro-AUC of 0.971, followed closely by VGG16 (0.967).
• Errors primarily occurred between adjacent HER2 grades, and Grad-CAM highlighted key stained regions.

Conclusions:

• A deep learning framework shows encouraging performance for predicting HER2 IHC scores.
• This approach can assist pathologists by enhancing scoring consistency and identifying cases needing further review.
• The study highlights the potential of AI to improve breast cancer diagnostics.