Blog
July 3, 2025
5 mins

Can an AI diagnostic help doctors make more precise treatment decisions for breast cancer?

In a recent Nature Communications article1 a preliminary version of Owkin’s RlapsRisk® BC uses AI to bring a new layer of insight to breast cancer diagnosis, helping reveal which patients may be at risk of relapse.

The clinical dilemma

Despite significant progress in diagnosis and treatment, breast cancer (BC) continues to be  the leading cause of death from cancer for women worldwide. A critical question for many patients, particularly when cancer is diagnosed early, is whether the harsh side effects of chemotherapy and the potential long-term impact on quality of life can be safely avoided. As there are a growing array of targeted treatments, patients also wonder if they are eligible to receive precision therapies, like CDK4/6 inhibitors.

For oncologists, determining the optimal therapeutic strategy for each patient requires careful evaluation of multiple factors. Accurate risk assessment is fundamental to their decision-making process, as it directly influences the intensity and duration of recommended treatments. When their assessment indicates a high risk of distant relapse, they may need to consider more aggressive therapeutic approaches or extended treatment protocols. Conversely, for patients identified as low-risk, oncologists can confidently recommend less intensive treatment plans, helping patients avoid unnecessary therapies and their associated side effects while maintaining positive outcomes.

Challenges in predicting breast cancer relapse

Assessing a patient's likelihood of recurrence is not straightforward. Breast cancer manifests differently between various patient subgroups, spurned on by genetic differences, clinical factors, and disease markers. With ER+/HER2- invasive breast cancer, the largest subgroup consisting of approximately 70% of all breast cancers, oncologists have to sift through a wide spectrum of outcomes and treatment options.

Additionally, current methods for assessing risk, prognostic scores calculated from clinical and histopathological factors or molecular sequencing, don't confidently classify all subgroups of breast cancer patients. To varying degrees, they struggle to differentiate the intermediate risk population and perform poorly in specific subgroups. Additionally, some of these methods, such as genetic testing, can be expensive and are not always routinely available. Variations in test results, restrictive indications, and limited reimbursement in many healthcare systems present challenges to their widespread use.

AI diagnostic shows promise for improving relapse risk assessment

Today, prognosis assessment is evolving with AI and machine learning addressing biological and clinical challenges. Recent research spotlights deep learning's capacity to leverage WSI for predicting patient outcomes and identifying prognostic features for breast cancer2,3,4. Digital pathology is becoming an integral part of routine clinical practice, with biomarker assessments helping to streamline testing, refine prognosis,  and guide more informed treatment decisions.

Owkin has developed an AI diagnostic to predict the risk of recurrence in early invasive breast cancer patients (ER+/HER2-) directly from H&E whole slide images (WSI). In validation tests of a preliminary product version, RlapsRisk® BC gives a strong prognostic indication of risk of relapse from the H&E image alone. When RlapsRisk® BC integrates clinico-pathological factors, it effectively segments patients into low- and high-risk groups with strong discriminative power across the entire population and key clinical subgroups. This stratification performance holds even in subgroups with known difficult prognosis, such as grade 2. In this study, the RlapsRisk® BC prototype classified 78% of the patients in the validation cohort in the low-risk group and the remaining 22% were labeled high risk. As anticipated, there was a much higher percentage of patients who relapsed in the high-risk group (10%) and in the low-risk group there were minimal patients who relapsed (1.3%). This indicates that RlapsRisk® BC is high-performing classifier for risk assessment.  Additionally, the improved performance of the combined model compared to the clinico-pathological factors alone also shows that the RlapsRisk® BC assessment of the WSI is using additional information found within the tissue to give it's prognostic score1. This information is not captured or leveraged with traditional scoring methods.

Interpretability analysis highlights the relevance of digital pathology for risk assessment

Much like studying a rainforest requires looking beyond individual trees to understand how they interact with surrounding vegetation, soil conditions, and the broader ecosystem, pathologists and oncologists gain valuable information from examining the disease and its environment. While genetic tests examine specific molecular markers, RlapsRisk® BC analyzes the entire tissue landscape, including critical features like blood vessel formation and tissue architecture in its prognostic score.

The ability to detect subtle patterns in vascular structures and tissue organization can reveal important indicators of cancer progression that might otherwise go unnoticed. These insights, combined with traditional diagnostic methods, enable healthcare providers to make more precise and informed decisions about patient care, potentially identifying cases that require more intensive treatment approaches.

Conclusion

Between treatment advances for breast cancer and the limitations of current methods for assessing risk of recurrence, it can be challenging to navigate the therapeutic landscape to ensure patients receive the most effective care plan for their cancer. RlapsRisk® BC aims to help oncologists more precisely evaluate their patient's risk level for recurrence, by adding an independent prognostic factor into the assessment that stems from AI tissue analysis. Clinical validations of the product prototype demonstrate the potential RlapsRisk® BC has in better stratifying breast cancer patients against traditional methods, even in subgroups typically hard to define. RlapsRisk® BC paves the way for AI patient outcome prediction solutions to leverage digital pathology to unlock more precise treatment allocation for oncologists, and their patients.

Learn more about what's next for RlapsRisk BC. Register for our upcoming webinar here.

References
  1. Garberis, I., Gaury, V., Saillard, C. et al. Deep learning assessment of metastatic relapse risk from digitized breast cancer histological slides. Nat Commun 16, 5876 (2025). https://doi.org/10.1038/s41467-025-60824-z.
  2. Ibrahim, A. et al. Artificial intelligence in digital breast pathology: Techniques and applications. Breast 49, 267–273 (2020).
  3. Amgad, M. et al. A population-level digital histologic biomarker for enhanced prognosis of invasive breast cancer. Nat. Med. 30, 85–97 (2024).
  4. Ogier du Terrail, J. et al. Federated learning for predicting histological response to neoadjuvant chemotherapy in triple-negative breast cancer. Nat. Med. 29, 135–146 (2023).

RlapsRisk® BC v1 is not available on the market, and not for clinical use. Manufacturer: Owkin France. RlapsRisk® is a trademark of Owkin Inc.

Authors

Talia Lliteras
María Lola Álvarez

Testimonial

No items found.
Can an AI diagnostic help doctors make more precise treatment decisions for breast cancer?

No items found.
No items found.
No items found.