Jump to section
Use case highlights
In this use case, we show how Sonrai Discovery transforms complex pathology images into clear, analysis-ready tabular data using state-of-the-art imaging foundation models. By removing the usual barriers to integrating imaging into translational and clinical research, Sonrai Discovery lets researchers work across imaging, omics and clinical data, all within a single, centralised cloud environment.
Easy and Meaningful Imaging Features
From tensors to tabular
Processing whole slide images into Whole Slide Embeddings (WSEs) transforms complex, high-dimensional image data into tabular formats that can be processed with standard data science tools.
1000x Reduction
From gigabytes to megabytes
WSEs are much smaller than raw whole slide image data. Reducing the volume while maintaining the principal features enables faster experimentation and easier data handling.
50% Reduction
In computational costs
Once extracted, image features are tracked in Sonrai Discovery and reused for new embeddings, classification or segmentation applications, eliminating redundant compute over time.
The challenge
Translational research teams and data scientists in clinical research often work with complex, multi-modal datasets such as H&E and IHC imaging, RNA-Seq, proteomics and clinical data. While imaging data can be a rich source of insights — for example, in identifying or developing prognostic biomarkers — teams frequently lack the specialised infrastructure, analytical tools and domain expertise needed to extract value from these images and integrate them effectively with other data types.
Do these challenges sound familiar?
If you need advice on the challenges you're facing, our team are ready to help.
The strategy
Using the Sonrai Discovery platform, we applied advanced foundation models trained across a wide range of tissue types and stains, including IHC, to generate a dataset of image features from whole slide images (WSI). Sonrai Discovery was used to:
- Run the pretrained foundation models
- Manage the computational complexity
- Manage feature extraction from imaging data
- Save and store extracted features for multi-modal analysis
These extracted image features were processed into whole slide embeddings (WSEs) of each of the WSIs in the dataset. A dataset of WSEs transforms each slide into a single row in a tabular dataset, with columns corresponding to the feature patterns extracted from the image. WSEs are compatible with existing data science processes, and integrate easily with clinical and omics data — it's just a case of joining columns. Researchers can unlock multi-modal integration and analysis with a click of a button.
Within minutes, we were able to:
- Merge the WSE dataset with clinical survival data
- Apply Python's sksurv survival analysis within Sonrai Discovery using notebooks
- Use the survival analysis to train a model for image-derived prognostic biomarkers
- Improve the performance of the model using different algorithms in Jupyter Notebook within Sonrai Discovery, optimising it within minutes thanks to the light but powerful format of the WSEs
- Expand the model by integrating omics features with the image-derived features to build a more powerful multi-modal prognostic biomarker
- Generate Kaplan-Meier survival curves within seconds to visualise the prognostic value of imaging features
Further to this, we could combine imaging insights alongside findings from clinical and omics analysis in Sonrai Discovery Insights, delivering a complete multi-modal picture for decision-making.
The foundation model-derived features don't go to waste, either. Features derived from a given image are tracked in Sonrai Discovery and can be reused to develop classification, regression or segmentation models for other research projects in future.
Generating a WSE provides tissue segmentation as a byproduct — here, muscle tissue has been highlighted, generated using self-supervised clustering.
Unlocking Prognostic Biomarkers: Integrating Imaging and Omics with Foundation Models
Watch the on-demand webinar on multi-modal data and foundation models for biomarker discovery.
Results & impact
WSEs are a powerful tool that transform image-derived data into an analysis-friendly format, while maintaining predictive power. Imaging data transformed into WSEs is:
- Lightweight and quick to process
- Easily integrated with omics and clinical data
- Immediately ready for further analysis
Sonrai Discovery makes advanced AI-powered pathology analysis accessible and reproducible, bringing imaging, omics and clinical data together in a compliant, traceable, cloud-native research platform. By transforming raw slides into WSEs, data scientists and translational researchers can extract value from imaging data using state-of-the-art pathology foundation models without the engineering overhead.
Using WSEs, theories that would once have taken hours to investigate can now be trialled within minutes. This lightweight image format lowers the burden of experimentation, letting researchers fail fast, learn more and discover new possibilities.
Ready to see your imaging data in a new light?
Integrate, manage and analyse imaging, omics and clinical data to accelerate translational research and streamline pathology workflows.
Related Case Studies
Case Study
Integrating Imaging and Omic Data
How an oncology therapeutics developer connected imaging, RNA-Seq, proteomics and clinical data within days — a 297% increase in meaningful data points for patient stratification.
Case Study
Multi-Modal Biomarker Discovery
How Sonrai's integration of genomic, proteomic, imaging and clinical data supported multi-modal biomarker discovery.
Like what you see? Let's talk
We listen to your problems, and give you the confidence to make your decision.
Take the first step