What are the main challenges in Breast Cancer?
Breast cancer remains the most common cancer for females worldwide despite the medical and technological advances (World Cancer Fund, 2018). In fact, over 1 in 8 women will be diagnosed with it in their lifetime. However, there is hope, thanks to novel personalized treatment options such as targeted therapy and global screening campaigns. As a result, today’s 5-year survival rates are as high as 91%. On the other hand, the challenge we face in breast cancer is identifying patients at risk of recurrence. This will allow clinicians can alter their treatment preventatively. Once a patient has relapsed, their chances of survival decrease significantly. For example, 73% of patients who develop metastasis will die after 5 years.
How do we currently predict recurrence in Breast Cancer?
The current state-of-the-art technology to predict breast cancer recurrence includes repetitive testing for patients at a high-frequency (Magnetic Resonance Imaging ‘MRI’ or mammography) and non-routine and expensive techniques such as serial measurement of circulating tumor DNA. There is a clear need for accessible prognostic tests that can predict disease progression among early-stage patients. This will ensure patients receive the best treatment for their predicted disease course, manage the expected relapse population and further reduce the breast cancer mortality rate.
Owkin Collaborative AI Research
At Owkin, we believe that a combination of Artificial Intelligence (‘AI’) and collaborative research is critical to tackling the most challenging scientific questions, including breast cancer recurrence in patients. To this end, we have developed our Federated Research Ecosystem to bring academic and pharmaceutical industry researchers together to leverage Owkin biomedical AI capabilities in order to advance medical research and accelerate drug development.
Our ecosystem consists of four components:
- Firstly, the network (Owkin Loop),
- Secondly, the technology infrastructure (Owkin Connect),
- Thirdly, the AI software tool (Owkin Studio); and
- Finally, the expertise (Owkin Lab).
Projects are then articulated by therapeutic areas, centered around renowned researchers and unsolved research questions called AI Loops.
Owkin is committed to fostering awareness and innovation in breast cancer. As part of the Breast Cancer Awareness initiative in October, we want to introduce you to Owkin Breast Cancer AI Loop.
We will look specifically at the research questions of focus, the key opinion leaders and leading research centers involved, the novel datasets curated and analyzed, and the cutting-edge federated learning techniques deployed.
Owkin Breast Cancer AI Loop
If we look back at our research ecosystem’s four components. We can use them to explore the breast cancer AI Loop. The Owkin Loop is our worldwide network of leading academic medical centers. Within the Loop, we can access high-quality research-grade patient data and cutting-edge clinical expertise. Furthermore, our Breast Cancer AI Loop is a fantastic example of this network in action. The projects we describe here highlight the importance of all components of the research ecosystem.
Institute Gustave Roussy (‘IGR’) – The Race-AI Project
The Race-AI project is an 18/20-month project with the European Leader in cancer research, IGR. It is funded by the Region Ile de France. The project’s goal is to use AI to predict the recurrence rate in breast cancer patients.
Our Owkin Loop network includes a full dedicated team of experts to operationalize this Race-AI Project: the world-renowned clinician Dr. Fabrice Andre, pathologists such as Dr. Magali Lacroix-Triki and clinical research assistants dedicated to preparing the data to intensify our research.
As part of this Race-AI collaborative project, IGR (Institut Gustave Roussy) provides an expertly curated dataset of nearly 2,000 breast cancer patients.
The data is multimodal (clinical and pathological) data. It follows the patient’s response to various treatment options for over 15 years.
Owkin’s AI software tool, Owkin Studio allows Owkin and IGR to work on this dataset collaboratively without extracting the hospital’s data. IGR can upload its dataset and research questions into Owkin Studio. IGR clinicians work collaboratively with Owkin data scientists through the Owkin Studio interface.
Back at the Owkin offices. The data scientists within our Owkin Lab can access the de-identified data via a VPN. They then train and interpret machine learning models on the IGR data to predict breast cancer recurrence patients. IGR maintains full control of their data and every training action is traced to ensure transparency and security.
We are just at the beginning of this exciting collaboration. We believe it could improve patient outcomes and well-being by proposing a more efficient and less invasive method to predict breast cancer recurrence.
Is your research team is interested in getting involved with this project? To get access to the first insights and findings, please reach out via the link below.
Dr. Magali Lacroix-Triki, Pathologist at IGR, said, “We started working with owkin in 2019 to predict breast cancer recurrence and it’s been a very enriching experience. They bring AI expertise and tools to make our projects reach the highest levels. I trust them to handle data with care, with high-security standards, and to work with them on putting this into industrialization.”
Unique access to the IGR Race-AI dataset is incredible as the cohort represents such a large quantity of high-quality multimodal data points. However, at Owkin, we know that it is rare to find cohorts that capture data heterogeneity and integrate different modalities. The standard approach to collect datasets large enough for effective machine learning training is to aggregate many distributed datasets. However, we don’t believe in this paradigm of data aggregation. There are too many data security and privacy risks and governance questions. This makes it hard to reach a trust level satisfying all partners.
This is the reason we developed federated learning for healthcare. Federated learning is powered by Owkin Connect. Connect allows machine learning models to train on distributed data (data at more than one hospital) without moving it from behind the hospital firewalls. The algorithms travel between the hospitals, but the data remains where it belongs – on site. The resulting federated models maintain the performance seen with a data-pooling approach while protecting data privacy and identifying the dataset’s biases. Owkin Connect also promotes trust between all parties by tracing all actions and training steps via its incorruptible ledger.
Owkin Connect is not just a visionary idea but a proven technology used in live research projects.
One such project is the Healthchain consortium, led by Owkin. The goal of this project is to use federated learning to securely allow public and private consortium members to build collaborative AI models for dermato-oncology, anatomo-pathology, and fertility.
One project within this consortium investigates the response to the treatment of neoadjuvant chemotherapy in breast cancer patients. Institut Curie and Centre Léon Bérard (two French hospitals) provide the digital pathology datasets and specific research questions while Owkin deploys Owkin Connect to train the federated models across both institutions. In this scenario, both centers benefit from a more intelligent model (higher predictive power and generalizability) trained on both hospital datasets without compromising their patient data.
The project is progressing well; the federated models’ live runs are ongoing, with the first results expected to be published soon.
Pierre Etienne Heudel, an oncologist at the Centre Léon Bérard, said, “It’s thrilling to start such a project with Owkin and Institut Curie. Federated Learning is a pioneering technology, and we hope that this project will help show the impact it can have.
In conclusion, we have outlined two examples of projects within the Breast Cancer AI Loop and how they utilize the four components of Owkin unique Research Ecosystem to advance medical research.
We are proud that our Research Ecosystem gives us unique access to the world’s leading researchers and their curated datasets to build state-of-the-art multimodal models to answer the most pressing clinical questions. We are also excited that our pioneering research in multimodal AI and federated learning for healthcare – with the potential to break down research silos and foster a new form of collaborative research – is progressing well in real-life situations. It allows us to push our vision for AI in medical research. We believe that all actors should demand this level of security and efficiency when they leverage health data.
Get in touch if you want to set up you own federated data network to address a particular research question!