MADRID, Spain – The integration of Artificial Intelligence (AI) into clinical workflows has long been promised as a transformative force in medicine. At the 2026 European Society for Medical Oncology (ESMO) congress, held in Madrid from October 23–27, Lunit provided a compelling demonstration of this potential, showcasing how its SCOPE platform is shifting the paradigm of immuno-oncology research.
By presenting data from five distinct studies, Lunit has illustrated that AI is no longer merely a tool for image enhancement; it is a sophisticated instrument capable of quantifying complex biological patterns that were previously invisible or subjective. From non-small cell lung cancer (NSCLC) to pancreatic ductal adenocarcinoma (PDAC), the SCOPE suite is proving its versatility in deciphering the nuances of the tumor microenvironment (TME).
The Core of the Innovation: Understanding Lunit SCOPE
To grasp the significance of these findings, one must first understand the platform. Lunit SCOPE is a comprehensive digital pathology suite designed to bridge the gap between traditional histology and precision medicine. It functions as an analytical layer over standard pathology images, providing objective, reproducible data where human observation might otherwise be hindered by fatigue or inter-observer variability.
The platform is modular, featuring specific tools like:
- Lunit SCOPE PD-L1: A specialized module that automates the analysis of programmed death-ligand 1 (PD-L1) expression, a critical biomarker for immunotherapy eligibility.
- Lunit SCOPE IO: A high-level analysis tool for Hematoxylin and Eosin (H&E)-stained whole slide images (WSIs). It is engineered to evaluate the TME, including the spatial distribution of tumor-infiltrating lymphocytes (TILs), immune phenotypes, and the architectural composition of the tumor stroma.
By applying these modules to diverse clinical datasets, researchers are beginning to map the "hidden" geography of cancer, revealing how subtle cellular spatial arrangements correlate with clinical outcomes.
Chronology: A Multi-Faceted Research Journey
The data presented at ESMO 2026 represent the culmination of intensive multi-center collaborations. While the platform has been in development for several years, the 2026 congress serves as a pivotal checkpoint where the transition from "experimental" to "evidence-based" research is becoming clear.
- Pre-Congress: Lunit initiated a series of large-scale retrospective studies in Japan and elsewhere to stress-test the SCOPE algorithms against "gold-standard" human pathological assessment.
- October 23, 2026: The opening of the ESMO congress marked the official unveiling of these comparative studies, drawing significant interest from oncology researchers and digital health experts alike.
- October 24–27, 2026: The congress provided a platform for deeper technical discussions, focusing on how these AI-derived metrics can be integrated into existing clinical trials to refine patient selection criteria.
Supporting Data: Precision vs. Perception
The data presented in Madrid were highlighted by two particularly striking studies that underscore the discrepancy—and the synergy—between human expertise and machine intelligence.
The NSCLC Concordance Study
In a massive analysis involving 993 patients with surgically resected non-small cell lung cancer, researchers set out to determine how closely AI aligns with pathologists in determining PD-L1 Tumor Proportion Scores (TPS).
The results revealed an overall agreement rate of 73%. While this number may appear moderate at first glance, the devil is in the details. The lowest concordance was observed in cases categorized as "PD-L1 negative" by pathologists. Crucially, when the AI flagged a tumor as "PD-L1 positive" (TPS ≥1%) despite a pathologist’s score of <1%, a staggering 97% of those cases were classified as "low expression" (TPS 1–10%) by the AI.
This suggests that the AI is not necessarily "wrong," but rather more sensitive at the lower end of the detection spectrum. This finding is a goldmine for clinical research, as it identifies a subset of patients who might previously have been excluded from immunotherapy trials due to "negative" PD-L1 status, but who may actually benefit from treatment.
The PDAC Microenvironment Study
In a separate study of 181 patients with pancreatic ductal adenocarcinoma (PDAC), Lunit SCOPE IO was utilized to examine the "tumor-stroma ratio"—the proportion of cancerous cells versus the supportive, fibrous connective tissue surrounding them.
The researchers found that the prognostic value of these ratios is not universal. In the general population, a high tumor-stroma ratio was linked to worse progression-free survival. However, when the data were filtered by molecular subtype—specifically those with KRAS G12D mutations—the inverse was true. Furthermore, high proportions of endothelial cells (the cells lining blood vessels) were found to be beneficial for survival only in tumors with low stromal content.
This complexity highlights the limitation of traditional "one-size-fits-all" pathology. Lunit’s platform has demonstrated that tumor biology is context-dependent; the same biomarker or structural feature can mean two entirely different things depending on the underlying molecular mutation.
Official Responses and Strategic Vision
Brandon Suh, CEO of Lunit, emphasized that the data presented at ESMO 2026 serve as a testament to the maturation of AI in the medical field. "AI is helping researchers uncover new insights from pathology images across different cancer types," Suh stated.
He further elaborated on the company’s forward-looking strategy: "These studies reflect the diverse ways Lunit SCOPE can support cancer research. We will continue to support researchers and expand collaborations that advance our understanding of cancer biology and precision oncology."
For Lunit, the goal is not to replace the pathologist but to augment their capabilities. By handling the tedious, high-dimensional quantification of cell counts and spatial patterns, the SCOPE platform allows clinicians and researchers to focus on higher-level decision-making.
Implications for the Future of Oncology
The findings presented in Madrid carry profound implications for the future of clinical practice and drug development:
1. Refinement of Biomarker Testing
The discrepancy in PD-L1 assessment suggests that current threshold-based diagnostic criteria for immunotherapy may be too blunt. Moving forward, AI could allow for "continuous" rather than "binary" biomarker reporting, potentially leading to more personalized dosing and patient selection.
2. De-risking Clinical Trials
Drug developers often struggle with the "heterogeneity" of patient populations. By using AI to identify specific architectural phenotypes within the tumor—such as those linked to KRAS mutations in the PDAC study—researchers can design more homogenous trial cohorts. This increases the likelihood of detecting a signal of efficacy, thereby reducing the risk of costly phase III failures.
3. A New Understanding of the TME
The interaction between stroma, endothelial cells, and immune infiltrate is one of the most complex areas of current oncology research. The ability of the SCOPE platform to quantify these relationships in routine H&E images—which are already part of every standard diagnosis—means that historical archives of pathology slides can now be "re-mined" for new biological insights without the need for expensive, invasive new biopsies.
4. Bridging the "Black Box"
One of the major criticisms of AI in medicine is the "black box" nature of deep learning. However, by providing quantifiable, spatial data that aligns with known biological features, Lunit is helping to build trust in these tools. The ability to visualize why the AI reached a specific conclusion (e.g., highlighting specific cell clusters in a WSI) is essential for clinical adoption.
Conclusion
The data from ESMO 2026 indicate that we are entering a new era of digital pathology. The Lunit SCOPE platform is no longer a peripheral tool; it is becoming a central component of the research ecosystem. By transforming qualitative images into quantitative data, AI is providing the missing link between cellular-level tumor morphology and patient-level clinical outcomes. As these tools continue to evolve, the ability to tailor cancer treatments to the specific, AI-mapped landscape of an individual’s tumor will likely move from an academic ambition to a clinical standard.
About the Author
