University of Notre Dame Australia (PANACEA-HF program) · Us2.ai
zero — Imaging failure rates
Vendor-reportedHealthcare · Diagnostics · Computer vision · Pilot
Diagnostics
The Netherlands Cancer Institute (NKI) · NVIDIA
Vendor-reportedHealthcareNetherlandsPilotPublic sectorComputer vision
Vendor-reported. The customer is named and the numbers are quoted from the source page, but the account comes from the vendor. No independent confirmation.
Reported by nvidia.com (vendor self-report). Checked against the source page on 2026-09-06. 1 of 1 figures below appear on that page word for word.
The problem
Adapt radiotherapy treatment to current anatomy for precision tumor targeting and better treatment planning. Classical CBCT reconstruction methods yield poor image quality.
What was deployed
- Vendor
- NVIDIA
- Products
- NVIDIA AI Enterprise, VMware vSphere, NVIDIA RTX 8000, NVIDIA A100
- Technique
- Computer vision
- Build or buy
- Bought and customised
- Deployment
- On-premise
- Data used
- cone-beam computed tomography (CBCT) scan projection data
- Timeline
- 1 months
What changed
Each row is quoted from the source. Figures we could not find on the page in those words are marked — they are kept, not deleted, so you can judge them.
five minutesTime to reconstruct full volume at 1mm resolution on a single GPUfrom 14 minutes
“A full volume at 1mm resolution took just five minutes to reconstruct on a single GPU, compared to 14 minutes on the bare-metal RTX 8000 machine.”
Quoted word for word from the source
“With AI models and hardware from two or three years ago, we’d have to down-step the resolution to free up memory for training. Now, with the power of current GPUs, we can develop end-to-end systems that reconstruct high-resolution volumes directly from the projection data.”
Difficulties and limits
- Classical CBCT reconstruction methods yield poor image quality compared to dedicated CT scans.
- Training models to reconstruct 3D volumes directly from projection data at clinically relevant resolution is memory-hungry — a single CBCT volume at 1mm resolution occupies about 1GB of memory once gradient information is taken into account.
- Motion tracking of tumors caused by body movement during respiratory cycles is not yet addressed by the current system; researchers note improving this as a possible future direction.
Sources
- nvidia.com — vendor self-report
Similar deployments
Aletha Health · Rapid Innovation
Vendor-reportedHealthcare · Diagnostics · Computer vision · In production
University of Medicine and Pharmacy Hospital, Ho Chi Minh City · Us2.ai
about 10 minutes — diagnosis time
Vendor-reportedHealthcare · Diagnostics · Computer vision · Pilot
HappyPath · Oodle
30 minutes — Render integration setup time
Vendor-reportedTechnology & software · Diagnostics · LLM · In production
MUSC Health · Notable
110000 — monthly digital registrations completed
Vendor-reportedHealthcare · Process automation · Classical ML · Scaled · 2022
Houston Methodist · ActiumHealth
18000000 — staffing capacity created
Vendor-reportedHealthcare · Customer support · LLM · In production