Our Technology
We develop novel AI models to uncover imaging biomarkers indicative of cancer prognosis. We use them to train advanced time-to-event models that accurately predict patient outcomes and treatment response.
Pioneering AI Methods
Multi-modal learning and self-supervised learning enable us to create groundbreaking tools to enhance the accuracy and efficiency of clinical prognosis, ushering in a new era for precision oncology.
Key Characteristics of Ataraxis Platform
Clinically actionable models predicting patient outcomes and treatment response
To personalize patient care, we leverage information from multiple sources, capturing the heterogeneity of patients and their disease. With complex methods learning from multiple modalities, we uncover patterns imperceptible even to human experts that are strongly correlated with cancer progression and can be used to help inform treatment decisions.
Powered by pan-cancer foundation models
Our first foundation model for digital pathology, Kestrel, is pretrained with hundreds-of-millions of pathology images from cancer tissue specimens. It has learned to extract strongly predictive morphological features captured at the microscopic level from whole pathology slides. Kestrel is developed with cutting-edge methodology, including vision transformers and self-supervised learning, as well as proprietary methods to optimize and scale these models which we are publishing at top AI venues.
Fast, accurate, robust, and extensible
Our tests, including Ataraxis Breast, can be delivered in an unprecedented 1 hour within receiving specimens, and predictions can be made using data from already existing specimens without any need for additional procedures.
Ataraxis Breast has already demonstrated superior performance relative to a standard-of-care genomic assay. Furthermore, it works across breast cancer subtypes, including those for which there are no diagnostic tools currently recommended by clinical guidelines.
Importantly, our technology is not limited to any particular indication, allowing us to rapidly extend our tests to answer new questions regarding the care of entirely new and underserved patient cohorts.