Using computational fluid mechanics and Artificial Intelligence for Risk Assessment and Guidance of Stroke Neuroradiological Interventions

Project reference: SIE_04_21  
First supervisor: Jorge Cardoso
Second supervisor: Thomas Booth 

Start date:  February 2022 or October 2022

Project summary:  The project aims to investigate methods to predict and evaluate intra-operative risks of neurovascular repair towards supporting surgical planning and decision-making through. the use of computational fluid mechanics and AI.

Stroke management currently lacks quantitative and statistical information as to minimise risks of complications. Stenting and coiling procedures also require predictive markers of hemodynamic changes in the neurovascular network prior to device deployment.

 

A predictive and risk-aware framework for surgical downstream effects would inform the surgeon and greatly support decision-making. With this view, this project aims to investigate the use of computational fluid mechanical models, enhanced by artificial intelligence, and leveraging multi-modal imaging and annotated angiographies, to predict the risks of clinical interventions. Novel computational fluid mechanic methods, such as Iso-Geometric Analysis (IGA), will be used due to their near-real-time computational complexity.

Limitations in the IGA model (turbulence, pulsatile flow, etc) will be mitigated through the use of AI-based model enhancement, allowing a fast and accurate understanding of physio-pathological mechanisms and of the organ-device interactions during surgery.  The ultimate goal of the project is to develop a risk-minimising predictive framework to optimise patient-specific neurosurgical pathways, with translational applications in supporting intra-operative guidance, motion-planning, simulation, navigation and augmented visualisation.

 
Project description:

Neurovascular diseases, such as stroke, are the biggest source of long-term neurological disability in first world countries. In an acute setting, it is necessary to assess the risk of a neurovascular event, and to promptly decide how best to intervene (e.g. thrombectomy vs thrombolysis in brain ischaemia).

Also, recent stenting and coiling procedures for stenotic arteries and for saccular aneurysms are being developed for major brain vessels, thanks to device miniaturisation and advanced intra-operative imaging modalities. Yet, these novel procedures are currently adopted to avoid ischaemia, rupture and haemorrhage on greedy evaluations (i.e. on the estimation of the luminal restriction, alternatively of the aneurysm size), irrespective of predictive markers, e.g. the local and global hemodynamic changes, in the patient-specific neurovascular network.

The surgical decision-making process mainly rely on the clinical experience, and such process is currently ill-informed regarding the quantitative and statistical evaluation of the risks associated to the procedure with the patient – i.e. how, and to which extent the chosen treatment/procedure is going to impact on the specific neurovascular and neurological functionality during intervention and surgical post-op.

 

The aim of this project is to predict and evaluate intra-operative and long-term risks of neurovascular repair procedures by means of fluid mechanical modelling enhanced by AI.

Methods such as Iso-Geometric Analysis (IGA) will be used for the computational modelling of the subject-specific neurovascular anatomy in a time-constrained setting, allowing the study of certain physio-pathological mechanisms (e.g. interactions and induced alterations during surgery).

 

Physical computational models (i.e. IGA) would characterise functional physio-pathological behaviour of the neurovascular network from its geometry and bio-mechanic characteristics, as well as the organ-device interactions, i.e. from the early catheter motion, to the device deployment and ultimately to the follow-up, with high-to-low-rank computational equivalents. The theoretical limitations of IGA models will be mitigated either by improving the solvers or through the use of AI for physics approximation [Chu et al. arxiv:705.01425].

After an accurate simulation of neurovascular flow and clinical intervention effects, statistical and predictive AI models will be used to address the inference of surgical downstream effects, e.g. thrombus removal, re-canalisation, stenting, and coiling procedures in interventional neuroradiology. In particular, AI models could account for estimating and classifying the restored localised and territorial blood supply in the brain upon measurements and simulations, as well as predicting early and long-term auto-regulation mechanisms impacting the neurological follow-up. Surgical risk factors and intra-operative downstream effects will be modelled over a large sample of surgical outcomes, clinically relevant biomarkers, as well as quantitative and surrogate measures on short- and long-term prediction horizons.

 

As a summary, with a large-scale dataset of multi-modal, high-resolution brain images and angiographies comprising clinical annotations of patients undergone neurovascular surgical procedures, this project design and develop a predictive framework able to optimise the patient-specific surgical pathway, minimising risks of intra-operative and long-term complications, with translational applications to intra-operative guidance, motion-planning, trajectory optimisation, simulation and augmented navigation.

Cardoso project image.png