Abstract Number: Esoc2026a1090 Highlighting Changes in Vessel Patency on Dsa Improves Emboli Detection and Reperfusion Assessment During Endovascular Thrombectomy

Abstract Background and aims Digital subtraction angiography (DSA) is used for guiding endovascular thrombectomy (EVT) in acute ischemic stroke. Accurate comparison of pre- and post-EVT DSA sequences is critical for assessing reperfusion and detecting complications such as embolization in new territory (ENT). Because of vessel overprojection, manual interpretation of the images during the procedure by the interventionalist is error-prone and time-consuming. We aimed to evaluate whether automated registration and visualization of changes in vessel patency can improve ENT detection and reperfusion assessment. Methods From the MR CLEAN Registry, 40 pre- and post-EVT DSA pairs were registered using a keypoint-based approach (FAST/BRIEF), combined with a similarity transformation estimation. Visually succesful image-pair registrations were retained for analysis. On these pairs, an automated arterial segmentation method (CAVE, 10.1016/j.compmedimag.2024.102392) was applied to extract arterial lumen segmentations. To assess clinical utility, we conducted a reader study with two experienced neurointerventionalists after highlighting pre- and post-EVT arterial differences, using in-house developed software (Figure 1-2). Each interventionalist graded extended thrombolysis in cerebral infarction (eTICI) scores and determined ENT presence in two sessions with two weeks in between: without overlays, and with arterial-change overlays. Results In the reader study, inter-rater agreement for eTICI scoring improved from κ = 0.704 to κ = 0.804 (P = 0.0784) when overlays were provided (Figure 3). ENT detection increased from 3 to 5 and 4 to 5 cases for the two radiologists, respectively. Conclusions Automated registration of cerebral DSA sequences facilitates visualization of arterial changes and may enhance consistency in reperfusion grading and ENT detection, potentially improving patient outcomes. Conflict of interest Marten Leenders: nothing to disclose. Frank te Nijenhuis: nothing to disclose. Danny Ruijters: nothing to disclose. Sandra Cornelissen: nothing to disclose. Frans Vos: nothing to disclose. Ruisheng Su: nothing to disclose. Theo van Walsum: nothing to disclose. Figure 1 - belongs to Methods Figure 2 - belongs to Results Figure 3 - belongs to Conclusions