HINT.GENT Summit – December 5, 2025
Towards responsible Artificial Intelligence (AI) in Healthcare: Integration, Evaluation and Adoption
Abstract
The HINT.GENT Summit, held on 5 December 2025 in Ghent, presented the evolving role of artificial intelligence (AI) in healthcare from the ethical, economic, legal, technical and clinical perspectives. Presentations by UGent experts emphasized the need for responsible, transparent and people‑centered AI integration in education, research and patient care. Ethical analyses argued for a pragmatic focus on present unmet clinical needs, cautioning against polarized utopian or dystopian narratives. Economic evaluations highlighted that while many AI applications appear cost‑effective, current evidence is limited by methodological weaknesses and a lack of real‑world validation. Legal perspectives situated healthcare AI within the EU AI Act’s high‑risk classification, underscoring the need for robust documentation, human oversight and alignment with GDPR and medical‑device regulations.
Technical sessions demonstrated operational AI systems across clinical and administrative workflows. These included large‑language‑model‑based discharge‑summary generation, AI medical scribes integrated into electronic health records, AI‑supported maxillofacial surgical planning using 3D printing and intraoperative CT, iterative AI‑assisted medical‑image annotation platforms, and orthopedic decision‑support systems leveraging remote monitoring, digital twins and robotics. In oncology, digital pathology tools and multimodal deep‑learning approaches showed improved diagnostic consistency and treatment‑response prediction, with emerging capabilities to infer gene expression from histology alone.
During the panel discussion, participants identified key adoption challenges, including insufficient data interoperability, lack of standardized documentation, unclear liability frameworks, and the need for institutional governance structures enabling meaningful human oversight. Trust, transparency and co‑production with clinicians were highlighted as essential determinants of successful uptake.
The summit concluded that AI is transitioning from experimental technology to operational clinical tool, yet its safe and equitable adoption requires rigorous evaluation, ethical stewardship, economic justification and alignment with legal frameworks. While AI can augment clinical expertise across multiple domains, its effectiveness ultimately depends on thoughtful integration, robust oversight and sustained interdisciplinary collaboration.
- Introduction
Artificial intelligence has rapidly transitioned from experimental tool to operational technology across multiple domains of healthcare. HINT.GENT, the Health Intelligence Network Ghent, aims to advance responsible, transparent and human‑centered AI in healthcare by fostering collaboration between clinicians, researchers, engineers and policymakers. The HINT.GENT Summit (5 December 2025) was organized to present the latest insights on AI’s ethical, legal, economic and clinical dimensions. The program opened with an ethical examination of AI’s impact on trust, bias and medical decision‑making, followed by a health‑economic analysis which assessed the value and cost‑effectiveness of AI‑supported interventions. The presentation showcased the plans and vision for the new BRAIN AI hub at UGent Campus Zwijnaarde, highlighting more opportunities for innovation and collaboration in artificial intelligence research. An accessible explanation of the EU AI Act was presented, outlining its high‑risk classification and oversight requirements for healthcare applications. The summit further showcased clinical applications across documentation automation, surgery, imaging, orthopedics and digital pathology, illustrating the breadth of AI integration. A final panel discussion synthesized these perspectives, identifying key challenges such as interoperability, accountability and adoption barriers, in alignment with HINT.GENT’s mission to guide responsible AI implementation in healthcare.
- Conceptual and Strategic Frameworks for AI in Healthcare
2.1 Ghent University’s Vision and Responsible AI Principles
In the opening keynote by prof. Petra De Sutter, the Rector of Ghent University, emphasized Ghent University’s commitment to integrating AI responsibly across education, research, and healthcare delivery. She characterized AI as an augmentative technology that accelerates knowledge generation and supports clinical decision-making, provided that its deployment adheres to transparency, accessibility and human-centeredness. Initiatives such as AI literacy training, faculty guidelines, interdisciplinary AI networks and the emerging BRAIN AI Tech Hub were presented as different initiatives of Ghent Universities aim to implement AI responsibly in education and research.
2.2. Ethical Frameworks and the Need for Pragmatism
Ethical considerations were further developed by Prof. Ignaas Devisch, who argued against deterministic narratives framing AI as either salvation or threat. Instead, a pragmatic, needs-based approach was proposed, emphasizing that ethical discourse should focus on current clinical challenges such as communication difficulties, administrative burden, and diagnostic ambiguity rather than on hypothetical future scenarios. Ethical responsibility was underscored as involving not only the scrutiny of AI, but also the design of systems capable of resisting bias, acknowledging uncertainty, and operating transparently.
2.3. Economic Evaluation and Health Technology Assessment
Prof. Lieven Annemans examined AI from the perspective of health economics, asserting that AI technologies must conform to established principles of value for money. Drawing on a systematic review of AI cost-effectiveness studies, it was reported that while many systems appear cost-saving or cost-effective, evidence is frequently weakened by methodological gaps, including incomplete cost accounting and insufficient evaluation of diagnostic accuracy. It was argued that any AI intended for clinical use should be assessed according to established Health Technology Assessment (HTA) criteria spanning safety, effectiveness, cost, ethics, societal impact, and organizational implications. Real‑world evidence was highlighted as the decisive requirement for widespread adoption.
2.4. The BRAIN: UGent’s new AI hub
Prof. Rik Van de Walle presented the BRAIN AI Tech Hub as a large‑scale, interdisciplinary infrastructure designed to consolidate AI research and innovation within Ghent University. The initiative includes a physical and collaborative ecosystem intended to unite hundreds of researchers, companies and clinicians to accelerate clinically relevant, application‑driven AI development. Ongoing research domains including DNA and protein analysis, sleep‑apnea detection, cataract‑surgery support, digital‑twin modelling, uncertainty quantification, wearable monitoring and robotics were presented as examples of how integrated expertise can drive high‑impact progress. The BRAIN Hub aims to be a strategic foundation for positioning Flanders as a European leader in responsible and translational health‑AI innovation.
2.5 Legal Governance Under the EU AI Act
Prof. Tom Goffin presented a legal analysis situating healthcare AI within the EU AI Act. It was clarified that the AI Act functions primarily as a product-safety regulation, with most healthcare applications expected to fall into the “high-risk” category. This classification obliges developers and deployers to implement robust documentation, risk management and human oversight. However, it was underscored that compliance with the AI Act does not guarantee responsible clinical use. Additional considerations such as liability, data protection, patient rights, and professional autonomy must be governed through institutional policies, training, and clinical governance structures.
- Applied AI: Clinical and Administrative Implementations
3.1. Large Language Models for Medical Documentation
Different clinical applications of AI were presented at the summit, with most applications covering large language models (LLMs) in documentation workflows. The FRAIT project at UZ Gent demonstrated a systematic approach to generating clinician-specific discharge summaries using LLMs. Through a structured evaluation involving 31 clinicians across three hospitals, the study found that while most summaries were clinically acceptable, measurable shortcomings related to completeness and hallucinations persisted. Clinicians expressed that full integration within existing electronic health records (EHRs) is essential for adoption, highlighting integration rather than model capability as the primary determinant of usability.
3.2. AI-Based Scribing and Workflow Automation
The companies Cavel and Squire presented operational AI scribing systems capable of converting spoken dialogue into structured clinical documentation. These systems support multiple provider types and interface directly with common Belgian EHR platforms. Both Cavel and Squire acknowledged ongoing challenges, including hallucinations, lack of standardization in medical note‑taking, and risks of professional deskilling, underscoring the continued necessity of human oversight.
3.3. Surgical Applications: Digital Twins, 3D Planning, and Intraoperative Verification
A multidisciplinary surgical team led by prof. Renaat Coopman provided detailed accounts of AI-enabled surgical planning in complex maxillofacial oncology. Using Materialise’s planning software and 3D printing capabilities, the team developed patient‑specific cutting guides and virtual reconstructions. This workflow enabled them to assemble mandibular reconstructions from fibular grafts directly at the donor site, significantly reducing ischemia times. Moreover, intraoperative CT allowed immediate verification of alignment, enabling sub‑millimeter precision. Early integration of AI-based segmentation suggests future potential for full or partial automation of preoperative planning stages.
3.4. AI-Augmented Imaging and Annotation Platforms
Medannot demonstrated the utility of a cloud-based annotation and model‑training environment designed for clinical users. By enabling iterative training on small datasets and integrating annotation workflows with hospital PACS systems, the platform accelerates the development of bespoke segmentation models. Applications were shown ranging from diagnostic assistance to intraoperative augmented‑reality visualizations. This work illustrates how AI can be co‑developed with clinicians to increase accuracy and usability over time.
3.5. Orthopedic Decision Support, Rehabilitation Monitoring and Robotics
The MoveUp platform exemplified how longitudinal patient-reported data, wearable sensor data and AI‑based intake tools can inform personalized orthopedic care. The system supports preoperative triage, postoperative rehabilitation and long-term outcome monitoring. Clinical examples showed how AI-derived indices help distinguish patients who benefit from conservative therapy from those who require surgery. Additional demonstrations included CT‑based digital twins for knee arthroplasty, robotic assistance and augmented-reality guidance, illustrating the integration of AI across the operative continuum.
3.6. Precision Oncology Through Digital Pathology and Multimodal AI
The summit concluded its technical sessions with advanced applications in digital pathology and computational oncology. Prof. Piet Ost presented evidence demonstrating that AI systems can reduce diagnostic variability in prostate cancer grading and improve prognostic stratification beyond traditional Gleason scores. The clinical value of FDA‑authorized systems, such as Page Prostate and ArteraAI, was emphasized particularly in predicting the likely benefit of endocrine therapies.
Prof. Kathleen Marchal extended these findings by showing that deep learning models can infer gene expression and even spatial transcriptomic patterns directly from histopathology images. These results suggest a future in which multimodal AI systems can provide molecular-level insights without requiring invasive or costly laboratory procedures, thereby enabling more accessible precision medicine.
4. Adoption Challenges and System-Level Considerations
The summit concluded with a panel discussion on how to support responsible adoption of AI in healthcare. The panel was moderated by prof. Tom Braekeleirs and included Vincent Seynhaeve (Pfizer), Marijke Schroos (Microsoft), Karlien Hollanders (FPS and patient representative), Simon Malfait (UZ Gent), and dr. Dipak Kalra (iHD). The concluding panel discussion identified common obstacles to AI adoption, including fragmented data infrastructure, insufficient interoperability and limited access to high-quality datasets. Panelists emphasized that trust must be cultivated through demonstrable transparency, documented bias analyses, explicability and adherence to human‑oversight requirements.
Participants also noted that successful AI deployment depends upon clinical co-production, clear evidence of value and alignment with organizational priorities. Regulatory uncertainty and evolving liability considerations further necessitate institutional governance structures capable of supporting AI supervision. A recurring theme was that while AI is expected to impact nearly all clinical roles, human expertise, judgment and critical thinking remain central.
5. Discussion
The HINT.GENT Summit revealed a healthcare ecosystem in which AI adoption is accelerating but remains uneven across domains, shaped by sociotechnical, regulatory and organizational factors. Several cross‑cutting themes emerged that merit scientific reflection.
First, the summit’s presentations underscored that AI’s technical maturity is ahead of its infrastructural maturity. Many systems such as AI scribes, digital pathology models and surgical planning platforms function effectively in controlled or well‑integrated environments. However, widespread deployment is constrained by interoperability deficits, variable data quality and the absence of harmonized documentation standards. Clinicians repeatedly indicated that integration with Electronic Health Records (EHRs) and existing workflows is a more significant adoption barrier than model performance.
Second, the proceedings illustrated a persistent gap between AI’s regulatory compliance and its clinical responsibility. The EU AI Act provides a foundation for product safety but does not fully address liability, professional autonomy or the complexities of clinical decision-making. This mismatch places substantial pressure on healthcare institutions to define governance structures that ensure meaningful human oversight, allocate supervision time and establish accountability mechanisms. Without institutional frameworks that operationalize these responsibilities, high-risk AI could exacerbate rather than mitigate clinical risks.
Third, the economic and ethical discussions together highlighted that value creation in AI is multidimensional. While some AI systems demonstrate cost savings or improved efficiency, such outcomes are contingent on complementary organizational changes, including redesigned workflows and cross‑disciplinary collaboration. Ethical considerations such as transparency, explainability and trust also emerged as prerequisites for adoption. Trust is not solely a function of accuracy; it depends on clinicians’ ability to scrutinize AI outputs, understand uncertainty and override recommendations when necessary.
Fourth, clinical case studies emphasized that AI augments expertise but does not replace it. In surgical planning, oncology and orthopedics, AI systems provide valuable insights, accelerate planning or reveal otherwise inaccessible molecular information. Yet expert supervision remains essential, not only to validate outputs but also to contextualize them within broader clinical reasoning. Several speakers stressed that AI-induced deskilling is a real risk if clinicians become over-reliant on automated systems. This concern underscores the need for continuous professional development and the integration of AI literacy within medical education.
Finally, the panel discussion highlighted a tension between rapid innovation and cautious adoption, reflecting broader societal debates about technological change. While many participants advocated accelerating deployment for use cases with demonstrable clinical and operational value, others warned that premature implementation without robust oversight could undermine patient trust and safety. The discussions suggested that the path forward requires balanced governance: fostering innovation while safeguarding patient rights, clinician autonomy and data integrity.
6. Conclusion
The HINT.GENT Summit demonstrated that artificial intelligence is entering a phase of practical deployment across diverse areas of healthcare, including documentation automation, surgical planning, diagnostic pathology and personalized oncology. Although substantial technical progress was evident, responsible adoption requires sustained attention to ethical, economic and legal frameworks, as well as institutional governance and clinician engagement.
Based on the insights of the summit of 2025, HINT.GENT aims to continue to facilitate multidisciplinary health‑AI innovation and work on future priorities: enhancing interoperability, generating real‑world evidence, strengthening human oversight mechanisms and supporting co‑production between clinicians and technologists. As several speakers noted, AI should be understood not as a replacement for human expertise but as a set of tools whose effectiveness and safety depend fundamentally on thoughtful implementation and critical use.
This event was supported by:
