01Input · imaging
Start from routine imaging
Models are built from the scans patients already have. CT is our starting point; time-resolved CT and other modalities add motion and function. No extra acquisition, no dedicated protocol.
Technology
Every model starts from one patient's own anatomy.
We combine machine learning and physics. AI reconstructs the anatomy from imaging; physical laws describe how blood flows and how tissues and devices deform within it; AI again makes those physics fast enough to use outside a supercomputer.
02AI · anatomy
Reconstruct the anatomy
Neural networks turn the images into a 3D computational model of the anatomy: a surface and a volume mesh that keep the patient's own shape and are ready for measurement and simulation. The user reviews the result before anything is computed on it.
03Morphology
Assess the morphology
From the 3D model we describe the anatomy in quantitative terms: dimensions, curvature, volumes, how it moves over the cardiac cycle and how it compares with other patients. Which descriptors matter depends on the anatomy and on the clinical question.
04Physics · simulation
Simulate what images cannot show
Physics-based models compute what a scan cannot measure: blood flow, pressure and wall shear stress, the mechanics of the vessel wall, or the effect of a procedure or a device on the patient's anatomy. AI-built reduced-order models make these simulations fast enough for a workstation. Results are model estimates, and their reliability depends on the inputs.
- Blood flow
- Wall biomechanics
- Procedures and devices
What is personalised
Measured, assumed, estimated.
A model built on a patient's geometry is not automatically a full model of that patient's physiology. We say which quantities come from the images and which depend on assumptions.
- From the patient's images
- Anatomy, dimensions and shape and, with time-resolved imaging, motion.
- Entered by the user or assumed
- Boundary conditions such as inflow and outflow, tissue and blood properties, the device or procedure being studied.
- Estimated by the model
- Flow, pressure, wall stresses, and the response of the anatomy to a procedure.
- Decided by the clinician
- Interpretation, and every clinical decision.
Physics and AI
Why simulation needs AI to leave the lab.
High-fidelity flow and structural simulations take hours on dedicated hardware. That is fine for a study, not for a workstation in a hospital.
Segmentation
Neural networks reconstruct the anatomy. They run on the local computer, and the user reviews their output.
High-fidelity models
Established flow and structural solvers, with mesh morphing, generate reference simulations on many anatomies.
Reduced-order models
Those reference simulations train compact models that approximate the result for a new geometry in a fraction of the time.
Applications
One product today. A wider research programme.
VenusX is where the technology is already a usable product. The same methods are being developed for other questions; they are research, not products.
Product · VenusX
Thoracic aorta from CT
Sizing: segmentation, centerline, diameters and report, available for research evaluation. Flow: model-based hemodynamics, in development.
VenusX- Research
Aneurysm growth from shape
Shape features beyond diameter, extracted from imaging, to study how fast an ascending aneurysm grows.
- Research
Procedure planning
Reduced-order models to support endovascular planning and intra-operative navigation.
- Research
Wall and valve mechanics
Fluid–structure models of vessel walls and heart valves, including heart motion and growth over time.
- Research
Virtual patient populations
Sets of realistic 3D anatomies to test devices in silico before physical prototyping.
Engineering services · for device manufacturers
Test your device in silico, on real anatomies.
The team behind VenusX runs computational projects for medical device manufacturers, researchers and clinicians: from patient-specific anatomies to virtual deployment and flow.
- 01
AI-powered anatomical segmentation
Automatic extraction of anatomical structures from CT and MRI.
- 02
Virtual patient populations
Sets of realistic 3D anatomies for in-silico trials.
- 03
Device design, optimisation and testing
Simulate and refine a device before physical prototyping.
- 04
Patient-specific simulations
Flow and structural models built on individual anatomies.
- 05
Predictive models
AI and physics-based models of risk and disease progression.
Patents and IP
Methods we own.
The core of our modelling, from reconstructing the anatomy to making physics fast, is developed in-house and protected.
- Patent applications filed on computational modelling and AI methods.
- 2
- VenusX and the models behind it are designed, built and maintained by our own team.
- In-house
Quality and security
How we build software.
- Developed under LivGemini's quality management system, certified to ISO 13485, following the IEC 62304 software life cycle.
- Automated unit, integration and system tests, with traceability from requirements to verification.
- Releases come with a software bill of materials, vulnerability scans and static analysis.
- Processing runs on the local computer. No network connection is opened during an analysis, verified in system testing.
Discuss the technology with our team.
For research groups and companies interested in patient-specific modelling, in VenusX or beyond it.