HER2 Scoring

The ASCO/CAP guidelines define four discrete scores based on membrane staining characteristics observed under light microscopy:

• Score 0 (Negative): No staining observed, or incomplete, faint membrane staining in ≤10% of invasive tumor cells.
• Score 1+ (Negative): Faint, barely perceptible incomplete membrane staining in >10% of tumor cells.
• Score 2+ (Equivocal): Weak to moderate complete membrane staining in >10% of tumor cells, or strong complete staining in ≤10%. This result requires reflex testing via in situ hybridization (ISH).
• Score 3+ (Positive): Strong, complete, circumferential membrane staining in >10% of invasive tumor cells. Patients are eligible for anti-HER2 targeted therapy.

A Score 2+ (Equivocal) finding triggers mandatory reflex testing using fluorescence or chromogenic in situ hybridization (FISH/CISH) to assess HER2/neu gene amplification at the DNA level. ISH testing evaluates:

• HER2/CEP17 Ratio: The ratio of HER2 gene signals to chromosome 17 centromere signals. A ratio ≥2.0 indicates amplification.
• Average HER2 Copy Number: The mean number of HER2 signals per nucleus. ≥6.0 signals/cell is considered amplified.
• Dual-Probe ISH: The standard approach uses differentially labeled probes for HER2 and CEP17, allowing simultaneous visualization and accurate ratio calculation. Computational pathology models can also predict ISH amplification status directly from H&E morphology, potentially reducing the need for reflex testing.

HER2 expression is often spatially heterogeneous within a tumor, posing a significant challenge for manual and computational scoring. Key considerations include:

• Regional Variability: Some tumor areas may show strong 3+ staining while adjacent regions are 1+ or negative. The ASCO/CAP guidelines require scoring the area with the highest intensity.
• Basolateral vs. Circumferential Staining: True 3+ staining must be complete and circumferential. Basolateral (U-shaped) patterns are characteristic of score 2+.
• Crush Artifact and Edge Effects: Tissue processing artifacts can create false-positive membrane staining at tissue edges or crushed areas. Deep learning models trained with artifact detection modules can exclude these regions from analysis.
• Microscopic vs. Macroscopic Heterogeneity: Computational models can quantify heterogeneity at both the cell-level and regional-level, providing an H-score or heterogeneity index that may have prognostic significance beyond the binary classification.

Accurate HER2 scoring is the gateway to life-prolonging targeted therapies. The clinical cascade includes:

• Trastuzumab (Herceptin): A monoclonal antibody binding to the HER2 extracellular domain, inducing antibody-dependent cellular cytotoxicity.
• Pertuzumab: Binds a different HER2 epitope, preventing dimerization with other HER family receptors. Used in combination with trastuzumab.
• Antibody-Drug Conjugates (ADCs): Agents like trastuzumab deruxtecan (T-DXd) and trastuzumab emtansine (T-DM1) deliver cytotoxic payloads specifically to HER2-expressing cells. Notably, T-DXd has shown efficacy even in HER2-low (IHC 1+ or 2+/ISH-negative) tumors, expanding the therapeutic relevance of precise, quantitative scoring.
• Tyrosine Kinase Inhibitors: Small molecules like lapatinib and tucatinib that inhibit the intracellular kinase domain.

Manual HER2 scoring suffers from significant inter-observer variability, particularly at the 1+/2+ boundary. Studies report concordance rates as low as 70% between pathologists for equivocal cases. Computational standardization addresses this through:

• Continuous Scoring: AI models output a continuous membrane completeness and intensity score rather than a discrete 0-3+ bin, enabling more nuanced clinical decision-making.
• Heatmap Visualization: Quantitative spatial maps overlay the slide image, showing exactly which regions contributed to the score, providing built-in explainability for pathologist review.
• Training Set Diversity: Robust models are trained on multi-institutional data with stain normalization to generalize across different scanners, staining protocols, and tissue preparation methods.
• Proficiency Testing: Automated scoring systems can serve as a reference standard in external quality assessment (EQA) programs, benchmarking individual pathologist performance against a consistent computational gold standard.

The foundational laboratory method for HER2 scoring. IHC uses enzyme-linked antibodies to visually detect the HER2 protein on the cell membrane.

• Scoring Scale: Pathologists assign scores of 0, 1+, 2+, or 3+ based on staining intensity and completeness of membrane staining.
• Equivocal Range: A score of 2+ is considered equivocal and typically requires reflex testing with In Situ Hybridization (ISH).
• Computational Automation: Deep learning models now automate the quantification of membrane staining intensity to reduce inter-observer variability.

A reflex molecular test used to resolve equivocal IHC 2+ results by directly counting HER2 gene copies.

• Mechanism: Fluorescent probes bind to the HER2 gene and chromosome 17 centromere (CEP17) to calculate a HER2/CEP17 ratio.
• Amplification Threshold: A ratio ≥ 2.0 or an average HER2 copy number ≥ 6.0 signals is considered positive.
• AI Integration: Computational pathology models can predict FISH amplification status directly from H&E morphology, potentially bypassing the need for the assay.

A newly defined therapeutic category expanding the binary view of HER2 status to include patients previously classified as negative.

• Definition: Tumors with IHC scores of 1+ or 2+ with negative ISH results.
• Therapeutic Relevance: This classification identifies patients eligible for Enhertu (trastuzumab deruxtecan), an antibody-drug conjugate.
• Computational Challenge: Distinguishing between IHC 0 (negative) and IHC 1+ (low) is notoriously difficult for human pathologists, representing a high-value target for precise AI quantification.

The significant discordance rate in manual HER2 scoring that drives the clinical need for automated, standardized AI systems.

• The Problem: Studies show only moderate agreement between pathologists when scoring borderline IHC 2+ cases, leading to misclassification and inappropriate therapy.
• Root Causes: Variability arises from subjective interpretation of staining intensity, heterogeneous tumor expression, and differences in training.
• AI Solution: Deep learning models provide a continuous, quantitative HER2 membrane completeness score, acting as a consistent second reader to reduce diagnostic drift.