Neurological signs identified by Ai related to stroke patient recanalization
Résumé
Background and Aims: Thrombus dimensions are critical factors influencing early recanalisation following thrombolysis in stroke patients and the Susceptibility Vessel Sign (SVS), a ferromagnetic artefact, reflects the composition of the thrombus [1] (iron content and therefore red blood cell content). However, their manual estimation becomes challenging for distal occlusions, those located beyond the proximal M2 segment of the middle cerebral artery. We present a novel artificial intelligent (AI) tool that segments distal thrombi from MRI images, automatically calculates thrombus dimensions and investigates their relationship with the AOL recanalisation status one hour after treatment. Successful recanalisation is defined as AOL ⩾2b.
Methods: Our deep learning model combines key imaging modalities: Diffusion Weighted Imaging (DWI) which highlights the lesion and Susceptibility weighted Imaging (SWAN) where thrombi are visible. Trained on 300 annotated cases, the model recurrently segments clots. Enabling fully automated analysis, thrombus length is calculated using the maximum Feret diameter (greatest distance between the two parallel planes) and thrombus width is derived from its orthogonal dimension. Additionally, the SVS is computed as the ratio of the thrombi width to the minimum width (per artery).
Results: Our model achieves 70% thrombus volume accuracy with a mean error of 1.1±0.8mm between actual and estimated lengths. Statistical analysis reveals that thrombi in non-recanalized patients are significantly longer (t-test, p = 0.001). Moreover, a larger SVS is significantly associated with successful recanalization (p = 0.024).
Conclusion: The proposed AI tool offers an automated accurate method for thrombus segmentation and dimension estimation that can offer valuable insights for predicting recanalisation success.
Disclosure of interest: All authors: nothing to disclose