Will AI Replace Oncologists? Planning Is Becoming More Automated, but Cancer Treatment Still Requires Clinical Judgement

Radiotherapy contouring is becoming more automated

AI can outline tumours and organs at risk, reducing the time clinicians spend drawing structures manually.

Contours still need review because anatomy, prior treatment and unusual disease can create errors. An automated boundary is a starting point, not a final prescription.

Treatment planning is an optimisation problem with clinical constraints

Software can search dose distributions more efficiently and compare plans against technical objectives.

Oncologists still decide which compromise is acceptable when tumour control and normal-tissue exposure compete. The priorities depend on disease, prognosis and patient values.

Imaging can support response assessment

AI can measure lesions and compare scans over time, helping identify change consistently.

Treatment response is not always captured by one measurement. Inflammation, treatment effect and mixed disease patterns can complicate interpretation, so clinical context remains essential.

Prognostic models can inform rather than decide

Models can estimate outcomes using tumour, patient and treatment data. This can support conversations about likely benefit.

Predictions describe groups, while the person in front of the oncologist may differ in important ways. Clinicians need to explain uncertainty rather than turning a probability into a promise.

Systemic treatment involves changing evidence

Cancer drugs and biomarkers evolve rapidly. AI can help match patient characteristics with current guidelines and trials.

Oncologists still weigh toxicity, comorbidity, prior treatment and the strength of evidence before recommending therapy.

Toxicity management requires ongoing judgement

Symptoms during radiotherapy or systemic treatment can indicate expected side effects, infection, organ toxicity or disease progression.

AI can support monitoring, but clinicians need to determine which change requires dose adjustment, admission or stopping treatment altogether.

Cancer care is multidisciplinary

Oncologists work with surgeons, radiologists, pathologists, specialist nurses and other professionals. Treatment decisions often emerge from multidisciplinary review.

AI can summarise evidence, but specialists still resolve disagreement and decide which sequence of treatments is feasible.

Conversations about prognosis are deeply human

Patients may need to understand whether treatment is curative, palliative or unlikely to help. The same statistics can lead different people to different choices.

Oncologists need empathy, honesty and awareness of what the patient values. A generated explanation cannot own that relationship.

Post-deployment monitoring is essential

RCR guidance stresses that AI tools need ongoing monitoring after implementation. Performance can change as scanners, populations and workflows change.

Clinical teams must detect drift and know when the tool should be suspended or recalibrated rather than assuming regulatory approval guarantees permanent safety.

The resilient oncologist becomes a clinician who can challenge AI

Planning and administrative tasks will automate further. Durable value lies in treatment selection, toxicity management, multidisciplinary judgement and communication under uncertainty.

AI can increase capacity and consistency, but RCR's current approach keeps clinical expertise and accountability at the centre of cancer care.

Pathology and molecular data increase rather than simplify decisions

Cancer classification now includes histology, mutations, biomarkers and other features that can alter treatment. AI can organise this information and surface relevant evidence.

The oncologist still decides whether the evidence applies to the patient's stage, prior therapy and overall condition. More data increases the need for synthesis.

Clinical trials create options that guidelines may not capture

Patients with uncommon cancers or exhausted standard treatments may be eligible for experimental therapies. Matching systems can identify trials more efficiently.

Oncologists need to explain uncertainty, eligibility, travel, burdens and alternatives so a trial is understood as research rather than guaranteed benefit.

End-of-life decisions resist optimisation

There are times when more treatment is technically possible but unlikely to improve the life the patient wants. Those decisions involve prognosis, symptoms, family context and values.

AI can estimate outcomes, but it cannot decide what trade-off between time, side effects and quality of life is acceptable for another person.

Oncology workload may shift toward supervision of automation

As contouring, documentation and image review automate, oncologists may spend less time on repetitive preparation and more time on complex decisions and patient communication.

That can increase capacity without making expertise less important. The remaining work becomes the part where uncertainty and accountability are highest.

Comorbidities change what treatment is tolerable

Heart, lung, kidney and other conditions can limit otherwise standard cancer treatment. Oncologists need to understand whether a patient can tolerate the proposed dose or combination.

AI can identify risk factors, but the final plan depends on multidisciplinary judgement and how much benefit is expected relative to competing health risks.

Radiotherapy delivery still needs verification

Even a high-quality automated plan has to be transferred accurately to treatment machines and checked against patient positioning and anatomy.

Clinical teams use imaging, physics and quality assurance to ensure the delivered treatment matches the intended plan. Automation strengthens that workflow without removing accountability.

Follow-up separates treatment effect from recurrence

Symptoms and imaging after treatment can be difficult to interpret because healing and toxicity can resemble active disease. Oncologists compare timing, scans and clinical change before deciding whether further treatment is needed.

AI can support classification, but ambiguity remains common in exactly the cases where the decision matters most.

Patient preference can legitimately change the treatment plan

Two people with the same cancer may choose differently when they understand likely benefit, side effects and impact on daily life. Oncologists help make those preferences clinically informed rather than assuming one modelled optimum fits everyone.