IPM Take
This is where AI becomes more than a workflow tool. The sharper signal is patient stratification: using digital pathology to quantify HER2 expression and map the tumour microenvironment in ways that could support treatment decisions. The promise is real, but so is the governance problem. AI-generated biomarker intelligence needs validation, standards and clinical trust before it can become part of access.
Executive Summary
At ASCO 2026, Lunit presented studies using AI-powered HER2/IHC biomarker quantification and spatial tumour microenvironment analysis. The company highlighted research across HER2-positive advanced biliary tract cancer, NSCLC, adenoid cystic carcinoma, microsatellite-stable metastatic colorectal cancer and advanced gastric cancer. In one biliary tract cancer analysis, AI-based whole-slide HER2 quantification identified a subgroup with a higher response rate to a HER2-targeted regimen.
Why it matters
- Diagnostics / pathology: AI biomarker interpretation needs validation, quality control and integration into routine pathology workflows.
- Clinicians: AI-derived stratification is useful only when it clearly informs treatment choice or trial eligibility.
- Data / AI leaders: The next test is clinical utility, not only algorithmic performance.
AI in cancer pathology is often sold as speed. Faster slide reading. Faster triage. Faster workflow. That is useful, but it is not the whole story.
The more interesting movement is toward biological interpretation. If AI can quantify HER2 expression more precisely, map immune architecture or define spatial tumour patterns, it could help identify which patients are more likely to respond to targeted or immune-based treatment. That moves AI closer to the treatment pathway.
But this also raises the bar. A biomarker used for treatment selection cannot be a black box. It needs reproducibility, external validation, clear thresholds, quality assurance and clinician confidence. Pathologists need to know what the tool is measuring. Oncologists need to know how to act on the output. Payers need to know whether it improves outcomes or simply adds another layer of cost.
For IPM, AI pathology is not just a technology story. It is a readiness story: can health systems turn digital slide intelligence into safe, equitable and reimbursed treatment decisions?
