This invention is a computational technology utilizing a machine learning model to accurately predict single-cell types from standard hematoxylin and eosin (H&E) stained digital pathology images. This non-invasive method significantly enhances personalized cancer treatment by providing unparalleled insight into tumor heterogeneity and predicting patient response to therapy without needing expensive, complex assays.
Description
The core of this technology is a machine learning model trained on annotated pathology images, which performs single-cell classification directly from routinely used H&E-stained tissue slides. This process identifies and classifies different cell types within the tumor microenvironment with single-cell resolution. Once the detailed cellular composition is determined, the model integrates these single-cell predictions with clinical data to analyze tumor heterogeneity.
This innovative approach allows the technology to move beyond simple pathology or diagnostic classification. By linking the specific cell-type findings in the tumor microenvironment to therapy outcomes, the technology enables precise predictions of cancer therapy resistance and response, making it a powerful tool for guiding treatment decisions in clinical settings and supporting advanced research into tumor biology. The model's versatility has been demonstrated across different cancer types, including head and neck and liver cancer.
Applications
- Personalized Oncology Treatment: Guiding clinicians in making more precise, individualized treatment decisions based on predicted therapy response and resistance.
- Drug Development and Clinical Trials: Identifying predictive biomarkers and stratifying patients in clinical trials for new cancer therapies.
- Digital Pathology Software: Integration into commercial digital pathology systems to enhance diagnostic and prognostic capabilities.
- Tumor Biology Research: Serving as a valuable tool for researchers studying tumor heterogeneity and immune-tumor interactions in unprecedented detail.
- Low-Resource Clinical Settings: Providing single-cell resolution insights in settings where specialized, expensive assays are not feasible.
Advantages
- Single-Cell Resolution from Standard Images: Achieves single-cell classification from widely available and routinely used H&E slides, unlike most technologies that require complex, expensive techniques like single-cell RNA sequencing.
- Prediction of Therapy Response: Provides a detailed cellular composition of the tumor, which is linked to clinical outcomes to predict patient response and resistance to cancer therapies, aiding personalized medicine.
- Highly Accessible and Cost-Effective: Offers a non-invasive, cost-effective solution that integrates into existing pathology workflows, lowering the barrier to access for both research and clinical applications.
- Scalability and Speed: Uses digital pathology images, allowing it to be easily scaled for large datasets and integrated into existing digital pipelines for faster processing times.
- Broad Applicability: While primarily focused on one cancer type, the technology has been successfully applied to a second and can potentially be adapted for use in other types of cancer, broadening its impact.
Invention Readiness
This computational technology has been defined as a concept, and a prototype currently exists. The core machine learning model has been trained on annotated pathology images and has demonstrated its versatility by being successfully applied to multiple cancer types, including head and neck and liver cancer. Further studies would involve external validation across a broader range of clinical cohorts and cancer types to fully adapt the underlying model and demonstrate its full impact in routine clinical practice.
IP Status
Research Tool
Related Publication(s)
Kürten CHL, Kulkarni A, Cillo AR, et al. Investigating immune and non-immune cell interactions in head and neck tumors by single-cell RNA sequencing. Nat Commun. 2021;12(1):7338. doi:10.1038/s41467-021-27619-4
Dadey RE, Li R, Griner J, et al. Multiomics identifies tumor-intrinsic SREBP1 driving immune exclusion in hepatocellular carcinoma. J Immunother Cancer. 2025;13(6):e011537. doi:10.1136/jitc-2025-011537
