Deep generative computed perfusion-deficit mapping of ischaemic stroke.

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Focal neurological deficits in ischemic stroke are linked to impaired brain perfusion due to vascular occlusion.
• Current methods often rely on lesion location, which may not capture early predictive signals.
• Computed perfusion maps from CT and CT angiography (CTA) offer a potential upstream view.

Purpose of the Study:

• To develop and validate a deep generative model for inferring neural substrates of stroke deficits from perfusion maps.
• To disentangle the distinct patterns of disrupted perfusion and neural dependence in acute ischemic stroke.
• To assess the clinical and scientific value of acute CTA-derived perfusion maps for stroke phenotyping.

Main Methods:

• Computed perfusion maps from CT and CTA data of 1393 acute ischemic stroke patients.
• Application of deep generative perfusion-deficit inference to localize neural substrates of NIHSS sub-scores.
• Analysis focused on disentangling perfusion topology from neural dependence without prior lesion knowledge.

Main Results:

• The deep generative model successfully replicated known lesion-deficit relationships.
• The approach identified novel neural dependents associated with stroke deficits.
• High anatomical fidelity was achieved, suggesting clinical utility of CTA-derived perfusion maps.

Conclusions:

• Acute CTA-derived computed perfusion maps, analyzed with deep generative inference, offer significant value in stroke phenotyping.
• This imaging-based approach can reveal functional-anatomical relationships within the critical pre-interventional window.
• The method holds promise for powering expressive models in acute ischemic stroke research and clinical practice.