AI has rapidly become one of the most talked-about technologies in medicine, but can it actually make a difference in oncology? Headlines promise algorithms that can detect cancer earlier than clinicians, predict treatment outcomes, or even replace entire diagnostic workflows. At the same time, oncologists, researchers, and product teams are confronted with a very different reality: fragmented data, regulatory barriers, questionable model transparency, and clinical workflows that are far more complex than tech narratives suggest.
The truth sits somewhere in between the hype and skepticism. AI is already transforming parts of oncology, from image-based screening to toxicity prediction and remote symptom monitoring. But many widely circulated assumptions remain inaccurate or overstated, and misunderstanding them can lead to wasted resources, unrealistic product roadmaps, or stalled clinical adoption.
This article separates what’s real, what’s false, and what’s actually coming next for AI in oncology. We’ll explore the capabilities that already work in clinical practice, the misconceptions slowing progress, the data infrastructure required for reliable AI, and the near-term developments that will reshape cancer care. Whether you’re a clinician, researcher, product manager, or digital health innovator, this framework will help you understand where AI can truly add value, and where caution is still essential.
Oncology sits at the intersection of extreme complexity and enormous potential, which is why it is both the hardest and the most promising domain for artificial intelligence. Unlike other specialties, cancer is not one disease but hundreds, each with its own biology, progression patterns, genetic mutations, and environmental influences. Tumor heterogeneity means that two patients with the same diagnosis can respond completely differently to treatment. Add to this the vast range of treatment pathways, like surgery, chemotherapy, immunotherapy, radiation, targeted agents, and the combinatorial complexity quickly surpasses what clinicians or traditional statistical tools can fully track.
AI offers a powerful way to make sense of this complexity. It is uniquely suited to make sense of this complexity because it can integrate and learn from multimodal datasets, including:
But operationalizing AI in oncology remains challenging. Data is deeply siloed across hospitals, EHRs, research centers, imaging archives, and biobanks. Real-world datasets are often incomplete, noisy, or unrepresentative. Regulatory expectations for safety and explainability are far stricter than in other fields. As a result, successful AI in oncology requires not only strong algorithms but robust data infrastructure, rigorous validation, interdisciplinary collaboration, and clinical-grade governance.
Despite the hype surrounding artificial intelligence in cancer care, the real value of AI today lies in a set of focused, clinically grounded capabilities that enhance oncology rather than replace it. Let’s go through the strongest areas
AI-powered image analysis helps radiologists and pathologists identify subtle abnormalities on CT scans, mammograms, MRIs, digital pathology slides, and even dermoscopy images. These tools do not diagnose cancer on their own, but they improve sensitivity, reduce oversight errors, and accelerate triage in high-volume clinical settings.
By analyzing structured EHR data, laboratory markers, tumor genomics, and in some programs even wearable signals, machine-learning models can estimate recurrence risk, forecast treatment toxicity, and identify which patients are likely to require acute care. These predictive insights support personalized follow-up schedules and more efficient resource allocation. See for yourself how we at Thryve provide health risk assessment!
AI performs best when answering narrow, well-defined clinical questions, such as matching a tumor’s molecular profile to targeted therapies or surfacing guideline-concordant options based on a patient’s history. These systems act as decision-support layers rather than autonomous decision-makers, ensuring oncologists retain full clinical authority.
Modern platforms combine patient-reported outcomes with real-world data from wearables to detect early signs of deterioration, such as worsening fatigue, declining mobility, abnormal sleep patterns, or escalating pain. This enables earlier interventions, fewer emergency visits, and safer outpatient management.
AI is already transforming the administrative backbone of oncology. Natural-language processing tools can summarize notes, prepare tumor board packets, streamline documentation, and automate follow-up workflows, allowing clinicians to spend more time with patients instead of paperwork.
In short, today’s AI is strong in pattern recognition, risk estimation, monitoring, and workflow automation, not in making independent treatment decisions.
Alongside legitimate breakthroughs, oncology is filled with exaggerated claims about what AI can achieve today. One of the most persistent myths is that AI can diagnose cancer better than doctors. In reality, AI systems outperform clinicians only in very narrow, highly controlled tasks, such as identifying specific image patterns in clean, high-quality datasets. As soon as real-world variability enters the picture (motion artifacts, comorbidities, rare tumor types, atypical presentations), trained oncologists dramatically outperform any model in holistic diagnostic reasoning.
Another misconception is that AI can instantly personalize treatment. True precision oncology requires the integration of genomics, imaging, pathology, comorbidities, patient preferences, toxicity risks, and longitudinal outcomes. No model today can synthesize all of these reliably without significant human oversight. AI can support personalization, but it cannot independently produce a safe or comprehensive treatment plan.
We also hear claims that language models could replace tumor boards. This is far from reality. Tumor boards require collaborative interpretation of complex medical histories, liability-bearing decisions, and ethical considerations, all grounded in structured clinical data that LLMs do not reliably process or contextualize. At best, LLMs can summarize notes or retrieve guideline excerpts; they cannot lead case deliberations.
A related myth is that “more data automatically means better models.” In oncology, quality matters far more than quantity. A small, well-annotated dataset from a single cancer center often outperforms massive but noisy, heterogeneous datasets. Without standardization, interoperability, and harmonized labeling, additional data can degrade model accuracy rather than improve it.
Finally, wearable devices alone cannot detect cancer, a narrative that resurfaces frequently in media headlines. While wearables provide valuable real-world insights into fatigue, sleep, mobility, or heart rate patterns, no clinical evidence suggests they can directly diagnose or screen for cancer.
The biggest barrier to meaningful AI in oncology isn’t algorithmic capability; it’s data quality. Cancer care generates some of the richest medical data of any specialty, yet much of it is fragmented, inconsistent, and disconnected. AI models cannot learn reliably when biomarkers vary across labs, imaging formats differ between institutions, and patient histories are scattered across incompatible systems. Even within a single hospital, tumor markers, imaging results, genomics, and treatment records may live in separate silos that rarely communicate. Check our blog post on working with fragmented data to find more detailed information!
Most oncology datasets also lack longitudinal continuity. A model may know what happened during diagnosis and treatment, but not how the patient lived months or years later. Did they recover? Did symptoms worsen? Did comorbidities change the picture? Without long-term tracking, AI cannot understand patterns that unfold slowly, such as relapse, toxicity, or survivorship challenges.
Another gap is the absence of behavioral and lifestyle data. Factors like sleep, fatigue, physical activity, and heart rate variability profoundly affect treatment tolerance and recovery, yet they are rarely included in oncology datasets. This omission leaves AI blind to everyday realities that shape patient outcomes.
The future of oncology AI depends on integrating clinical data, labs, imaging, genomics, and continuous real-world data from wearables. Only when these signals are harmonized and standardized can models deliver reliable, clinically meaningful insights.
Ultimately, AI is only as good as the data it receives. Without high-quality, interoperable, continuous data streams, even the most advanced algorithms will fall short.
AI in oncology is moving toward a more pragmatic and clinically grounded future. Instead of grand claims about replacing oncologists, the next wave of progress will come from systems that support clinicians, improve workflows, and fill data gaps that limit today’s models.
Multimodal AI Models: They combine imaging, genomics, lab values, clinical notes, and even wearable data into a unified understanding of each patient. By integrating these data types, AI will be able to improve risk stratification, guide treatment choices, and predict how patients respond to therapies far more accurately than single-modality models.
Treatment Tolerance & Toxicity Prediction: Oncology teams struggle to anticipate who will tolerate chemotherapy, immunotherapy, or radiation and who may experience severe side effects. By analyzing trends in sleep, HRV, physical activity, inflammatory markers, and clinical history, AI could flag vulnerable patients early and enable personalized dosing or supportive interventions.
Relapse detection: Possible through passive monitoring will soon become possible as well. Subtle real-world activity changes, heart rate patterns, and symptom reports often shift weeks before clinical signs appear. AI systems trained on continuous wearable data could prompt earlier follow-ups and prevent avoidable disease progression.
Survivorship Care: Long after treatment ends, patients suffer from fatigue, neuropathy, hormonal changes, sleep disturbances, and mental health strain. AI-driven tools can monitor recovery, identify late effects sooner, and personalize rehabilitation pathways. For example, we collaborated with the Surviouship clinic to improve health outcomes for women who beat gynecological cancer.
Operational AI Improvement: from triaging patients based on urgency, to predicting no-shows, to optimizing the extremely strained scheduling systems inside oncology centers. In population health, AI could refine screening programs by identifying high-risk individuals earlier and allocating resources more efficiently.
When oncology AI finally gains access to integrated clinical, lab, imaging, and real-world signals, the next era of intelligent cancer care will truly begin.
For oncology AI to mature, it needs stable, high-quality, longitudinal data, and this is where Thryve provides the foundational layer. Thryve’s unified biometric models harmonize data from hundreds of wearables and health devices, ensuring that signals like HRV, sleep, activity, and temperature are standardized, comparable, and clinically meaningful across populations.
In oncology research and survivorship programs, continuous real-world data is increasingly essential. Thryve enables secure, high-resolution activity and recovery tracking that complements clinical, lab, and imaging datasets, helping researchers understand fatigue, toxicity, cardiometabolic stress, and overall functional status with far greater depth.
For oncology teams, research groups, and digital therapeutics, the next step is clear: focus on AI where it meaningfully improves detection, prediction, treatment tolerance, and long-term survivorship. If your organization is building oncology AI or integrating real-world data, Thryve provides the infrastructure to help you do it safely, reliably, and at scale. We provide:
Book a demo with Thryve and see how AI can treat real patients, save real lives, and provide real outcomes!
Friedrich Lämmel is CEO of Thryve, the plug & play API to access and understand 24/7 health data from wearables and medical trackers. Prior to Thryve, he built eCommerce platforms with billions of turnover and worked and lived in several countries in Europe and beyond.
