I'm a medical physicist at Henry Ford Cancer. I study how language-model agents plan radiation treatment, and where they go wrong.
Henry Ford Cancer, Detroit · Assistant Professor, Michigan State University
Pick a number of beams. Each beam on its own is weak. The dose is only high where they cross, and that's how radiotherapy spares healthy tissue.
Beams
Radiation can't tell healthy tissue from cancer. Most of my work is about getting the dose where it should go, and knowing how much actually got there.
I'm a board-certified medical physicist (MCCPM) at Henry Ford Health in Detroit, where I work in MR-guided adaptive radiotherapy and radiosurgery. I'm also an assistant professor at Michigan State University. Before that, I was a staff physicist at Sunnybrook Health Sciences Centre in Toronto.
My PhD was in radiation dosimetry and detector development. My research now asks a practical question about language models that plan treatments: when a plan goes wrong, does the fix need a better model, or would a clearer setup and a simple check catch it?
Planning a treatment usually takes a lot of trial and error: adjust the objectives, recalculate the dose, check it, and try again. SAGE hands that loop to a language-model agent working in the Eclipse planning system, with a physicist reviewing every plan. We tested it on prostate cases and on 41 radiosurgery cases.
Our simulators let trainees rehearse linac safety and HDR prostate brachytherapy in virtual reality before they do either on real equipment, with real patients.
An MR-linac images the patient during treatment, but its magnetic field changes how dose detectors respond. To measure absorbed dose directly, we built a water calorimeter that works inside the magnet. Calorimetry measures dose through the tiny temperature rise that radiation causes in water.
FLASH delivers the whole dose in a fraction of a second. In animal studies, healthy tissue often tolerates it better than a normal-speed dose, while tumors still respond. I'm involved in FLASH research at Henry Ford. In our preprint with Verdi Vanreusel, Stephen Brown and colleagues, we tested flexible scintillating sheets that show the dose across a surface in real time.
Treatment plans assume scattered radiation does the same biological damage as the main beam, but low-energy scatter can do more. For my PhD, I simulated how much that varies, then built a prototype detector for LET (how densely radiation deposits its energy) from lead-doped plastic scintillators. They glow blue under radiation, a bit like Frodo's sword Sting near orcs.
Scientific Reports. First author.
Medical Physics 53(9), e70654. First author.
Brachytherapy 25(5), 728–735. Senior author.
arXiv preprint 2504.15824. Co-author.
Medical Physics 47(12), 6458–6469. Second author. Sylvia Fedoruk Prize.
Medical Physics 46(9), 4215–4223. First author.
Chapter 7 in Recent Advancements and Applications in Dosimetry, ed. Maria F. Chan. Nova Science Publishers. First author.
Free tools for people who work in radiation oncology. They're quick references and teaching aids, not a substitute for your clinic's protocols.
More than 150 normal-tissue dose limits from QUANTEC, HyTEC, PENTEC and UK SABR, with BED and EQD2 worked out for you.
SBRT and SRS protocols for lung, brain, spine, prostate and liver, in one place.
Hands-on lessons for clinicians learning to use language models safely.
Why stepping back from a radiation source cuts your dose so quickly, shown with moving particles.
Which websites block AI? A tracker that checks every day how sites treat AI crawlers, search bots and assistants, and explains the answer in plain words.
Language models debate each other, switch sides, and get scored by a panel of judges. The demo runs on a free server, so it can take a minute to wake up.
Department of Radiology, College of Human Medicine
MR-guided adaptive therapy and radiosurgery.
IMRT QA and reference dosimetry lead.
CAMPEP-accredited.
Radiation dosimetry and detector development.
Henry Ford Health, for the SAGE radiosurgery research.
American Association of Physicists in Medicine, for virtual and augmented reality research.
Best paper in Canadian medical physics, for MR-linac water calorimetry.
Best student paper, Canadian Radiation Protection Association.
Toronto Metropolitan University.
Awarded to 20 of more than 7,000 applicants across Canada.
If you work on treatment planning, dosimetry or AI in the clinic, or you've found a problem in one of my tools, I'd be glad to hear from you.