Automatic Detection and Segmentation of the Acute Vessel Thrombus in Cerebral CT

Lucas, Christian; Schöttler, Jonas J.; Kemmling, André; Aulmann, Linda F.; Heinrich, Mattias P.

doi:10.1007/978-3-658-25326-4_19

Cited by 6 publications

(5 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Qazi et al 12 used linear regression to build statistical models to predict patient-specific optimal Hounsfield unit thresholds, which replaced a universal single Hounsfield unit threshold for thrombus segmentation favored by Riedel et al 28 However, these thrombus density threshold-based methods are subject to image-intensity variability, and their generalizability is a concern. Lucas et al 29 proposed a cascaded neural network to segment thrombi. Unfortunately, this method was restricted to 2D images and limited to the MCA 1 ICA region, used fixed ROIs, and was developed using a small data set (the segmentation network was trained on only the 216 positive cases).…”

Section: Discussionmentioning

confidence: 99%

Automated Segmentation of Intracranial Thrombus on NCCT and CTA in Patients with Acute Ischemic Stroke Using a Coarse-to-Fine Deep Learning Model

Zhu

Bala

Zhang

et al. 2023

AJNR Am J Neuroradiol

View full text Add to dashboard Cite

BACKGROUND AND PURPOSE: Identifying the presence and extent of intracranial thrombi is crucial in selecting patients with acute ischemic stroke for treatment. This article aims to develop an automated approach to quantify thrombus on NCCT and CTA in patients with stroke. MATERIALS AND METHODS:A total of 499 patients with large-vessel occlusion from the Safety and Efficacy of Nerinetide in Subjects Undergoing Endovascular Thrombectomy for Stroke (ESCAPE-NA1) trial were included. All patients had thin-section NCCT and CTA images. Thrombi contoured manually were used as reference standard. A deep learning approach was developed to segment thrombi automatically. Of 499 patients, 263 and 66 patients were randomly selected to train and validate the deep learning model, respectively; the remaining 170 patients were independently used for testing. The deep learning model was quantitatively compared with the reference standard using the Dice coefficient and volumetric error. The proposed deep learning model was externally tested on 83 patients with and without large-vessel occlusion from another independent trial. RESULTS:The developed deep learning approach obtained a Dice coefficient of 70.7% (interquartile range, 58.0%-77.8%) in the internal cohort. The predicted thrombi length and volume were correlated with those of expert-contoured thrombi (r ¼ 0.88 and 0.87, respectively; P , .001). When the derived deep learning model was applied to the external data set, the model obtained similar results in patients with large-vessel occlusion regarding the Dice coefficient (66.8%; interquartile range, 58.5%-74.6%), thrombus length (r ¼ 0.73), and volume (r ¼ 0.80). The model also obtained a sensitivity of 94.12% (32/34) and a specificity of 97.96% (48/49) in classifying large-vessel occlusion versus non-large-vessel occlusion. CONCLUSIONS:The proposed deep learning method can reliably detect and measure thrombi on NCCT and CTA in patients with acute ischemic stroke.ABBREVIATIONS: AIS ¼ acute ischemic stroke; ASSD ¼ average symmetric surface distance; DC ¼ Dice coefficient; DL ¼ deep learning; EVT ¼ endovascular therapy; HD95 ¼ 95th percentile of the Hausdorff distance; IQR ¼ interquartile range; LVO ¼ large-vessel occlusion; MeVO ¼ medium-vessel occlusion; STAPLE ¼ simultaneous truth and performance level estimation R andomized controlled trials in patients with acute ischemic stroke (AIS) have demonstrated the efficacy and safety of endovascular therapy (EVT) compared with medical therapy in patients with large-vessel occlusion (LVO). 1,2 Identifying the presence, location, and extent of thrombi on NCCT and CTA images is important when selecting patients with AIS for reperfusion therapy. Thrombus characteristics such as location, length, volume, and permeability are helpful in predicting recanalization after both thrombolysis and EVT. 3,4 Recent studies have also shown that thrombus radiomics is able to predict recanalization with IV alteplase 5 and first-attempt recanalization with thromboaspiration. 6

show abstract

Section: Discussionmentioning

confidence: 99%

Automated Segmentation of Intracranial Thrombus on NCCT and CTA in Patients with Acute Ischemic Stroke Using a Coarse-to-Fine Deep Learning Model

Zhu

Bala

Zhang

et al. 2023

AJNR Am J Neuroradiol

View full text Add to dashboard Cite

show abstract

“…The poor density agreement suggests that the overlap between the segmentation results and the ground truth is suboptimal. Lucas et al [16] used cascaded CNNs and limited the search area to the MCA and ICA regions. They used their initial segmentation results as input to a classifier network to detect the affected hemisphere and used this information to further limit the search area to one hemisphere.…”

Section: Discussionmentioning

confidence: 99%

“…Convolutional neural networks (CNN)s have shown great promise in medical image segmentation. Lucas et al [16] use two cascaded CNNs to segment thrombi on NCCT. The first network is a U-Net architecture that takes a region of interest (ROI) containing middle cerebral artery (MCA) and ICA regions as input and segments all the candidate regions in it.…”

Section: Introductionmentioning

confidence: 99%

Fully Automated Thrombus Segmentation on CT Images of Patients with Acute Ischemic Stroke

et al. 2022

View full text Add to dashboard Cite

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.

show abstract

“…It should be mentioned that there are other complex methods for the segmentation of the thrombus in NCCT [10][11][12]. Some of these methods incorporate vessel characteristics and information about contralateral anatomy.…”

Section: Discussionmentioning

confidence: 99%

Identifying Thrombus on Non-Contrast CT in Patients with Acute Ischemic Stroke

Qazi

Wilson

et al. 2021

Diagnostics

View full text Add to dashboard Cite

The hyperdense sign is a marker of thrombus in non-contrast computed tomography (NCCT) datasets. The aim of this work was to determine optimal Hounsfield unit (HU) thresholds for thrombus segmentation in thin-slice non-contrast CT (NCCT) and use these thresholds to generate 3D thrombus models. Patients with thin-slice baseline NCCT (≤2.5 mm) and MCA-M1 occlusions were included. CTA was registered to NCCT, and three regions of interest (ROIs) were placed in the NCCT, including: the thrombus, contralateral brain tissue, and contralateral patent MCA-M1 artery. Optimal HU thresholds differentiating the thrombus from non-thrombus tissue voxels were calculated using receiver operating characteristic analysis. Linear regression analysis was used to predict the optimal HU threshold for discriminating the clot only based on the average contralateral vessel HU or contralateral parenchyma HU. Three-dimensional models from 70 participants using standard (45 HU) and patient-specific thresholds were generated and compared to CTA clot characteristics. The optimal HU threshold discriminating thrombus in NCCT from other structures varied with a median of 51 (IQR: 49–55). Experts chose 3D models derived using patient-specific HU models as corresponding better to the thrombus seen in CTA in 83.8% (31/37) of cases. Patient-specific HU thresholds for segmenting the thrombus in NCCT can be derived using normal parenchyma. Thrombus segmentation using patient-specific HU thresholds is superior to conventional 45 HU thresholds.

show abstract

Automatic Detection and Segmentation of the Acute Vessel Thrombus in Cerebral CT

Cited by 6 publications

References 8 publications

Automated Segmentation of Intracranial Thrombus on NCCT and CTA in Patients with Acute Ischemic Stroke Using a Coarse-to-Fine Deep Learning Model

Automated Segmentation of Intracranial Thrombus on NCCT and CTA in Patients with Acute Ischemic Stroke Using a Coarse-to-Fine Deep Learning Model

Fully Automated Thrombus Segmentation on CT Images of Patients with Acute Ischemic Stroke

Identifying Thrombus on Non-Contrast CT in Patients with Acute Ischemic Stroke

Contact Info

Product

Resources

About