A New Era in Diagnosing HER2-Low Breast Cancer with Artificial Intelligence

In breast cancer, the level of HER2 protein expressionplays a critical role in determining which targeted treatments patients may benefit from. Many patients previously classified as “HER2-negative” actually had low or ultralow HER2 expression, potentially making them eligible for recently developed antibody–drug conjugates such as Enhertu . However, conventional pathology methods may not always distinguish these differences.

At the 2025 ASCO Annual Meeting , abstract 1014 presented a new international multicenter study showing that artificial intelligence (AI) support substantially improves the accuracy of HER2 classification. AI makes a particularly important difference in correctly identifying HER2-low and HER2-ultralow categories.

Study Overview: A New Era in Diagnosis

The study involved 105 pathologists from 10 countries and was conducted through the digital education platform ComPath Academy . Participants assessed HER2 immunohistochemistry (IHC) scores according to the 2023 ASCO/CAP guideline and evaluated 20 digital breast cancer cases using additional categories: “HER2-ultralow” (IHC 0 with membrane staining) and “HER2-null” (IHC 0 without membrane staining).

The AI System: ComPath Academy and Mindpeak Infrastructure

The AI infrastructure used in this study was developed by Germany-based Mindpeak GmbH and integrated into ComPath Academy , a platform specifically designed for digital pathology education. It provides case-based decision support to pathologists and aimed to improve HER2 IHC scoring accuracy in breast cancer.

Although the algorithm's technical architecture (such as CNN or Transformer) was not specified directly, the study's structure suggests the following technological features:

• An image-classification-based machine-learning algorithm was used.

• The model was trained to perform tumor-cell-level analysis and detect membrane staining in individual cells.

• It particularly focused on distinguishing borderline categories such as IHC 0, HER2-low and HER2-ultralow .

• The AI outputs provided explainability , probably involving visualization layers, such as heatmaps or attention maps, integrated into the decision-support system.

Mindpeak GmbH is a leading developer of AI applications for digital pathology. It produces deep-learning solutions, particularly for breast, lung, and prostate cancers, that automate IHC scoring . These algorithms are mostly based on convolutional neural networks (CNNs) .

In conclusion, the AI system used here is CNN-based image-analysis software specifically trained for HER2 immunohistochemistry (IHC) assessment. Integrated into the digital education platform, it improved both diagnostic accuracy and agreement between pathologists.

🔍 What Changed with AI Support?

• In the first two assessments, without AI assistance , accuracy was: 89.1%

• In the third assessment, with AI assistance , accuracy was: 96.1%

• Interpathologist agreement coefficient (concordance):

  • Without AI: 0.506

  • With AI: 0.798

• Correct classification by clinical category:

  • Without AI: 90.1%

  • With AI: 95.0%

• Misclassification of HER2-low or HER2-ultralow cases as HER2-null:

  • Decreased by 24.4% with AI

These figures show that AI not only improves diagnostic quality but also substantially reduces diagnostic errors that determine access to treatment.

Why Is This So Important?

Breast cancers with low HER2 expression open the door to promising new treatments for thousands of patients who previously lacked a treatment opportunity. However, this potential can only be realized through accurate diagnosis.

Approximately 65% of tumors described as HER2-negative actually show some HER2 expression and are now classified as HER2-low or HER2-ultralow. These patients may benefit from HER2-targeted drugs, but correct diagnosis is essential. For the first time, multicenter findings from this study demonstrate AI's potential to close this diagnostic gap.

Commentary

Accurate HER2 scoring helps patients receive the best treatment. This study shows that an AI-supported approach improves diagnostic accuracy and provides benefits in situations directly affecting treatment decisions. AI is not here to replace physicians; it helps us deliver high-quality, personalized care by working more intelligently and efficiently.

Conclusion: More Hope for Women Through AI

This study again demonstrates how complex HER2 classification is in breast cancer and how critical it is to treatment decisions. AI is becoming a support tool that can save lives as well as time. Like any new technology, it offers concrete clinical benefits when implemented correctly.

The future of breast cancer is evolving into a world where data, as well as cells, come under the microscope. AI-supported diagnostic systems may allow more patients to receive the right treatment at the right time.