A clinician's perspective on the responsible integration of artificial intelligence into cancer care and radiotherapy practice.
Selected news, blog posts and commentary on artificial intelligence in radiation oncology. This section is designed to be updated with recent posts and source links.
GE HealthCare announced FDA 510(k) clearance for MIM Contour ProtegeAI+ 2.0, an AI-enabled auto-contouring tool for radiation therapy planning. This is a commercial regulatory announcement, so departments should still review the public FDA record, local validation needs, contour-editing workflow, and audit requirements before clinical reliance.
Read updateA Cureus review summarised AI applications in prognostic modelling, adaptive radiotherapy monitoring, response assessment and toxicity prediction. The authors emphasised that routine clinical use remains limited by inconsistent methods, limited external validation, workflow integration challenges and the need for clear governance.
Read updateAn arXiv preprint described a workflow-integrated LLM system that generated physician-specific daily summaries and potential clinical trial matches for radiation oncology clinics. The evaluation is preliminary and survey-based, so it should be interpreted as early workflow evidence rather than proof of clinical accuracy or patient benefit.
Read updateSiemens Healthineers announced CE marking for AI contouring capabilities integrated with the Varian Eclipse treatment planning system. This is a vendor announcement and does not replace local commissioning, clinical review of generated contours, or monitoring for site-specific performance and failure modes.
Read updateAI should augment, not replace, clinical judgement. Its safest role in radiation oncology is as a structured assistant for information retrieval, drafting, checking, summarising and prompting clinicians to consider overlooked risks, while preserving expert clinical decision-making and patient-centred care.
I am interested in how digital tools can support safer, clearer and more efficient oncology practice. This does not mean embracing every new technology uncritically. Evidence matters more than hype. The most important question is not simply whether a tool can do something, but whether it should be used in that setting, and how safely.
Digital tools in oncology introduce genuine benefits, including faster literature review, more consistent contouring and structured documentation. They also introduce real risks, including bias, hallucination, over-reliance, opaque decision-making and patient safety concerns. A thoughtful, evidence-based approach is essential.
Seven principles I apply when considering AI tools in clinical and educational contexts.
AI augments; it does not replace clinical judgement
The radiation oncologist remains responsible for diagnosis, treatment intent, planning decisions and patient communication. AI tools assist with specific tasks; they do not assume clinical responsibility.
All AI outputs require expert verification
AI systems can produce plausible but incorrect outputs (“hallucinations”). Every AI-generated result (whether a literature summary, a contour or a treatment plan) must be reviewed and verified by a trained clinician before use.
Safety and transparency are non-negotiable
AI tools used in clinical settings must be transparent about their capabilities and limitations. Clinicians must understand what a tool can and cannot do. Black-box AI used without understanding introduces unacceptable risk.
Patient confidentiality must be protected
Patient data must never be entered into unvalidated AI tools, consumer AI platforms, or systems without appropriate privacy and governance safeguards. Data protection obligations in healthcare are non-negotiable.
Bias and inequity must be considered
AI systems trained on non-representative datasets may perform poorly for underrepresented populations. Clinicians should be aware of the training data behind any AI tool used in clinical practice, particularly in imaging and auto-contouring.
Clinical context remains essential
AI operates on data inputs and patterns. It does not know the patient in front of you, their values, their comorbidities, their social context or their goals of care. Clinical context, gathered through consultation and relationship, cannot be replaced.
Evidence matters more than hype
AI tools in oncology should be evaluated with the same rigour as any other clinical intervention. Peer-reviewed evidence, prospective validation and clinical outcome data matter more than vendor claims or conference demonstrations.
A non-exhaustive overview of current and emerging AI applications across the radiation oncology workflow.
Literature Review & Evidence
AI tools for systematic literature search, summarisation of trial evidence, and structured evidence synthesis. Reduces time burden on busy clinicians without replacing critical appraisal.
Clinical Documentation
AI-assisted drafting of consultation letters, treatment summaries and clinical notes. Requires careful verification for accuracy and clinical appropriateness before use.
Auto-Contouring
Deep learning models for automated organ-at-risk and target volume delineation. Requires physician review and correction. Evidence on accuracy varies by site and model.
Treatment Planning Support
AI-assisted plan optimisation, dose prediction, and plan quality assessment. Potential to improve planning efficiency and consistency, particularly in high-volume settings.
Safety & Quality Assurance
AI tools for anomaly detection, plan review flags, and workflow safety checks. May complement existing QA processes without replacing them.
Patient Education
AI-assisted generation and translation of patient information materials. Useful for improving health literacy and accessibility of information, with clinician oversight.