AI in radiotherapy needs more than technology
Workflow, validation, governance and professional adoption
AI becomes clinically meaningful when imaging intelligence, language intelligence, workflow design, validation, regulatory readiness, governance and professional adoption are organised as one coherent clinical transformation.
The radiotherapy pathway is the starting point
Referral & triage
Simulation & pre-treatment imaging
Consult preparation
Completeness, urgency and missing information.
Consult preparation
Simulation & pre-treatment imaging
Consult preparation
Patient context, prior imaging and open questions.
Simulation & pre-treatment imaging
Simulation & pre-treatment imaging
Simulation & pre-treatment imaging
Image quality, immobilisation and preparation readiness.
Image and planning readiness
Image and planning readiness
Simulation & pre-treatment imaging
Images, patient record data and route requirements ready for planning.
Contouring
Image and planning readiness
Treatment planning
OARs, targets and clinical review logic.
Treatment planning
Image and planning readiness
Treatment planning
Structured intent, treatment orders and plan evaluation.
Pre-treatment QA & release
Pre-treatment QA & release
Pre-treatment QA & release
Clinical, dosimetric and physics checks
Treatment delivery
Pre-treatment QA & release
Pre-treatment QA & release
Delivery, adaptive signals and workflow exceptions.
Follow-up
Pre-treatment QA & release
Follow-up
PROMs, toxicity signals and learning loops.
Three principles for clinical AI transformation
Clinical value
Governed release
Workflow impact
AI must address a real clinical, workflow, safety or capacity problem.
Workflow impact
Governed release
Workflow impact
Every AI output changes work, handovers, review tasks and responsibility.
Governed release
Governed release
Governed release
Clinical use requires validation, training, release decisions and lifecycle monitoring.
Good enough, relevant enough, safe enough
The acceptance chain helps professionals judge whether AI output is clinically acceptable. The key question is not whether something looks different, but whether the difference is clinically relevant and what action should follow.
Clinical release is not the same as go-live
Defined scope
Which pathway, model, version, modality and patient group are in scope?
Validation route
What evidence, testing, acceptance criteria and reviewer feedback support release?
Human oversight
Who verifies what, when, and with which escalation route?
Lifecycle monitoring
Which signals detect drift, unsafe use, user behaviour or workflow degradation?
