RlapsRisk™ BC is a CE-marked AI diagnostic to help pathologists and oncologists assess the risk of relapse of early breast cancer patients.

Revolutionize breast cancer care management by democratizing access to actionable medical insights

Clinical context

The incidence of breast cancer is high and approximately 10% of all patients will relapse after their initial treatment each year. Once a patient has relapsed, their chance of survival decreases dramatically. Currently, clinicians lack the tools to identify these high-risk patients at an early stage. Current methods either lack robustness or accuracy or they are expensive and non-routine such as genomic sequencing.

Scientific question

Is it possible to identify high risk of relapse patients from HES Pathology Images with AI? (In the HER2-, ER+ Subgroup)

Development milestones

In July 2019, Owkin wins the ‘AI for Health’ challenge receiving a €1.2m grant from Region Île de France for a breast cancer & histology project.

Abstract presented

Abstract presented

Abstract presented

Accurately discriminate between low and high risk breast cancer patients (ER+/HER2) using digital pathology slides (HES) to determine the right treatment pathway.

  • Suitable for adults with primary invasive breast cancer (ER+/HER2-)

  • Cumulative sensitivity is significantly greater than those obtained by standard clinical scores

RlapsRisk BC achieves 78% sensitivity and 80% specificity for post-treatment time-dependent accuracy at 5 years, outperforming current clinical scores in practice¹

ScoreCumulative Sensitivity*Dynamic Specificity
RlapsRisk BC0.780.80
PredictUK0.410.84
CTS00.700.79
*Cumulative Sensitivity/Dynamic Specificity are natural extensions of sensitivity/specificity to the setting of time-to-event outcomes, such as the metastasis-free interval (MFI), understood as the time to distant relapse occurrence from initial surgery. In use here as they easily accommodate time-dependent outcome status as well as right-censoring

Survival by Risk Group

Training Cohort

P-value < 0.01.        Sample size N = 1424

When run on our training cohort, RlapsRisk BC’s performance demonstrates strong discrimination between risk groups, better informing oncologists on the risk classification of their patients to aid in treatment decisions.

Blind Validation Cohort

P-value < 0.01.         Sample size N = 676

Through blind validation on a separate cohort, RlapsRisk BC demonstrates that it is well calibrated with similar performances on data it was not trained on.

Clinical Impact

Further analysis shows how well defined the low risk population is – with a distant 5 year relapse rate of 0.3%. This indicates RlapsRisk BC can impactfully support oncologist decision making.

AI powered diagnostics that seamlessly integrate into the pathologist and oncologist workflow

The following is an example of an AI Dx optimized workflow

Due to variability across pathology labs, workflows may vary by institutions and disease indications

AI powered diagnostics that seamlessly integrate into the pathologist and oncologist workflow

The following is an example of an AI Dx Optimised Workflow

Due to variability across pathology labs, workflows may vary by institutions and disease indications

  • Workflow agnostic – deployable within any IMS system or shared directory

  • PDF report with intuitive design

Contact us

Thanks to the solution we now have a better understanding of the underlying mechanism of highly aggressive tumors and the treatment needs for these patients. Identifying very high-risk patients earlier will enable us to adjust the therapeutic strategy for more favorable patient outcomes.

Professor Fabrice André

Director of Research, Gustave Roussy

RlapsRisk BC is CE-IVD marked for diagnostic use in the EU. In all other countries including the United States, the use of RlapsRisk BC is limited to Research Use Only, not for use in diagnostic procedures. For detailed information on regulatory or safety information, system configuration and indications for use, contact Owkin.

Images shown may represent the range of product, or be for illustration purposes only, and may not be an exact representation of the product.

(1) Garberis IJ, Gaury V, Drubay D, et al. Blind validation of an AI-based tool for predicting distant relapse from breast cancer HES stained slides. Poster presented at: European Society for Medical Oncology (ESMO); May 9th - 13th 2022; Paris France.