The 2025 ESMO AI & Digital Oncology Congress, held in Berlin from November 12 to 14, highlighted the accelerating role of artificial intelligence across the oncology care continuum. Although imaging and pathology remain the most established fields for AI adoption, this year’s programming revealed a decisive shift toward workflow-integrated AI that enhances clinical operations, supports trial efficiency, and addresses the realities of patient monitoring. 

Across three days of sessions, side-room conversations, and industry demonstrations, one theme was clear. AI is evolving from experimental add-on technology into a practical clinical teammate, but scaling its impact will require robust validation, seamless integration, and a sustained focus on clinician trust.

‍

AI across the care continuum

A major focus of this year’s event was the growing maturity of AI in diagnosis, point-of-care decision support, and prognosis. These domains have become central to real-world deployment. Clinicians repeatedly emphasized the challenges they face daily, including overwhelming data volumes, administrative overload, fragmented systems, and the increasing complexity of trial participation. 

Their perspective was consistent and candid. They need AI that reduces burden, improves efficiency, and supports clinical judgment without creating additional work. Many expressed strong interest in AI systems that streamline documentation, surface real-time insights during patient encounters, and simplify trial-related workflows such as eligibility interpretation, patient recruitment, medical record review, and concomitant medication evaluation. 

Medable showcases, "Harnessing AI Agents to Overcome the Complexity of Oncology Trials"

Rana Khan, Senior Technical Consultant, presented "Harnessing AI Agents to Overcome the Complexity of Oncology Trials" on Thursday, November 13th. He explained to attendees how despite major advances in technology and collaboration, clinical development is still progressing too slowly. Drug approvals have plateaued at roughly 50 per year, which is far below what is needed to address thousands of unmet diseases. The challenge is even greater in oncology, where trials run longer, generate far more data, and require frequent and costly protocol amendments.

His argument noted that real acceleration l only come from intelligent, interconnected AI agents, as these can amplify human teams rather than add to their workload. Medable’s Agent Studio provides a platform for these agents to analyze data, take cross system actions, support proactive site management, and operate with adjustable levels of human oversight. By focusing on accuracy, consistency, and full traceability, Agent Studio helps organizations address true operational bottlenecks while meeting regulatory expectations. This creates a scalable path toward more autonomous clinical development that reduces administrative friction, empowers researchers, and accelerates the delivery of life saving therapies without compromising quality.

‍

Imaging and pathology AI advances

Breast cancer screening continued to serve as a showcase for AI’s impact, supported by compelling evidence from several high-profile studies. The MASAI Trial reported a 29 percent increase in cancer detection, a 44 percent reduction in radiologist workload, a substantial improvement in identifying invasive and aggressive subtypes, and only a modest rise in recall rates. 

These findings were complemented by the EDITH Trial in the United Kingdom, the most ambitious multi-center and multi-vendor prospective AI breast screening study to date. Led by Professor Fiona Gilbert, EDITH is designed to address generalizability and performance across diverse technologies and clinical settings. Despite growing enthusiasm, speakers noted that workflow efficiencies alone cannot justify widespread deployment. 

Economic modeling suggests that AI-supported screening programs must achieve at least a five percent reduction in interval cancers in order to be cost-effective. Even so, several Nordic countries have already begun implementing AI-supported screening programs, which underscores the momentum building in imaging-based oncology.

AI-enabled modernization of clinical trials

Another central theme of the congress was the use of AI to modernize clinical trial design and execution. The complexity of oncology trials has increased dramatically over the past two decades, yet eligibility criteria often remain unchanged and are frequently copied forward without meaningful revision. AI tools showcased at the event demonstrated the potential to improve critical components of trial development, including the summarization of scientific literature, the refinement of overly restrictive eligibility criteria through real-world data, and the prediction of trial success probabilities. 

Digital twins and virtual patient simulations attracted considerable attention because they allow researchers to test alternative scenarios without exposing real participants to unnecessary risk. Tools such as PathFinder illustrated how intelligent modeling can expand eligible populations while improving statistical power.

Synthetic control arms, supported through real-world datasets and algorithmic matching, are also gaining traction as an alternative to traditional comparators. Despite the availability of more than a dozen commercial patient-trial matching platforms, real-world enrollment remains below ten percent, which suggests that operational challenges rather than technological limitations remain the primary barrier to improving recruitment.

The rise of digital patient avatars

One company, TwinEdge Biosciences, introduced one of the more ambitious applications discussed at the congress. Their presentation highlighted the structural inefficiencies of early-stage drug development, including the low approval rate for clinical oncology programs, the billions of dollars spent each year on unsuccessful trials, and the large number of patients who contribute to studies that do not ultimately succeed. 

TwinEdge’s digital patient avatar platform, built from more than 10,000 individualized tumor and patient profiles, aims to bridge the gap between limited early-phase data and the robust population-level insights needed for late-stage development. By creating virtual patient populations that reflect real heterogeneity, the technology offers a path toward more informed decision-making in drug development.

‍Agentic AI shifting from concept to clinical use

A growing number of sessions explored the emergence of agentic AI systems capable of autonomously supporting clinical teams. These discussions examined practical and regulatory considerations, including the need for CE-marked solutions (products that have been certified as compliant with European Union standards for health, safety, and environmental protection), clear guardrails on autonomy, and new standards for validating systems that continuously learn and adapt. 

Clinicians expressed both enthusiasm and healthy skepticism, noting that the promise of agentic systems to reduce cognitive and administrative burden is substantial, but adoption will depend on explainability, reliability, and sustained education. 

The prevailing sentiment was that agentic AI represents the next major frontier in digital oncology and that it has the potential to materially reshape clinical workflows once trust and regulation advance further.

Trust, safety, and validation at the forefront

Throughout the conference, the importance of trustworthiness emerged as a recurring theme. Technology developers emphasized the necessity of eliminating hallucinations, ensuring high-quality training data, preventing overfitting, and embedding safeguards that avoid the pitfalls associated with poorly curated information. 

The consensus across academia and industry was that the future of clinical AI depends as much on rigorous governance frameworks and transparent validation processes as on improvements in the underlying models.

Industry leadership perspectives

Bristol Myers Squibb offered a strategic perspective on AI’s expanding role in oncology. The company highlighted opportunities for AI to enhance prevention through early risk detection, to elevate diagnostic accuracy by reducing false negatives, and to strengthen treatment decision-making through multi-agent clinical support systems. 

Broader commentary suggested that medical diagnosis could become one of the first domains in which highly advanced AI systems achieve transformative capability, particularly through collaborations between biotechnology organizations and major technology firms.

Advances in digital home-based cancer care

The congress also drew attention to the growing importance of tools that support patients outside the clinic. Careology’s presentation demonstrated how digital oncology platforms can reduce the “blind spot” between clinical visits by enabling patients to track symptoms, mood, and medication adherence while passively sharing vital information through wearable devices. 

Clinicians gain access to real-time dashboards that highlight emerging concerns, facilitate earlier intervention, and reduce avoidable hospitalizations. This patient-centered model reflects a broader trend toward extending oncology care into the home, supported by technology that blends consumer-friendly interfaces with clinical-grade monitoring.

A shift from hype to practical integration

Across formal sessions, panel discussions, and informal conversations in hallways and exhibition areas, a consistent narrative emerged. The future of AI in oncology will be defined not only by novel algorithms but also by the ability to integrate these systems into real workflows. Clinicians no longer seek abstract AI-driven insights or experimental prototypes. 

Instead, they want solutions that reduce administrative load, save time, and increase their capacity to focus on patient care. Achieving this vision will require coordinated efforts across clinical, regulatory, and technological domains, supported by transparent evaluation and clear communication.

Conclusion - Integration as the next chapter of oncology AI

The ESMO AI & Digital Oncology Congress 2025 demonstrated that AI’s role in cancer care is expanding rapidly, with momentum building across imaging, trial design, digital patient support, and point-of-care decision-making. The defining characteristic of the next phase will be integration rather than invention. Organizations that succeed will be those that close the gap between AI’s potential and its practical implementation, delivering solutions that meet clinicians where they are and support the full complexity of oncology practice. In this evolving landscape, AI is positioned not to replace the clinician but to elevate clinical care by illuminating blind spots, reducing friction, and strengthening the continuity of treatment.

‍