Unlocking Prognostic Biomarkers: Integrating Imaging and Omics with Foundation Models
Learn how image-derived features can be integrated with omics and clinical data to improve prognostic modelling - with a live demonstration on Sonrai Discovery using Paige's Verkhov foundation model.
Featured in partnership with
Watch On Demand
Instant access. No commitment.
See how image-derived features improved prognostic performance when combined with clinical and molecular variables in a pancreatic cancer cohort.
The Challenge
Histopathology images contain rich biological signal, but turning a multi-gigabyte whole slide image into something that can sit alongside omics and clinical data in an analysis is a genuine technical barrier.
The Approach
Pathology foundation models convert image tiles into feature embeddings - compact numerical representations that capture morphology patterns learned from millions of slides.
The Outcome
Once embeddings are generated, image-derived features can be combined with omics and clinical variables in standard tabular analysis - enabling new prognostic and biomarker discovery workflows.
From Pathology Images to Prognostic Biomarkers
This is the process James walks through in detail during the session - and the one demonstrated live on the TCGA pancreatic cancer dataset.
Questions This Webinar Helps Answer
The session covers practical answers for teams looking to bring imaging data into their multi-modal analysis.
What Is a Pathology Foundation Model and How Does It Work?
Dr Siqi Liu explains how Paige's foundation model was trained on over 3 million whole slide images across 40+ tissue types, and what makes a foundation model different from a task-specific pathology AI tool.
How Do I Turn Image Data Into Something I Can Analyse?
James Leech walks through the full pre-processing pipeline - from whole slide images to feature embeddings to whole-slide embedding representations that fit into a standard tabular dataset.
Can Image Features Actually Improve Prognostic Models?
See a live demonstration on a TCGA pancreatic cancer cohort where adding Verkhov 2 image-derived features alongside clinical and MSI status data improved prognostic performance.
How Do I Combine Imaging, Omics and Clinical Data in One Analysis?
James Leech explains how image embeddings are converted into a standard tabular format that sits alongside RNA-seq and clinical variables - so existing analysis methods and tools can be applied directly.
What You'll Learn
Foundation Models for Pathology
Dr Siqi Liu introduces Verkhov 2 - how it was trained, how it compares on public benchmarks, and real-world applications including pan-cancer detection and a digital biomarker panel covering over 1,600 biomarkers.
From Images to Analysis-Ready Data
James Leech explains the full imaging workflow - feature extraction, whole-slide embeddings, clustering-based morphology mapping, and how image-derived features integrate with omics and clinical data on Sonrai Discovery.
Live Demo and Q&A
A live walkthrough of the TCGA pancreatic cancer dataset on Sonrai Discovery - generating image embeddings, defining tissue clusters and identifying a feature with independent prognostic value, followed by audience Q&A.
TCGA Pancreatic Cancer Cohort
See how Verkhov 2 image embeddings were combined with omics and clinical data from a TCGA pancreatic adenocarcinoma cohort - and how a single image-derived feature was identified with apparent independent prognostic value.
- Verkhov 2 feature extraction from whole slide images
- Whole-slide embeddings via clustering and Gaussian mixture models
- Integration with RNA-seq, MSI status and clinical variables
- Cox regression to identify prognostic image features
Improved Prognostic Performance
Adding image-derived features to clinical variables and MSI status improved the C-index from approximately 0.78 to 0.83 in a cohort of over 1,300 patients across stage I-III pancreatic cancer.
Converting Whole-Slide Images into Analysis-Ready Data
Tiles across the dataset were clustered into 16 morphology groups, then mapped back onto slides to validate tumour and stroma regions - reducing gigabytes of image data to a compact, analysis-ready format.
Candidate Prognostic Feature
A feature from a stroma-associated cluster demonstrated prognostic signal beyond tumour stage and hypoxia score within the analysed cohort - identified through Cox regression on the integrated dataset.
About Verkhov 2
- Trained on 3M+ whole slide images, 40+ tissue types
- Ranked #1 on public pathology benchmarks (Kaiko leaderboard)
- Verkhov 1 fully open-sourced under Apache 2.0
- Paige received the first FDA breakthrough designation including a foundation model
Built for Teams Working Across Imaging, Omics and Clinical Data
Particularly relevant if your team is exploring how imaging data can add value to existing omics and clinical analysis.
Translational Scientists
Wanting to understand whether histopathology images contain prognostic signal that complements existing omics and clinical biomarkers.
Bioinformaticians and Computational Biologists
Looking for a practical route to integrate image-derived features into existing tabular workflows without building a computer vision pipeline from scratch.
Pathologists
Interested in how foundation models can support semantic annotation, morphology-based clustering and collaboration with computational teams.
Biopharma Research Teams
Evaluating multi-modal data integration approaches for biomarker discovery and translational research programmes.
Meet the Speakers
AI and pathology expertise from Paige, alongside machine learning and platform expertise from Sonrai.
Dr Siqi Liu
Vice President of AI, Paige
Dr Liu leads AI product development at Paige, with a PhD in biomedical image computing from the University of Sydney. He leads the team behind Paige's pathology foundation models.
LinkedIn ↗
James Leech
Machine Learning Expert, Sonrai Analytics
James develops clinical-grade machine learning models using Python and AWS. He leads imaging workflow integration on Sonrai Discovery and presents the live TCGA demonstration.
LinkedIn ↗
Amélie Viratham
Imaging Expert, Sonrai Analytics
Amélie hosts this session, drawing on her imaging expertise to frame the discussion between Paige and Sonrai's teams and guide audience Q&A throughout.
LinkedIn ↗Access the Full Session
Watch Dr Siqi Liu introduce Verkhov 2, James Leech's walkthrough of the imaging workflow, and the live TCGA pancreatic cancer demonstration on Sonrai Discovery - including the prognostic performance improvement and audience Q&A.
Watch On Demand
Enter your work email to access the recording.