Thrombus imaging characteristics are associated with treatment success and functional outcomes in stroke patients. However, assessing these characteristics based on manual annotations is labor intensive and subject to observer bias. Therefore, we aimed to create an automated pipeline for consistent and fast full thrombus segmentation. We used multi-center, multi-scanner datasets of anterior circulation stroke patients with baseline NCCT and CTA for training (n = 228) and testing (n = 100). We first found the occlusion location using StrokeViewer LVO and created a bounding box around it. Subsequently, we trained dual modality U-Net based convolutional neural networks (CNNs) to segment the thrombus inside this bounding box. We experimented with: (1) U-Net with two input channels for NCCT and CTA, and U-Nets with two encoders where (2) concatenate, (3) add, and (4) weighted-sum operators were used for feature fusion. Furthermore, we proposed a dynamic bounding box algorithm to adjust the bounding box. The dynamic bounding box algorithm reduces the missed cases but does not improve Dice. The two-encoder U-Net with a weighted-sum feature fusion shows the best performance (surface Dice 0.78, Dice 0.62, and 4% missed cases). Final segmentation results have high spatial accuracies and can therefore be used to determine thrombus characteristics and potentially benefit radiologists in clinical practice.
(1) Background: For acute ischemic strokes caused by large vessel occlusion, manually assessed thrombus volume and perviousness have been associated with treatment outcomes. However, the manual assessment of these characteristics is time-consuming and subject to inter-observer bias. Alternatively, a recently introduced fully automated deep learning-based algorithm can be used to consistently estimate full thrombus characteristics. Here, we exploratively assess the value of these novel biomarkers in terms of their association with stroke outcomes. (2) Methods: We studied two applications of automated full thrombus characterization as follows: one in a randomized trial, MR CLEAN-NO IV (n = 314), and another in a Dutch nationwide registry, MR CLEAN Registry (n = 1839). We used an automatic pipeline to determine the thrombus volume, perviousness, density, and heterogeneity. We assessed their relationship with the functional outcome defined as the modified Rankin Scale (mRS) at 90 days and two technical success measures as follows: successful final reperfusion, which is defined as an eTICI score of 2b-3, and successful first-pass reperfusion (FPS). (3) Results: Higher perviousness was significantly related to a better mRS in both MR CLEAN-NO IV and the MR CLEAN Registry. A lower thrombus volume and lower heterogeneity were only significantly related to better mRS scores in the MR CLEAN Registry. Only lower thrombus heterogeneity was significantly related to technical success; it was significantly related to a higher chance of FPS in the MR CLEAN-NO IV trial (OR = 0.55, 95% CI: 0.31–0.98) and successful reperfusion in the MR CLEAN Registry (OR = 0.88, 95% CI: 0.78–0.99). (4) Conclusions: Thrombus characteristics derived from automatic entire thrombus segmentations are significantly related to stroke outcomes.
1 November 2025 in Stroke: Vascular and Interventional Neurology
BACKGROUND: In acute ischemic stroke, the infarct core and hypoperfused regions are key indicators for assessing and prognosticating patients. They are typically estimated with computed tomography perfusion (CTP). However, because noncontrast CT and CT angiography are more widely available, we trained a neural network to estimate the ischemic lesion from noncontrast CT and CT angiography scans. METHODS: In this retrospective study, an nnU‐Net model was trained to estimate infarcted and hypoperfused regions from noncontrast CT and CT angiography using reference standards from a commercial CTP software (StrokeViewer). We included data from 859 patients for training and 137 for testing. We used data from the Collaboration for New Treatments of Acute Stroke consortium, including MR CLEAN (Multicenter Randomized Controlled Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands)‐NO‐IV, MR CLEAN‐MED, MR CLEAN‐LATE, and MR CLEAN‐Registry, and a local cohort. In addition to testing our model against StrokeViewer, we also compared our results with 3 other commercial CTP software packages. RESULTS: Our model achieves a Dice of 0.45 (95% CI, 0.39–0.50) for core and 0.66 (95% CI, 0.62–0.69) for hypoperfused region, underestimating core volume by −9.3 mL (95% CI, −12.5 to −6.1) and hypoperfused region volume by −12.9 mL (95% CI, −21.1 to −4.7) compared with StrokeViewer. When comparing the 4 CTP software packages together, the average of their 2‐by‐2 agreement ranges from a Dice of 0.22 to 0.28 for core, and a Dice of 0.50 to 0.56 for hypoperfused region. This is similar to the average agreement of nnU‐Net with these 4 software packages (average Dice 0.27 for core and 0.56 for hypoperfused). Furthermore, nnU‐Net produces fewer connected components (1.3 for core, 1.6 for hypoperfused) than the average of the 4 CTP software packages (60.8 for core and 110.8 and hypoperfused), indicating more cohesive segmentations. CONCLUSION: Our model's performance in segmenting infarct core and hypoperfused regions from noncontrast CT and CT angiography is comparable to commercial CTP software packages, with potentially fewer segmentation artifacts. It can therefore be used when CTP is not available.
Background/Objectives: Computer-aided vessel occlusion detection for acute ischemic stroke is typically developed on early-phase CT angiography (CTA). In late-phase CTA, enhanced venous contrast can mimic arterial occlusions and lead to incorrect occlusion detections. To counteract this, we propose a constraint-based post-processing framework that leverages anatomical and temporal vascular information from CT perfusion (CTP) scans. Methods: Vascular anatomy and relative contrast-arrival times were encoded as time–vessel maps. Three complementary constraints worked on these maps to remove incorrect detections based on distance-to-vessel truncation points, contrast-arrival time, and spatial priors. Constraints were fixed on a 32-case late-phase development subset and applied to an external held-out cohort of late-phase CTA (N = 354). To enable application without CTP, an auxiliary CTA-to-time–vessel map generator was designed to estimate maps directly from CTA. Results: In the external cohort, the framework reduced incorrect occlusions per scan from 3.9 (95% CI 3.6–4.2) to 2.1 (95% CI 1.9–2.3), preserving occlusion-level sensitivity at 80% (95% CI 69–91). The patient-level AUROC was 83 for the baseline versus 88 with post-processing. Improvements were also observed in detectors exposed to late-phase CTA, where post-processing reduced incorrect occlusions per scan from 2.8 to 1.9. As only 51 occlusion-positive cases were present in the external cohort, confidence intervals were wide. Consequently, these estimates should be interpreted with caution. Conclusions: Anatomically and temporally informed post-processing reduced incorrect occlusion detections in late-phase CTA without measurable loss of sensitivity, potentially improving robustness against CTA phase changes in external institutions.