Healthcare and Life Sciences
UNSW Sydney Accelerates Stroke Diagnosis in Regional Australia With MONAI
University of New South Wales (UNSW) Sydney—Image Gallery
Objective
Stroke is the second leading cause of death and the most common cause of adult disability in the world. With global stroke cases projected to double by 2050, timely diagnosis and treatment are now more critical than ever. In Australia, the University of New South Wales (UNSW) Sydney Telestroke Service is pioneering a new approach—leveraging AI, MONAI, and NVIDIA GPU-accelerated computing to deliver life-saving insights in real time, particularly for patients in regional and remote communities.
Key Takeaways
- Using MONAI Label and NVIDIA GPUs, NSW Telestroke Service reduced hemorrhage volume analysis time from 20 minutes to seconds.
- Powered by NVIDIA GPU-accelerated MONAI Label with Auto3DSeg, NSW Telestroke Service automated lesion segmentation across 23 regional hospitals statewide.
- With MONAI Label integrated with XNAT and powered by NVIDIA GPUs, NSW Telestroke Service created a unified AI platform for real-time stroke analysis.
Bridging the Gap in Regional Stroke Care
The UNSW team, led by Dr. Ken Butcher, professor of neurology at UNSW Sydney and medical director of the NSW Telestroke Service, provides rapid virtual access to stroke specialists for 23 regional hospitals across New South Wales. These communities, often hours from metropolitan centers, previously lacked ready access to expert stroke diagnosis and care.
“Without brain imaging, there is no such thing as modern stroke medicine,” explains Dr. Butcher. “You can’t even differentiate the two most basic subtypes of stroke—hemorrhagic and ischemic—without a CT or MRI scan.”
This distinction is vital: hemorrhagic strokes (bleeding in the brain) and ischemic strokes (blocked arteries) need vastly different treatments but frequently present with identical symptoms. Rapid, accurate diagnosis—within a crucial therapeutic window—is essential to guide the right care.
Imaging Bottlenecks: From Qualitative to Quantitative
CT scans can be acquired quickly, but quantitative analysis—such as precisely measuring brain hemorrhage volume—has long been limited to research settings. Manual segmentation is time-consuming, labor-intensive, and inconsistent.
“I’ve spent years training students and doctors to manually segment brain images,” Dr. Butcher notes. “It’s painstaking work. Every time someone graduates, I have to start over. There had to be a better way.”
“Thanks to MONAI and NVIDIA, we can now generate accurate hemorrhage volumes in seconds—something that used to take up to 20 minutes, and we’re doing it in real time, during clinical consultations. Currently, we have models for intracranial hemorrhage and diffusion-weighted MRI lesions, with others in development.”
Dr. Ken Butcher
Professor of Neurology at UNSW Sydney and Medical Director of the NSW Telestroke Service
A Breakthrough With MONAI and NVIDIA
That better way came through MONAI Label and Auto3DSeg—core modules of the open-source MONAI toolkit—integrated into a medical imaging viewer and powered by NVIDIA GPUs on AWS enterprise eHealth cloud infrastructure. With support from the Telestroke Project team and NVIDIA technical experts, Dr. Butcher’s team deployed an AI-assisted Telestroke solution that automates segmentation and measurement of CT brain lesions—moving advanced quantitative insights into daily clinical practice.
The workflow is built to be both powerful and efficient. CT images are sent to a MONAI Label server in a secure, cloud-native environment managed with NVIDIA GPU-accelerated computing on AWS. MONAI Label uses these GPUs for AI inference, quickly segmenting stroke lesions and delivering quantitative volume measurements. Results are sent back to the statewide image repository for immediate consultant review; anonymized PDF versions are simultaneously emailed to the entire Telestroke team.
Integration with open-source viewers—including 3D Slicer, OHIF, and ITK Snap—provides rapid visualization and annotation anywhere. MONAI Label’s modular server architecture, combined with XNAT for data management, makes the system flexible for on-premises or cloud deployment and easy to integrate into existing health IT environments. “Thanks to MONAI and NVIDIA, we can now generate accurate hemorrhage volumes in seconds—something that used to take up to 20 minutes,” says Dr. Butcher. “And we’re doing it in real time, during clinical consultations. Currently, we have models for intracranial hemorrhage and diffusion-weighted MRI lesions, with others in development.”
Streamlined AI Deployment for Stroke Imaging
The project team integrated MONAI Label with XNAT, a research PACS system. Images from the statewide repository, including the Telestroke PACS, are pushed to XNAT, automatically triggering MONAI Label for inference. The system generates DICOM segmentation objects, bounding boxes, PNG segmentation overlays, and PDF reports—all centrally stored, easily visualized with the OHIF viewer, and returned to PACS for clinical use.
Any inaccurate AI-generated segmentations are flagged and corrected in 3D Slicer, with corrections fed back for active learning and retraining. MONAI Label’s GPU-accelerated inference and retraining enable continuous improvement and fast workflow cycles.
Currently, UNSW runs two deployed models: one for hemorrhage detection and one for final ischemic infarct volume from MRI. These models are continually refined with new data and annotations. The team is also developing new approaches for contrast-enhanced CT, retinal imaging, and eye tracking for vertigo diagnosis. All approved models are made available publicly via the MONAI Model Zoo, enabling clinicians worldwide to adopt and adapt them.
“We’ve built a feedback loop. If the model gets something wrong, we correct it, retrain it, and redeploy it—all within our clinical infrastructure. That’s the future of AI in medicine.”
Dr. Ken Butcher
Professor of Neurology at UNSW Sydney and Medical Director of the NSW Telestroke Service
Real-Time Decisions, Better Outcomes
The Telestroke Service now delivers real-time volume measurement for hemorrhage across its network. Segmentation and measurement data arrive to the PACS and are securely shared with all relevant specialists. This workflow enables collaborative, state-wide decision-making.
“We review every case as a team at our weekly clinical review meeting,” Dr. Butcher shares. “For the first time, we can objectively assess whether a hemorrhage has grown and how much. The MRI model helps us see exactly how much brain tissue we saved after treatment. It’s not just research—it’s applied clinical care.”
For rural and remote hospitals, the platform is especially transformative—providing timely expertise and critical insights that can mean the difference between recovery and lifelong disability. Clinicians make faster, more confident decisions, improving patient outcomes and reducing pressure on emergency departments.
Stroke brain scan
A Platform for Continuous Learning and Innovation
One of the most innovative aspects of the project is its ability to learn and improve over time. Every AI segmentation is reviewed as part of UNSW’s weekly multidisciplinary meetings. If the model’s output needs correction, this is swiftly done in 3D Slicer and fed back for retraining via MONAI Label’s active learning capability—ensuring the system continuously improves.
“We’ve built a feedback loop,” says Dr. Butcher. “If the model gets something wrong, we correct it, retrain it, and redeploy it—all within our clinical infrastructure. That’s the future of AI in medicine.”
With this approach, the platform is expanding to other imaging domains, including ophthalmology. Continuous annotation, retraining, and validation ensure the models evolve alongside real-world clinical practice.
“This has been career-changing for me. It’s opened up a whole new world of research and clinical application. And it’s been incredibly satisfying to see it make a real difference in patient care. We’re not just building models—we’re building infrastructure that empowers clinicians. … Ultimately, we aim to combine AI-segmented lesions and volumes with additional clinical data to predict therapy response and improve decisions.”
Dr. Ken Butcher
Professor of Neurology at UNSW Sydney and Medical Director of the NSW Telestroke Service
A Clinician-Led Revolution
Dr. Butcher describes the collaboration with NVIDIA and MONAI as transformative: “This has been career-changing for me. It’s opened up a whole new world of research and clinical application. And it’s been incredibly satisfying to see it make a real difference in patient care.” He adds: “We’re not just building models—we’re building infrastructure that empowers clinicians. If you’re a doctor with an idea, you can now bring it to life. Ultimately, we aim to combine AI-segmented lesions and volumes with additional clinical data to predict therapy response and improve decisions.”
The AI-powered Telestroke Service stands as a model for how clinician-led innovation, open-source AI, and accelerated GPU computing can reshape healthcare delivery. As stroke cases rise, UNSW’s pioneering work creates a blueprint for faster, smarter, more equitable care—delivering the right clinical decisions, in real time, wherever patients are.
Learn more about NVIDIA solutions for medical imaging.