Will AI replace medical dosimetrists? The work is shifting, not vanishing. Planning systems already produce dose distributions in minutes, and auto-contouring tools draw first-pass structures for review. A person still shapes the plan around the case, checks it against the prescription, and carries responsibility when it goes to the treatment machine. That mix of fast math and human accountability is what the score on this page reflects.
Why the plan still needs a planner
Dosimetry is a small, high-stakes job. The Bureau of Labor Statistics counted about 3,410 medical dosimetrists in the United States, with median pay of $147,470 (BLS, 2025) and employment projected to grow about 5% from 2025 to 2035. In a field that size, every plan matters, and every error is visible.
Much of the day is calculation: computing dose distributions, working out monitor units, and testing beam arrangements against the oncologist’s prescription. Software has handled that arithmetic for years, and newer auto-planning tools handle more of the optimization too. The judgment sits around the math. Which trade-off do you accept when a target overlaps the spinal cord? Is this plan deliverable on the machine this patient is scheduled on? Does the setup hold if the patient cannot hold that position for twenty minutes?
The other anchor is review. Dosimetrists check contours, run plan quality assurance, document plan parameters, and sit with the radiation oncologist and medical physicist at simulation and plan approval. Those are accountability tasks. A clinic can let a model propose a plan, but a named person has to say the plan is safe to treat.
What AI does, what it helps with, what stays with people
The tasks our data puts in the “AI does it” group are the computational ones: generating dose calculations and optimizing beam weights within set constraints, and producing first-pass structure contours from imaging. Together those account for 0% of task time on this page. Our Coverage Score Method explains how that share is measured; the headline coverage figure for this job is 28 out of 100.
A larger block of the work is shared. Checking and editing auto-generated contours, comparing competing plans, and running routine chart and plan quality assurance all go faster with software, but each still ends in a human decision. That assisted group covers 62% of task time. The shift described by people in radiation oncology is real: less manual drawing, more inspection of what the tool drew.
Then there is the part that stays with people: 38% of task time. That includes working through a case at simulation with the oncologist and physicist, final plan review and sign-off, explaining constraints and compromises to the clinical team, and adapting a plan when anatomy, equipment, or the patient’s tolerance changes mid-course. Only a small slice of the job is physical, so robotics barely touches it either way.
What the evidence actually shows
Our evidence grade for head-to-head quality here is D. In plain terms, there is no direct, published test of an AI system against working medical dosimetrists on the same clinical caseload, so we publish no parity number for this job. Plenty of studies look at auto-contouring and auto-planning in radiation oncology, but they measure agreement with a reference, not whether a system can carry a planner’s full role.
What would settle it is specific: a multi-site study that takes real referrals, has AI and credentialed dosimetrists plan the same cases, and then has blinded oncologists and physicists score the plans for deliverability and approval. It would also need to report how often AI plans were edited before treatment, and how often automated chart checking caught errors a person missed, and missed errors a person caught. Until that exists, the honest reading is uncertainty about quality, not confidence either way.
Good to know: a tool that matches a reference contour in a study has not been shown to replace the person who signs the plan.
When the picture could change
Most likely between 2041 and 2056 (8 in 10 of our scenarios). The Replacement Year Method sets out how that window is built, and the chart above shows the spread.
Two things could pull it earlier. First, vendor auto-planning shipped as the default inside treatment planning systems, so the fast route becomes the standard route rather than an option. Second, staffing pressure: with a workforce this small, a clinic short one planner has a strong reason to lean on automation for routine sites like breast and prostate.
Two things hold it back. Regulatory and professional sign-off requires a qualified person to take responsibility for each plan, and that structure changes slowly. And case variety keeps biting: re-planning, unusual anatomy, prior radiation, and machine-specific limits are exactly where generic models do worst. You can compare how those forces land across similar roles on Compare Two Jobs, and read the scoring approach in full on our methodology page.
How to stay needed in medical dosimetry
Lean into the tasks that are hardest to hand over. Own plan review and quality assurance, including the judgment calls on auto-generated contours. Be the person at simulation who translates clinical intent into deliverable constraints. Take the complex re-plans and the adaptive cases that routine automation handles badly.
Two skills pay off. One is practical fluency with auto-planning and auto-contouring tools, including where they fail and how to document an override. The other is clinical communication: explaining a dose trade-off clearly to an oncologist, a physicist, and a therapist at the machine.
Radiation Therapists work the closest adjacent role at the delivery end. For imaging-side paths, see our pages on radiologic technologists and nuclear medicine technologists: Radiologic Technologists and Technicians and Nuclear Medicine Technologists. You can also see how the whole group scores on the health technologists family page, the healthcare sector page, or in our list of jobs that most need people.