From research labs to clinical practice, artificial intelligence (AI) and health technologies are reshaping how we detect, diagnose and treat cancer. Yet while the potential is clear, the road to their successful implementation is shaped by challenges in data quality, trust, regulation and adoption. How can we best address the critical issues and opportunities at stake?
Executive summary
- Data quality and context matter more than sheer data volume in making AI useful for oncology.
- Clinical trial data is standardized, but routine care data is fragmented and inconsistent. Improving capture is essential.
- Cutting-edge tools like RNA sequencing create new possibilities, but bridging the gap from measurement to actionable decisions requires clarity and trust.
- Adoption of AI varies widely across oncology, with radiology leading and molecular pathology just beginning.
- Rethinking patient consent, data infrastructure and reuse strategies will unlock greater value from healthcare data.
- Practical use cases in clinical practice include streamlining tumor boards, supporting evidence-based decision-making, and automating routine processes.
- Barriers include uneven data quality, regulatory complexity, lack of clarity, and the need for skilled personnel.
- The long-term promise lies in earlier detection, improved efficiency, and better patient outcomes through collaborative adoption.
Why AI in cancer care depends on data quality
A common phrase in AI is that it needs “data, data, data.” But sheer volume is not the answer. What determines the usefulness of AI in oncology is not just how much data we have, but how it is generated, structured and applied. Even a small quality gap in the data – or missing data points – can have outsized effects on outcomes. Poorly contextualized or incomplete records can send algorithms down the wrong path, undermining the very purpose of using AI in patient care.
”The data that goes into any AI is far more important than the model itself. A small quality gap in the data, a small amount of missing data, can play a vital role in changing the course of the model and having a much larger impact than we can think of.
Manish KhatriDirector, Clinical Science Data & AI Lead – Oncology Portfolio, Global Clinical Development, Novartis
In practice, this means paying attention to the full process: how data is captured, the standards guiding its collection, and the purpose for which it was originally gathered. The raw quantity of information is far less important than its reliability and the integrity of the methods behind it.
Clinical oncology data challenges
In clinical trials, data quality is carefully managed through protocols. But in everyday patient care, data is often fragmented, inconsistently recorded, or scattered across systems that cannot easily communicate with one another. In such cases, any AI tool built on top of this foundation risks amplifying uncertainty rather than clarifying it.
”Rubbish in, rubbish out. When we talk about routinely collected data, you need to understand why this data point is here and in which context. Otherwise, you will never get anywhere.
Benjamin KasendaSenior Consultant Medical Oncologist, University Hospital Basel
The key lies in improving how data is captured in routine settings. Without this foundation, even the most sophisticated AI model cannot deliver meaningful or reproducible insights.
From complex data to actionable decisions
The tools available for generating biological data are advancing rapidly. RNA sequencing, single-cell analysis and other technologies can produce extraordinary details about a patient’s cancer. But the challenge remains: how to translate that complexity into decisions that matter for the individual. More data does not automatically mean better decisions – it needs to be contextualized, explained and linked to actionable options.
”We focus on developing AI pipelines that contextualize multi-omics datasets, integrating them with a specific focus on explainable AI to increase trust.
Tim HeinemannSenior Computational Biologist, CSEM
This is where explainability becomes central. If clinicians are to trust AI tools, they must not only deliver accurate outputs but also make it possible to understand why a particular conclusion or recommendation was reached. Transparent, explainable models build confidence and make adoption far more likely.
Adoption of AI in oncology: fast and slow lanes
The uptake of AI varies significantly across medical specialties. Radiology and radiotherapy are ahead of the curve, helped by standardized imaging formats and mature tools for automated image analysis. By contrast, fields such as molecular pathology are only beginning to experiment with AI-driven methods. This uneven progress reflects both technical readiness and the complexity of implementing solutions in different areas of cancer care.
”Cancer is not a single disease but a spectrum. AI will have to adapt to that complexity, which is why progress looks very different depending on the specialty.
Caoimhe Vallely-GilroyDirector, DayOne
Rather than expecting one “super model” to manage every dimension of oncology, the most realistic approach is a suite of specialized tools that serve particular functions –for example, tools for imaging, genomics or decision support – that can progressively be integrated into broader workflows.
Redesigning healthcare data collection for AI
To realize the full promise of AI, healthcare systems will need to rethink how data is collected and reused. Consent processes must allow patients to contribute their data not only for immediate treatment, but also for future research and secondary applications. Hospitals will need infrastructure that enables secure, interoperable and scalable data sharing, while ensuring compliance with ethical and regulatory frameworks.
Anonymization and de-identification techniques will play an important role, allowing valuable datasets to be repurposed without compromising patient privacy. By designing for reuse from the outset, healthcare systems can unlock far greater value from the data already being generated.
Practical opportunities for AI in cancer care
Beyond research, some of the most immediate opportunities for AI lie in easing the burden on clinical teams and supporting decisions. For example:
- Streamlining tumor boards: AI can consolidate scattered information from patient records – such as scans, lab results and clinical notes – into structured summaries that save valuable time in multidisciplinary case discussions.
- Decision support: AI-based systems trained on guidelines and prior patient outcomes can suggest treatment options or highlight comparable cases, accelerating evidence-based decision-making while leaving ultimate responsibility with clinicians.
- Automating routine processes: From preparing clinical documentation to managing longitudinal patient records, AI can take over repetitive tasks and free up physicians to focus on direct patient care.
”Instead of spending five minutes debating guideline-based cases, AI could summarize the options in seconds, allowing clinicians to spend more time focusing on complex cases.
Benjamin KasendaSenior Consultant Medical Oncologist, University Hospital Basel
These kinds of use cases, focused on augmenting rather than replacing human expertise, are likely to gain traction first.
Barriers to adopting AI in healthcare
Despite the progress, several barriers continue to slow adoption. Data availability and quality remain uneven. Many models still lack sufficient performance for clinical use. And the “black box” nature of some AI systems raises concerns about explainability and accountability. On top of this, regulatory pathways are evolving, and healthcare providers face the challenge of investing in infrastructure and skilled personnel to implement solutions safely.
”Fast iteration may be possible in consumer AI, but in healthcare the regulatory guardrails are vital—and they make adoption slower but also safer.
Tim HeinemannSenior Computational Biologist, CSEM
Overcoming these hurdles will require collaboration across startups, hospitals, regulators and pharmaceutical companies. Education will also be vital: clinicians must be trained not just in how to use AI tools, but in how to evaluate them critically and select the right tool for the right context.
The future of AI and healthtech in oncology
Ultimately, the promise of AI in oncology is not only about accelerating discovery, but also about transforming the patient’s journey.
- “The data that humans alone cannot process at scale – AI can help us translate into insights that matter for patients.” — Manish Khatri
- “We are able to measure more and more data, and I look forward to a future where diseases are diagnosed much earlier, even before symptoms appear.” — Tim Heinemann
- “AI will save time – collecting information, preparing data, and making decisions.” — Benjamin Kasenda
- “The critical question is not whether the technology works, but how close it is to being usable and implemented in clinical practice.” — Caoimhe Vallely-Gilroy
These perspectives point to a future where data works harder for patients, caregivers and clinicians alike. Realizing it will take persistence, trust and cross-sector collaboration – but the impact could be profound.
This article was inspired by discussions at DayOne’s Open Mic: Next in Health event on “The role of healthtech and AI in cancer treatment.”
