GenoMed4All is building an open, federated data hub to explore new AI models and services for clinical support in blood disorders.
Haematological diseases comprise a large group of up to 450 disorders resulting from abnormalities of blood cells, lymphoid organs and coagulation factors, generally categorised as either oncological or non-oncological. Most have a genetic background, and they represent a significant public health challenge: haematological malignancies account for about 5% of cancers, most can cause chronic health problems, and many are life-threatening conditions. In 2016, the European Haematology Association (EHA) estimated the financial burden of blood disorders on European society to be approximately €22.5bn per year. Moreover, data scarcity remains a pressing issue: the number of available samples for blood disorders remains small and is characterised by a high level of fragmentation, which mostly stems from the sensitive nature of data.
GenoMed4All is the European response to this scenario: an EU-funded initiative to radically transform the way we approach diagnosis, prognosis and treatment in haematological diseases through the use of Artificial intelligence (AI).
Applying novel AI models to a combination of already established clinical parameters and advanced genomic profiling will help us explore innovative diagnostic, prognostic and therapeutic strategies. As such, we aim to advance personalised medicine approaches in haematology, pooling multimodal data sources, such as clinical, multi-omics (e.g., radiomics, genomics, metabolomics…) and real-world data (e.g., PROMs, fitness information, nutrition, environment…) via a secure and trustworthy platform to enhance diagnostic capacity, assess treatment options and predict outcomes in haematological diseases.
Our mission is to link Europe’s most relevant repositories for blood disorders, facilitating the standardised sharing of cross-border data, ensuring full compliance with data protection legislation and ethical principles, and thus demonstrating the potential and benefits of reliable and explainable AI technologies in personalised medicine.
For the past 4 ½ years, the project has mobilised a consortium of 23 partners from Spain, Italy, Germany, France, Cyprus, Greece and Denmark, covering the whole value chain of clinical, regulatory and ethics research, academia, healthcare, disruptive tech and digital service providers. GenoMed4All also benefits from the support, resources and active participation of ten clinical partners from ERN-EuroBloodNet, the European Reference Network for oncological and non-oncological rare haematological diseases. ERN-EuroBloodNet is a joint effort between the EHA, the European Network on Rare and Congenital Anaemias (ENERCA), the EURORDIS European Patient Advocacy Groups (ePAGS) and the EHA Patient Organisations Workgroup.
A federated platform for clinicians and researchers
One of the critical challenges for AI applications in a clinical setting resides in how to effectively access and process sensitive clinical data for algorithm development and training while still maintaining full privacy and keeping the data secure.
To address this issue, the team at GenoMed4All is currently working on the final deployment stages of our Federated Learning (FL) platform; a shared, distributed space where clinicians and researchers can work together in the definition, development, testing and validation of AI models to improve the way we currently diagnose and treat haematological diseases in the EU.
In this scenario, a federated infrastructure is a perfect fit since data is not required to leave the clinical data providers’ premises at any point. Our privacy-by-design approach relies on this FL framework to offer security in all data exchanges, model training and storage. Coincidentally, this approach also solves the issue around data fragmentation by effectively linking multiple health data sources for the purpose of training predictive AI models.
By allowing experts to access and analyse distributed clinical data without any physical transfer involved, our platform opens up new possibilities for developing AI-driven diagnostic and decision-support tools and ensures that sensitive patient data stays on-premises, effectively safeguarding patients’ rights.
For clinicians, GenoMed4All’s platform will act as a local decision support system to input new prospective and retrospective patient data, extracting insights from an ever-learning model. In this way, we aim to equip clinicians with predictive algorithms and data insights that can further support their daily practice. For researchers, the project offers an AI sandbox to explore datasets, develop, train and benchmark new AI models on real-world data and to test their clinical usability. In this federated framework, researchers can leverage data from multiple providers with minimal degradation on model performance, better generalisation and minimal bias.
Our ethics-first approach to collection, harmonisation and cross-border data sharing
Healthcare data holds enormous potential, but most of it remains dormant. The healthcare sector (both in the EU and on a global stage) is rich in data but poor in information, and there is a distinct lack of evidence-based information to guide research, policy decisions and regulations.
Different hospitals may generate, annotate and store data following widely different formats: they may choose to inform specific data fields and disregard others, ultimately making it so that even when data is accessible, it is difficult to discover and understand its underlying structure and meaning. Moreover, available data may be incomplete, thus negatively impacting its sharing and reuse potential and highlighting the pressing need for broadly applicable standards capable of interpreting it.
In light of this, advocating for data quality and reliability is a necessary requirement to provide data-driven solutions to unmet clinical needs. Automatic tools and AI algorithms can help both in the pre-processing of information and data curation.
Working to improve the accountability, transparency and usefulness of AI tools is also key to building trust among patients, caregivers and healthcare professionals. At GenoMed4All, we believe that a shift in how we introduce these tools into a clinical setting is needed to ensure their accountability, transparency and usefulness among healthcare professionals. In alignment with this belief, the project has built robust, ethical agreements to bring community members together through explainable AI, enabling collaborative, cross-border data sharing that is standard-compliant.
From an ethical perspective, we have developed strong data protection management protocols and risk assessment plans and generated a comprehensive knowledge base on research integrity and data reuse to support a governance model that reconciles both innovation and institutional due process. This is especially relevant when targeting rare disorders, which suffer from aggravated bias, inclusivity and representation issues and where available data for AI use might not always be easily available. To reinforce our commitment to this ethics-first approach, GenoMed4All has also produced a series of recommendations to guide the ethical development and deployment of AI in a clinical setting, in order to be fully aligned with the EU AI Act.
The project has also conducted extensive work on the development and deployment of a common data model and dedicated bioinformatics pipelines for data extraction and semi-automated data curation across participating institutions, supported by a shared FAIR (Findable, Accessible, Interoperable and Reusable) methodology for our data collection and transformation, in order to make sure that all data coming from clinical sites could be properly harmonised and anonymised before being integrated into the platform.
Through a process of sequencing and standardisation, GenoMed4All has also produced genomic standardisation guidelines to allow for future data interoperability so the wider research and clinical communities can openly access, reuse and expand upon these AI-driven models, ultimately advancing research and improving the standard of care for blood disorders in Europe.
Validating AI in blood disorders: Our use cases
Disruptive innovations in healthcare should always emerge from and be informed by clinical needs. For AI tools to be trusted and adopted, clinicians need to be involved right from the start, and AI practitioners need to fully understand the real-world context in which their applications will operate to effectively navigate it. Consequently, bridging the technical-clinical divide and validating our AI models in real-world conditions has played a critical role in ensuring their fit in clinical practice and in ultimately reaching the patient.
GenoMed4All’s multidisciplinary team has brought this theory into practice through a distributed piloting exercise across multiple clinical sites to evaluate both the performance of our federated learning framework and the effectiveness of our AI models in augmenting clinicians’ abilities to diagnose, select optimal therapeutical interventions and assess potential disease outcomes and treatment response throughout the patient journey in different blood disorders.
To accomplish this, GenoMed4All targets three use cases covering oncological (Myelodysplastic syndromes and Multiple Myeloma) and non-oncological (Sickle Cell Disease) haematological diseases. First, we identified critical unmet clinical needs for each use case, such as diagnosis, risk prediction and treatment decisions. Next, the team focused on the creation of ad hoc protocols for synthetic data generation (so-called virtual patients) to enhance and complement available use case data for model training and simulations, together with the necessary validation protocols to standardise AI testing. Then, for all three, the project has deployed and fine-tuned a series of AI algorithms and models working with tabular data, medical imaging and genomics-based predictions for the early identification of high-risk individuals, prediction and insights on disease development and to aid decision-making in risk stratification.
The first of our use cases focuses on Myelodysplastic syndromes (MDS), a group of bone marrow failure disorders that typically affect the elderly. Patients suffer from blood cytopenia (low blood cell counts) since their bone marrow is no longer able to produce enough healthy blood cells. The disease is also known as a form of blood cancer, and in some patients, it can evolve into Acute Myeloid Leukaemia (AML), which is usually fatal if not treated. For this group of disorders, GenoMed4All has identified a genomic signature that can predict the risk of progression to AML and has also successfully simulated early relapse scenarios using synthetic data.
Similarly, Multiple Myeloma (MM) is a type of bone marrow cancer originating in plasma cells, a type of white blood cell responsible for producing antibodies to fight off infections. In patients with MM, cancerous plasma cells accumulate in the bone marrow and produce abnormal proteins instead, which can lead to decreased blood cell numbers, bone and kidney damage. In this use case, the team at GenoMed4All has effectively combined genomic data with imaging techniques (e.g., PET and CT scans) to increase the prediction accuracy of overall and leukaemia-free survival.
Finally, Sickle Cell Disease (SCD) is a group of hereditary red blood cell disorders. It is a rare, chronic and life-threatening disease. In patients with SCD, red blood cells become C-shaped in resemblance to a sickle. These cells die early and tend to clog the blood flow when going through small blood vessels, so patients usually suffer from low red blood cell counts, infections, acute chest syndrome and strokes. In this context, GenoMed4All has developed a novel algorithm that uses AI to detect silent infarcts (extremely small lesions that are often hard to detect and very easily misinterpreted by neuroradiologists) using available patient MRI data.
GenoMed4All’s legacy: How AI can help patients with a rare disease
Seen under a global lens, rare disease numbers are staggering. Roughly 30 million people are living with a rare disease in the EU, and conservative estimations speak of a total of 300 million worldwide. This is exactly why precision medicine is key: when we shift our focus to the individual, all diseases become unique.
The integrative approaches that GenoMed4All has developed within our three target use cases can be extended to other types of blood disorders, and potentially to other areas of medicine in the future, such as lung cancer, diabetes and neurodegenerative disorders. On this topic, the team is already exploring how to expand the project’s reach to onboard additional clinical sites working on oncological and non-oncological haematological diseases through the ongoing support of ERN-EuroBloodNet and its membership base.
Moreover, the solid technical foundations set by GenoMed4All’s federated platform represent the means through which we can unlock interoperability, data reusability and knowledge transfer across Europe, connecting hospitals and research centres and filling in the gaps between clinicians and researchers to pave the way for precision medicine and a common European Health Data Space (EHDS). As such, principles like robustness, scalability and user experience are non-negotiable. After all, the project intends to serve an ever-growing community by providing tailored AI models, algorithms and real-world use cases to validate them in.
Ultimately, GenoMed4All aims to make a real impact in clinical practice through trustworthy, explainable, secure and reliable AI tools that can support clinical decision-making and usher in a new era of data-driven, privacy-preserving, patient-centric care.
Disclaimer
GenoMed4All has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 101017549.
Please note, this article will also appear in the 23rd edition of our quarterly publication.
Source link
