Leveraging artificial intelligence to strengthen cancer care in Belgium: a policy brief from the Belgian Europe’s Beating Cancer Plan Mirror Group

Abstract

Artificial intelligence (AI) is transforming health care, especially oncology, by enhancing early cancer detection and screening, diagnosis, treatment, patient monitoring, and research. Despite strong research activity, AI adoption in clinical practice in Belgium remains limited due to resource constraints, regulatory complexity, and fragmented governance. Belgium should act decisively now to ensure that it does not miss the opportunity to lead in responsible and effective AI integration in health care. Building on the European Union’s AI Continent Action Plan, now is the time to define a coordinated national strategy, one that can turn AI’s potential into real benefits for patients and health care systems across the country.

Highlights

• •AI adoption by Belgian cancer clinics faces major hurdles.
• •A stakeholder group has formulated policy recommendations to overcome these hurdles.
• •The recommendations are aligned with the EU’s AI Continent Action Plan.
• •A coordinated national AI strategy and regulatory sandboxes are needed.
• •AI training needs medical reporting to be structured, standardized, and synoptic.

Introduction

Artificial intelligence (AI) tools are being developed for a large spectrum of use case scenarios in clinical practice and are increasingly viewed as a key driver in modernizing the health care landscape, offering improved efficiency, enhanced patient outcomes, and potential cost savings. In the field of oncology, AI is being set up to play a vital role, offering innovative perspectives for early cancer detection and screening, diagnosis, treatment, patient monitoring, and research.

Radiology is currently the champion in AI,^1,2 where it is already used in daily practice. Indeed, AI’s utility is particularly evident in medical imaging, where physicians’ tasks involve segmentation, pattern detection, and classification. These tasks are time consuming for radiologists and, due to their repetitive nature, are prime candidates for AI-powered decision tools.2, 3, 4 The same is true for pathologists who can use AI tools for the accurate and validated identification of features or regions of histological slides.

By enhancing diagnostic accuracy and efficiency, oncological AI tools support health care professionals in interpreting health care data, spotting urgent cases, and as such reducing patient wait time. They also play a pivotal role in developing and tailoring personalized treatments. For example, in molecular imaging and targeted radionuclide therapies, AI-driven dosimetry and outcome modeling are helping refine patient selection and optimize treatment efficacy while minimizing toxicity. But the role of AI as catalyst for change in oncology does not stop with these innovations as medical devices. The rapid advancements of generative AI, in particular large language models (LLMs), also have unique implications and many applications in health care, especially for reasoning tasks. LLMs could soon change how medical reports are generated, as well as the form and content of these reports. AI could thus substantially transform clinical notes and data management workflows in oncology and in patient care in general. The risk is the uncontrolled clinical use of publicly available LLMs by health care professionals. Such use raises several concerns, in particular regarding patient data privacy and confidentiality. To address these, some institutions have started deploying their own LLMs, for example, Chopper at Cliniques Universitaires Saint-Luc.

While AI has matured in specific domains, as previously mentioned, especially in radiology (i.e. neurology, chest, musculoskeletal, and mammography),² successful clinical use is still held back. In Europe, ∼60%-80% of AI activity is still in the testing, pilot, or research phase. Only ∼20%-40% is implemented in clinical use (mostly limited tools for stroke triage, lung nodules, mammography, and fracture detection).⁵ This number may be even lower in Belgium, prompting discussions about the causes behind this slow uptake.

Methodology

The current status of AI in health, in particular in oncology, future needs, and priorities are presented in this brief based on the diverse experiences of members of the Belgian Europe’s Beating Cancer Plan (EBCP) Mirror Group⁶ thematic working group (TWG) on AI, as well as current evidence. The Belgian EBCP Mirror Group was established with the support of the Ministry of Health to facilitate the participation of Belgian stakeholders in EBCP projects. It has >400 members and is structured into several TWGs. The Belgian EBCP TWG on AI currently consists of 50 members, with representatives from hospitals, universities, cancer care organizations, patient advocacy, and representative organizations, as well as public institutes and industry. Online group meetings were held to bring together these experts and an online survey was conducted to identify priorities and gaps in the application of AI in oncology. The outcomes of the meetings and the survey have been synthetized by the TWG lead and members of the Cancer Centre’s Implementation, Monitoring, and Evaluation Team.

Belgium’s position in the European Union landscape

In April 2025, the European Commission presented its new AI Continent Action Plan, setting out a strategic agenda to accelerate the adoption and development of AI throughout the European Union (EU), with health care recognized as a priority sector.^7,8 By aligning regulatory measures with investment strategies, focusing on both regulations and information technology (IT) development, the EU aims to create a balanced environment where AI in health care can thrive responsibly. The plan is structured around the following five key pillars:

• •Computing power and infrastructure
• •Data for AI
• •Strategic AI innovation and adoption
• •AI skills and talent
• •Regulatory simplification.

Computing power and infrastructure (pillar 1)

The optimal use of AI tools has huge potential in health care but requires large IT infrastructures and large collections of secure, high-quality, and interoperable health data to be shared, which often comes with several obstacles. In this context, the European Health Data Space (EHDS),⁹ the Data Governance Act,¹⁰ the Data Act,¹¹ and the AI Act¹² Regulations, together with Belgium’s newly established Health Data Agency (HDA),¹³ represent important steps forward. Belgium, ready to host an AI Factory Antenna, has nevertheless recently expressed to Europe its interest in hosting an AI-optimized supercomputer (AI Factory), which could provide not only legal, regulatory, and ethical guidance [e.g. General Data Protection Regulation (GDPR)¹⁴ and Medical Device Regulation (MDR)¹⁵ compliance] but also the necessary infrastructure for AI and data innovation. The health and life sciences sectors have been defined as strategic areas for the AI Factory Antenna, which would include coordination with EU data initiatives such as the implementation of data spaces (e.g. 1+MG¹⁶ for genomics data), as well as relevant government data repositories, to enable secure AI training on decentralized health records.

Belgium is also involved in multiple European research consortia as follows:

• •CANDLE,¹⁷ an EU project that will provide an avenue toward a successful and highly desired data transformation in European cancer research and serve as a catalyst for the UNCAN.eu¹⁸ and European Cancer Patient Digital Centre¹⁹ platforms by advancing the development of a national cancer data node.
• •SHAIPED,²⁰ an EU project addressing the challenges of training, testing, and validating AI-based medical devices through real-world health data access.
• •Belgium plays also a strategic role in diverse European research initiatives such as EUCAIM,²¹ ODELIA,²² and Bigpicture,²³ which focus on imaging data and digital technologies such as multiomics AI and high-performance computing to combat cancer.
• •In addition, there is increasing interest from Belgian partners in methods for synthetic data generation as a GDPR-proof alternative for ‘real’ data (e.g. projects such as SYNTHIA²⁴ and PHASE IV AI²⁵).

Data for AI (pillar 2)

At the national level, hospitals are now expanding their data centers to manage the exponential growth of health data and ensure its quality. Sharing clinical documents for diagnosis, treatment, and coordination of care is enabled through the hub architecture of the Belgian eHealth Platform, allowing health care professionals within different regions to securely locate and retrieve patient information.

Projects for secondary data use are being deployed and an important achievement in this respect is the completion of the data capability projects by the Ministry of Health that have led to the creation of data platforms linked to (networks of) hospitals, such as the Federated Health Innovation Network,²⁶ Levilo Health, and the Institute of Analytics for Health.²⁷ These projects are strengthening Belgium’s data interoperability and analytics capacity. To comply with GDPR, these initiatives often aim for federated approaches for secondary data usage, which are built on the adoption of key components in health data standardization such as Systematized Nomenclature of Medicine Clinical Terms (SNOMED-CT)²⁸ for electronic health records, sometimes in combination with Fast Healthcare Interoperability Resources (FHIR)²⁹ messaging and the common data model Observational Medical Outcomes Partnership Common Data Model (OMOP-CDM).³⁰ Furthermore, solutions for data sharing across different medical professionals and patients have also been developed by the Be-SafeShare project.³¹

Strategic AI innovation and adoption, AI skills and talent, and regulatory simplification (pillars 3-5)

While AI tools should reach clinical practice rapidly, their successful adoption requires more than just technology alone. The implementation of these AI tools still depends on tackling essential challenges that go beyond research and the AI algorithms themselves. These challenges can be grouped into five main categories—medical, technical, organizational, financial, and regulatory (Table 1). The interplay (and overlap of) between different regulatory frameworks (MDR,¹⁵ AI Act,¹² Data Act,¹¹ etc.) and their interpretation at European and national levels complicate deployment. Moreover, Belgium currently lacks a national AI regulatory sandbox that would allow real-world testing of AI tools in clinical settings. Establishing such a sandbox—particularly for oncology and advanced therapies such as radioligand treatment—could accelerate innovation while maintaining patient safety and regulatory compliance.

Table 1.

Key challenges relating to the implementation of AI into clinical practice in Belgium

Category	Key challenges
Medical	Lack of clinical awareness, scientific validation, and legal/ethical doubts
Technical	PACS/IT integration, underfunded IT departments, cloud shift, and interoperability
Organizational	Slow procurement, lack of AI champions, conservative medical culture, and low public engagement
Financial	No reimbursement, and budget cuts and inflation
Regulatory	Complexity of MDR compliance, including evidence requirements and conformity assessment; and insufficient harmonized guidance on how MDR, AI Act, and Data Act interact

AI, artificial intelligence; IT, information technology; MDR, Medical Device Regulation; PACS, Picture Archiving and Communication System.

Policy recommendations

Despite strong research activity and growing awareness of AI’s potential in oncology, resource limitations, fragmented governance, and overlapping regulatory frameworks remain the main obstacles to AI adoption in clinical practice in Belgium. Currently, many hospitals in Belgium are still forced to implement cost-cutting measures. Since AI adoption in clinical routine will be primarily driven by economic factors (cost saving or risk minimizing), it is essential, together with top management support, to evaluate the cost–benefit ratio and develop smart investment pathways in AI that could ensure long-term sustainability and measurable clinical impact.

While Flanders, Wallonia, and Brussels have regional AI strategies (cf. Flanders AI,³² DigitalWallonia4.ai,³³ MedReSyst,³⁴ and FARI³⁵), Belgium lacks a unified national AI strategy or centralized governance body to ensure alignment and coherence across regions. This body should ensure coordinated decision making on AI adoption, regulation, reimbursement, data access, and integration with national digital health infrastructure. Today, many hospitals are taking initiatives themselves. However, these introduce many concerns regarding validation, maintenance, version control, security updates, and computational infrastructure. AI implementation in the health care sector requires more. It needs real top-down guidance, structural support, and a long-term vision. While national efforts offer support for research development, research alone is not sufficient; implementation efforts should build on the experience of existing users and validated pilot projects and scale up what already works in the Belgian context. AI needs to be developed in a deliberate, goal-driven, and patient-centered manner, ensuring its effective, sustainable use in clinical practice, rather than letting AI initiatives emerge randomly or only driven by technology.

Building on the five pillars of the European AI Action Plan,⁷ it is also important to identify Belgium’s priorities for the health care sector for both primary and secondary data usage. In particular, the ‘data for AI’ and ‘regulatory’ pillars (pillars 2 and 5) are closely linked to the implementation of the EHDS regulation. With the EHDS, Belgium must establish clear frameworks at the national level to ensure that this new legal basis facilitates access to high-quality and interoperable health data rather than creating additional barriers. It is essential to define the structure of the HDA Bodies within Belgium. While regional perspectives will undoubtedly influence the debate, models based on data type or category such as imaging, genomics, and molecular therapy data may offer clearer benefits and foster innovation.

Complementing these regulatory measures for the use and exchange of health data, it is imperative to implement structured, standardized, and synoptic reporting, as well as links to standardized systems such as SNOMED²⁸ (level 5 reporting). Investing in high-quality and interoperable data not only improves care delivery and reduces administrative burden but also strengthens interoperability and applicability of AI solutions across institutions and systems. In addition, comprehensive and high-quality datasets are a critical prerequisite for training and validating reliable AI models, especially for innovative therapies such as radiopharmaceutical and radioligand treatments, where image-based dosimetry and patient-specific modeling are essential. In this regard, targeted training on health care data for both primary and secondary usage, and on AI applications must be urgently provided not only to the public and health care professionals but also to private researchers, data scientists, and bioinformaticians to ensure a smooth, cross-sectorial connection between research and health care (cf. pillar 4).

The deployment of regulatory sandboxes is also essential from a practical standpoint. The European AI Act¹² explicitly encourages the establishment of regulatory sandboxes. Belgium could take a leading role here by taking the initiative to set up such environments. Well-designed sandboxes would accelerate both innovation and the clinical implementation of AI in clinical practice, as they allow AI technologies to be tested within safe, controlled, and ethically supervised environments. Currently, no national AI regulatory sandbox is operational and it remains unclear how these will be implemented, which represents an opportunity for Belgium to define a clear model and governance structure, including stakeholder roles and mechanisms for collaboration among these stakeholders. A key component of such a sandbox should be a national collaborative AI data platform that links diagnostic data, imaging, biomarkers, molecular data, and therapy outcomes across the Belgian health care system, ensuring broad and equitable benefits for all. Complementing this platform, the sandbox should also maintain a curated repository of specialized AI models that are securely evaluated, benchmarked, and deployed for specific purposes. Implementing such a platform and repository within a sandbox is essential for building local expertise, conducting local validations with real-world samples, and defining quality requirements for the health care system.

Immediate actions, such as the implementation of sandboxes, should therefore be initiated without delay, alongside medium-term to long-term objectives, including reimbursement models, validation procedures, and national infrastructure. Time is becoming a critical factor in this transformation. Delaying the implementation of foundational measures risks widening the gap between technological advances and clinical practice.

Successful AI clinical tools implementation represents a delicate transition that will also require specific institutional governance and support. The use of AI needs to become a structural component of the clinical work in the institution, rather than an option left to some individuals or services. A clear, documented procedure must be implemented with the input of all professionals involved (e.g. clinicians, data scientists, clinical informaticians, IT professionals, legal experts, and ethics representatives) to ensure appropriate monitoring, define clinical accountability, and address patient information and consent. Finally, it also has to be seen as a dynamic and continuous process rather than a one-time deployment. Ongoing validation, recalibration, and maintenance of AI tools will indeed be needed.

AI implementation in the health care sector will require resources but they could improve the efficiency, accuracy, treatment quality, and resource allocation, ultimately having a huge impact on hospitals and the services they deliver to the patients in our country. Additionally, AI implementation has the potential to generate medium-term to long-term cost savings, free up health care system resources, and alleviate the pressure faced by the system.