OwkinStudio
Studio Radio
Train machine learning models on CT-Scans
Freely accessible for researchers working on COVID-19
see case study
Building a bridge between medical researchers & data scientists
Engineered for you, medical researchers, Owkin Studio is a software platform dedicated to finding new biomarkers, building prognosis models, and predicting response to treatment from multimodal patient data.
Embedded into Owkin’s unique federated learning and AI environment, Studio enables you to intuitively create and manage machine learning-based research projects, from cohort & project management to results interpretation.
Unlike traditional black-box AI, Owkin Studio helps you to understand & interpret the results from a biological perspective. Collaborate hand-in-hand with our in-house scientists who will dedicate their expertise to your project.
Advance your research, publish your results, and join our mission to develop better drugs for patients.
Request access to Owkin Studio trialWorking with Owkin Studio
At Owkin, confidentiality and security of patient data matter above all. To respect this commitment, we developed and patented a unique set of technologies, served through Studio and fueled by state-of-the-art federated learning techniques.
Owkin Studio is deployed in your institution, on premise or in the cloud. The data stays behind your institution’s firewall and can be used to locally train machine learning models.
Take your project one step further and collaborate with Owkin Lab’s data scientists to design more advanced experiments or train additional algorithms, leading to the publication of relevant scientific results and medical discoveries in high impact journals.
Learn more about deployment
A ground breaking approach
How Owkin Studio accelerates your research
Intuitively set up AI projects
Biologically interpret results
Work with Owkin Lab
CASE STUDIES
Malignant Mesothelioma – Studio Pathology

New biomarkers in malignant mesothelioma
In 2018 and 2019, Owkin collaborated with Francoise Galateau-Sallé, Department of Biopathology at Cancer Center Léon Bérard, to identify new histological features predictive of the survival of patients suffering from malignant mesothelioma.
Our model, MesoNet, trained on the Mesopath/Mesobank cohort, located unkown regions in the stroma that were associated with low patient survival. MesoNet is an example of how machine learning can provide tools and interpretation features to identify new biomarkers.
This new biomarker was published in Nature Medicine, and research is now continuing to identify new targets to deliver better treatments for malignant mesothelioma.
Discover how the model works, explore the cohort and the results, and analyze the tiles that display low and high survival scores.
Explore MesoNet live in StudioCOVID-19 – Studio Radiology

Understanding Covid-19 severity markers with our ScanCovIA model
In May 2020, Owkin developed a machine learning model in partnership with Gustave Roussy, Hôpital Kremlin-Bicêtre & INRIA to predict the severity of SARS-CoV-2 infection from initial CT-scans and clinical variables.
The preprint of our paper is available here. The results show that beyond AI modeling, a composite score integrating selected radiological measurements with relevant clinical and biological variables provides the most accurate predictions, and can rapidly become a reference for severity prediction.
The model is open sourced and can be visualized on public data in Owkin Studio.
Get in touch to try Studio RadioHE2RNA – Studio Pathology
Predict RNA-seq expression of tumors from whole slide images
HE2RNA is a deep learning model built to predict RNA-seq expression of tumors from WSIs without the need for expert annotation. The model robustly and consistently predicted subsets of genes expressed in different cancer types, including genes involved in immune cell activation status and immune cell signaling.
HE2RNA is interpretable by design, and provides virtual spatialization of gene expression. Moreover, the transcriptomic representation learned by HE2RNA can be transferred to improve predictive performance for other tasks, particularly for small datasets.
This model was published by Nature Communications and has many high-impact applications, that range from direct evaluation of immune response to the augmentation of existing pathology cohorts with predicted gene expression.