AI Validated in Landmark Study to Refine Breast Cancer Scoring

WALTHAM, MA – February 03, 2026 – A landmark study published in the peer-reviewed journal Modern Pathology has provided powerful independent validation for the use of artificial intelligence in breast cancer diagnostics, a development poised to enhance treatment precision for thousands of patients. The blinded, multi-vendor study confirmed the clinical-grade capabilities of BostonGene’s AI platform in assessing the Human Epidermal Growth Factor Receptor 2 (HER2) biomarker, a critical step toward standardizing a process historically plagued by subjectivity.

The initiative, a collaboration with Friends of Cancer Research and supported by a consortium of pharmaceutical giants including AstraZeneca, Bristol Myers Squibb, and Merck, evaluated 10 different AI models. By benchmarking these technologies under a rigorous, independent framework, the findings establish a new level of confidence in AI-driven biomarkers and signal a significant shift in how advanced analytical tools are vetted for clinical and regulatory use.

Setting a New Standard for Diagnostic Trust

The study, titled “Agreement Across 10 Artificial Intelligence Models in Assessing Human Epidermal Growth Factor Receptor 2 (HER2) Expression in Breast Cancer Whole-Slide Images,” analyzed over 1,100 whole-slide images from more than 700 breast cancer patients. Its core purpose was to measure the agreement between different AI algorithms and compare them to the assessments of expert pathologists.

While the results showed moderate overall agreement across all 10 computational models, particularly in the nuanced HER2-low categories, the consensus was exceptionally high when identifying HER2-positive tumors. The models achieved an agreement of 97.3% in distinguishing these clear-cut cases, a critical task for guiding targeted therapy. This performance underscores AI's potential to automate and standardize the most definitive aspects of HER2 scoring.

For BostonGene, a developer of a leading AI foundation model for cancer biology, the results represent a key external validation. "This is not an isolated result,” said Nathan Fowler, MD, Chief Medical Officer at BostonGene, in a statement. “We continue to see independent, external validation of the AI and ML algorithms that power our foundation model. These blinded, real-world evaluations provide the high-stakes certainty that drug developers trust when accelerating life-saving therapies.”

The study’s design—blinded, collaborative, and involving multiple vendors—is crucial. It mirrors the stringent benchmarks used by drug developers and regulators, de-risking the path for integrating AI into companion diagnostic development and clinical trial execution. This ecosystem-level evaluation places BostonGene among a select group of companies operating at the highest scientific and technical thresholds for clinical-grade AI.

The Challenge of the HER2 Spectrum

The validation comes at a pivotal moment in oncology. HER2 status has long been one of the most important biomarkers in breast cancer, with HER2-positive disease, found in about 15% of cases, being treatable with targeted therapies. However, accurately scoring HER2 expression using traditional immunohistochemistry (IHC) on glass slides is a notoriously subjective process.

Pathologists visually assess the percentage of tumor cells and the intensity of staining, a method that can lead to significant inter-observer variability. Studies have shown that pathologists may disagree on a diagnosis in up to a third of cases, creating uncertainty for treatment decisions. This challenge has become even more acute with the advent of a new class of drugs called antibody-drug conjugates (ADCs). These therapies have proven effective not only in HER2-positive cancer but also in tumors with low levels of HER2 expression—a category now known as "HER2-low."

The ability to precisely identify HER2-low patients has opened up new therapeutic avenues, but it has also placed immense pressure on diagnostic accuracy. A tumor misclassified as HER2-negative (or HER2 0) could mean a patient misses out on a potentially life-saving treatment. The Modern Pathology study highlighted this very issue, noting that the lowest concordance among the AI models occurred in distinguishing between HER2 0, 1+, and 2+ scores—the exact spectrum where clinical decisions are becoming more complex. AI-driven image analysis promises a solution by offering an objective, quantifiable, and reproducible method for scoring HER2, removing human subjectivity and improving consistency, especially in these challenging borderline cases.

Navigating a Competitive and Regulated Landscape

BostonGene is a key player in a rapidly advancing and competitive field. Companies like PathAI, Paige.AI, and Ibex Medical Analytics are also developing sophisticated AI solutions to tackle diagnostic challenges in oncology. Paige, for instance, gained the first FDA approval for an AI pathology tool and is developing a novel method to predict HER2 status from standard H&E slides. PathAI is collaborating directly with pharmaceutical firms to create its own automated HER2 scoring tools.

This competitive innovation is occurring within an evolving regulatory environment. Both the U.S. Food and Drug Administration (FDA) and the European Medicines Agency (EMA) are actively establishing frameworks to govern AI and machine learning-based medical devices. These agencies are focused on ensuring the safety, efficacy, and transparency of AI tools through a "Total Product Lifecycle" approach that demands robust validation and real-world evidence.

The collaborative, multi-vendor nature of the HER2 study provides precisely the kind of data regulators seek. It demonstrates a commitment to transparency and scientific rigor, building a foundation of trust necessary for broader clinical adoption. By successfully participating in such a high-profile benchmarking initiative, companies like BostonGene not only validate their technology but also demonstrate their readiness to meet the stringent requirements for future regulatory submissions.

Fueling the Future of Drug Development

The involvement of pharmaceutical heavyweights in the study is telling. For drug developers, the "de-risking" of clinical trials is a primary strategic objective. The success of a new cancer therapy often depends on accurately identifying the patient population that will benefit most, and that requires reliable biomarkers. Inconsistent or subjective biomarker analysis can lead to failed trials, wasted investment, and delayed access to new medicines.

Validated AI tools offer a powerful solution. By providing consistent and accurate patient stratification, AI can help pharmaceutical companies design smarter, more efficient clinical trials. This is why companies like AstraZeneca, Amgen, and GlaxoSmithKline supported the initiative—they are not just observers but active participants in building the infrastructure for a new era of precision medicine.

BostonGene’s approach extends beyond single-biomarker analysis. The company’s platform is built on a foundation model that integrates multiomic data—including genomics, transcriptomics, and spatial biology—with clinical information. This multidimensional approach allows for a deeper characterization of the tumor and its immune microenvironment, providing insights that can inform decisions across the entire drug development lifecycle, from early-stage target discovery to optimizing clinical trial design. This holistic strategy positions the technology not merely as a diagnostic aid but as a comprehensive engine for advancing cancer research and therapy development. The validation of its HER2 scoring algorithm is a testament to the precision of its underlying model, strengthening industry confidence in its ability to deliver clinically meaningful AI.