Ataraxis AI Steps Forward in Precision Cancer Care With $4 Million in Seed Funding

Through partnerships with hospitals, the Ataraxis team trained their models on thousands of historical medical images from over 4,500 breast cancer patients. The AI model, named Kestrel, combines predictions from multiple models, each averaging out potential errors to improve overall reliability. In retrospective analysis, Ataraxis Breast demonstrated up to 30% greater accuracy in identifying risk levels than existing methods, potentially allowing thousands of patients to avoid unnecessary chemotherapy.

Ataraxis team; Source: Ataraxis press release

Jan Witowski:

“If you don’t benefit from chemotherapy, you want to avoid it. And we show in the study and more broadly that we’re able to avoid unnecessary chemotherapy in potentially tens of thousands more breast cancer patients every year.”

The company was also designed with efficiency in mind. “I still have over $2 million in the bank,” Witowski noted, reflecting a resourceful approach to developing AI tools with minimized computational demands. Going forward, Ataraxis intends to expand its diagnostic scope beyond breast cancer, with plans to deploy Ataraxis Breast for use by early 2025, following further validation studies.

Read the full interview by Alex Knapp, Senior Editor at Forbes, on Forbes.com


Technology

The Ataraxis Breast test, as reported by the company and detailed in their preprint arXiv publication “A Multi-Modal AI Model for Breast Cancer Prognosis”, integrates high-resolution digital pathology features with clinical data to generate a continuous risk score predictive of cancer recurrence. Using self-supervised learning, Kestrel—a pan-cancer foundation model—is designed to capture morphological patterns strongly associated with disease progression across diverse breast cancer subtypes, including triple-negative cases. Ataraxis reports that this AI model improves precision by aggregating predictions across multiple models to ensure robust risk stratification, supporting clinicians in informed treatment decisions.

From the preprint: The Ataraxis AI test combines high-resolution pathology features from Kestrel, a self-supervised foundation model, with clinical data to predict breast cancer recurrence and mortality.

The company further claims that Ataraxis Breast can provide results within a single business day from standard H&E-stained slides, requiring no additional lab procedures or sample preparation. According to their study, Ataraxis reports a 50% reduction in prediction error compared to standard genomic assays, which they note is significant, particularly for patient subtypes currently underserved by available diagnostic tools. Designed for efficient integration, Ataraxis Breast reportedly minimizes tissue use and interfaces directly with EHR systems, offering a streamlined workflow intended to enhance precision medicine through clinical utility and improve timely intervention.

See also: In the Largest Study, AI Found To Increase Efficiency of Breast Cancer Screening—When Combined with Human Intelligence