moRAVIS Lab

Research

Three method streams, one clinical question

Everything below feeds the same endpoint: identifying the vulnerable vascular bed before it causes a stroke, using imaging that has already been acquired for standard care.

Overview

The lab is deliberately narrow. We do not chase every stroke question; we build one measurement pipeline properly and test it hard. Image analytics extracts the signal, computer vision makes the extraction reproducible at scale, and simulation tells us whether what we measured is physically what we think it is. Clinical trials and a prospective imaging database keep the whole thing honest.

Methods

  • Density-to-dwell-time reconstruction
  • Scanner and protocol normalisation
  • Partial-volume correction at the vessel wall
  • Stasis mapping across the bifurcation
01

Image Analytics

CT image post-processing

Recovering quantitative haemodynamic information from CT acquisitions that were never designed to carry it.

A single-phase CTA is conventionally read as an anatomical snapshot. It is not. Contrast arrival, washout and residual opacification across a bifurcation encode how long blood dwells in each part of the lumen. Our post-processing stack reconstructs that dwell-time field from the density distribution, corrects for injection protocol and cardiac output, and produces a per-voxel stasis map that is comparable between patients and between scanners. The hard parts are normalisation and partial-volume correction near the wall, which is exactly where the clinically interesting flow lives. [PLACEHOLDER - add pipeline detail, validation cohort and reproducibility figures]

Projects in this stream

  • Quantifying stasis in the carotid bifurcation from routine CTA

    active

    The flagship project. A reconstruction method that turns a standard single-phase CTA into a map of contrast dwell time, and a prospective test of whether that map predicts stroke.

    More detail +

    This is the work the lab was built around. We show that the spatial distribution of contrast density across a carotid bifurcation is not noise but a readout of local residence time, and that regions of prolonged stasis co-locate with the sites where intraluminal thrombus actually forms. The method requires no additional radiation, no additional contrast, and no change to the acquisition protocol, which is the entire point: it can be applied retrospectively to every CTA already sitting in a hospital archive. Current work is the prospective validation cohort. [PLACEHOLDER - add cohort size, primary endpoint and timeline]

Methods

  • Automated carotid bifurcation segmentation
  • Cross-subject anatomical registration
  • Plaque, web and thrombus phenotyping
  • Test-retest reliability analysis
02

Computer Vision on Imaging

Segmentation, registration, and phenotyping at scale

Making the measurement reproducible enough that it can be run on thousands of scans without a human in the loop.

A method that needs an expert to trace the carotid will never reach clinic. We train segmentation and landmark models to isolate the bifurcation, establish a common coordinate frame across patients so that measurements are anatomically comparable, and automatically phenotype plaque, webs and intraluminal thrombus. The models are evaluated against consensus expert reads, and just as importantly against their own test-retest behaviour on repeat scans of the same patient. [PLACEHOLDER - add architecture, training data description and performance]

Projects in this stream

  • Non-atherosclerotic carotid disease: webs, dissection and floating thrombus

    active

    Imaging phenotyping of carotid pathology that the stenosis percentage completely fails to describe.

    More detail +

    [PLACEHOLDER] Carotid webs and floating thrombi cause stroke at minimal measured stenosis, which makes them the cleanest possible test case for a haemodynamic rather than anatomical marker. Replace this paragraph with your own description.

Methods

  • Patient-specific CFD (velocity, WSS, residence time)
  • Pulsatile waveform modelling
  • 3D-printed flow phantoms
  • Image-versus-simulation agreement
03

Mathematical Simulation

CFD and physical phantoms

Establishing that what we measure in the image is the physics we believe it is.

Every imaging-derived quantity needs ground truth. We build patient-specific computational fluid dynamics models from the segmented geometry, solve for velocity, wall shear stress and residence time under physiologically realistic waveforms, and compare the simulated stasis field against the one recovered from the image. In parallel, 3D-printed patient-specific bifurcations run on a programmable pulsatile pump and are scanned on the clinical CT, which closes the loop between known flow, acquired image, and reconstructed measurement. [PLACEHOLDER - add solver, mesh strategy and phantom validation results]

Projects in this stream

  • Patient-specific flow phantoms for CT haemodynamics

    active

    3D-printed carotid bifurcations on a pulsatile rig, scanned on the clinical scanner, to establish ground truth for image-derived stasis.

    More detail +

    [PLACEHOLDER] Geometry is taken from a patient CTA, printed, and perfused with a blood-mimicking fluid under a programmable waveform. Because the imposed flow is known exactly, the phantom gives us the one thing patient data never can: a true value to compare the reconstruction against. Replace this paragraph with your own description of the rig, materials and protocols.

Methods

  • Prospective outcome-linked imaging registry
  • Investigator-initiated study design
  • Adjudicated endpoint committees
  • Multi-centre data sharing agreements
04

Clinical Trials & Imaging Database

Prospective validation

A prospectively built, outcome-linked imaging database, and the trials that run on top of it.

Retrospective signal is cheap and frequently wrong. The lab maintains a prospective imaging database linking de-identified CTA and follow-up imaging to adjudicated clinical outcomes, which is what allows any candidate marker to be tested as a predictor rather than a correlate. On top of it sit investigator-initiated studies in acutely symptomatic carotid disease, non-atherosclerotic carotid pathology, and recurrent stroke risk stratification. [PLACEHOLDER - add REB numbers, enrolment figures and active protocols]

Projects in this stream

  • Management equipoise in acutely symptomatic carotid stenosis

    active

    Characterising how physicians actually decide on imaging and revascularisation for the hot carotid, and where the evidence runs out.

    More detail +

    [PLACEHOLDER] Acutely symptomatic carotid stenosis sits in a genuine evidence gap, and practice varies enormously between centres and between individual physicians at the same centre. This programme documents that variation, identifies the specific decision points where clinicians report equipoise, and uses those points to design the trials that would actually resolve them. Replace this paragraph with your own description.

How the lab runs

How a study moves through the lab

A question arrives from clinic. We retrospectively test it against the imaging database. If the signal survives, we build the phantom and the CFD model to establish that it is real physics and not a reconstruction artefact. Then, and only then, it becomes a prospective protocol.

Imaging database
De-identified CTA, CT perfusion and follow-up imaging linked to clinical outcome, built prospectively under REB approval. [PLACEHOLDER - describe scale and governance]
Compute
GPU workstation and cluster time for segmentation, registration and CFD solves. [PLACEHOLDER]
Flow phantom rig
Patient-specific 3D-printed carotid bifurcations on a programmable pulsatile pump, scanned on the clinical CT. [PLACEHOLDER]