Introduction Clot composition is variable and may influence the efficacy of mechanical thrombectomy (MT) devices for endovascular treatment of acute ischemic stroke (AIS) from large or medium vessel occlusions. Predictive methods may help the neurointerventionalist in appropriate device selection.

Aim of Study To explore the potential of machine learning to accurately determine clot histological characteristics.

Methods Data from non-contrast CT (NCCT) and CT angiography (CTA) were collected from 10 AIS patients with middle cerebral artery occlusions. Automated clot analysis software (Nicolab, Amsterdam, the Netherlands) was used to evaluate clot perviousness by analyzing clot attenuation increase (CAI) using each patient‘s NCCT and CTA datasets. Predetermined CAI thresholds were applied to classify clots as impervious, semi-perviouse, and pervious. Histological analysis was done on the clot specimens and quantified for correlation with clot perviousness.

Results The software categorized the clot samples as follows: 50% were semi-pervious, 30% were impervious, and 20% were pervious. Statistical analyses did not reveal a significant correlation between histological clot composition and software-determined perviousness. However, fibrin-dominant (>90%) clots were observed in both impervious and pervious groups. Conversely, semi-pervious clots exhibited mixed and red blood cell content, ranging from 24.5% to 51.9%. Notably, an impervious sample was identified to contain mature fibrin, highlighting potential variability within clot composition.

Conclusion Despite the absence of significant correlations, a trend emerged, indicating that clots with extreme perviousness characteristics may be predominantly composed of a single component. Our findings suggest there may be a potential for machine learning to optimize MT.

Disclosure of Interest yes This study was sponsored by Nicolab.