Abstract 76: IHCExplore: An AI-driven computational pathology platform for accurate and scalable immunohistochemistry scoring.

Abstract Precise scoring of biomarker expression using immunohistochemistry (IHC) influences the success of precision therapeutics in oncology, such as antibody-drug conjugates (ADCs). However, manual IHC scoring by pathologists is limited by inter- and intra-pathologist variability, poor scalability, and limitations in visually accessible (e.g., spatial) information. IHCExplore* is an artificial intelligence (AI)-enabled tool that provides comprehensive and reproducible characterization of protein expression from an IHC specimen. IHCExplore leverages PLUTO, a pathology foundation model,1 and was trained on a dataset of 24,390 whole slide images (WSIs) from 123 distinct staining assays. The model identifies cell types (cancer cells and lymphocytes) and segments tissue regions into cancer epithelium, cancer-associated stroma, and necrosis. It further segments the nucleus, cytoplasm, and membrane of individual cells to quantify protein expression at subcellular resolution. Model-derived features quantify expression at the slide-level and classify cancer cells into unstained, low-, medium-, and high-intensity categories. Spatial features describe the distribution of these cells within the tumor microenvironment. Additional quantitative readouts include continuous H-score, membrane: cytoplasm intensity ratio, expression heterogeneity score, and spatial proximity score (to assess bystander activity).As a proof of concept, IHCExplore features were used to calculate PD-L1 tumor proportion score (TPSIHCExplore) as follows: total cancer cells with membrane staining above a threshold/total cancer cells. In a cohort of non-small cell lung cancer WSIs (N=597) stained with multiple PD-L1 clones (SP263, SP142, 28-8, 22C3), TPSIHCExplore was compared to manual consensus TPS from N=5 pathologists using intraclass correlation coefficient (ICC). After adjusting the staining threshold from a preset cutoff (4) to a calibrated cutoff (7), the ICC of TPSIHCExplore compared to consensus increased from 0.73 to 0.91, a value on par with the ICC of the average annotator compared to consensus (0.90). Therefore, outputs from routine IHCExplore deployments can recapitulate existing biomarker scoring with simple threshold optimization.Thus, IHCExplore is a scalable and accurate AI-driven solution for IHC scoring. The model’s ability to generalize across tumor types and staining modalities, along with its fine-grained cell compartment segmentation, positions it as a powerful tool for biomarker discovery, assay optimization, and potential deployment as a companion diagnostic or scoring-assist system for precision oncology. *IHCExplore is For Research Use Only. Not for use in diagnostic procedures. 1 Juyal, D. et al. (2024) PLUTO: Pathology Universal Transformer. arXiv:2405.07905 Citation Format: Kelsey Luu, Ciyue Shen, Chintan Shah, John Shamshoian, Blake Martin, Daniel Shenker, Nhat Jackson nyman, Le, Zahil Shanis, Syed Ashar Javed, Matthew Bronnimann, Robert Egger, Harsha Pokolla, Andrew H. Beck, Benjamin Glass, Jennifer Hipp, Ryan Leung, Jacqueline Brosnan-Cashman, Santhosh Balasubramanian, Bahar Rahsepar, Emma Krause. IHCExplore: An AI-driven computational pathology platform for accurate and scalable immunohistochemistry scoring abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 76.