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

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 = 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 = 0.73), and volume (r = 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.

Outcomes with Endovascular Treatment of Patients with M2 Segment MCA Occlusion in the Late Time Window

BACKGROUND AND PURPOSE

Randomized trials in the late window have demonstrated the efficacy and safety of endovascular thrombectomy in large-vessel occlusions. Patients with M2-segment MCA occlusions were excluded from these trials. We compared outcomes with endovascular thrombectomy in patients with M2-versus-M1 occlusions presenting 6-24 hours after symptom onset.

MATERIALS AND METHODS

Analyses were on pooled data from studies enrolling patients with stroke treated with endovascular thrombectomy 6-24 hours after symptom onset. We compared 90-day functional independence (mRS ≤ 2), mortality, symptomatic intracranial hemorrhage, and successful reperfusion (expanded TICI = 2b-3) between patients with M2 and M1 occlusions. The benefit of successful reperfusion was then assessed among patients with M2 occlusion.

RESULTS

Of 461 patients, 367 (79.6%) had M1 occlusions and 94 (20.4%) had M2 occlusions. Patients with M2 occlusions were older and had lower median baseline NIHSS scores. Patients with M2 occlusion were more likely to achieve 90-day functional independence than those with M1 occlusion (adjusted OR = 2.13; 95% CI, 1.25-3.65). There were no significant differences in the proportion of successful reperfusion (82.9% versus 81.1%) or mortality (11.2% versus 17.2%). Symptomatic intracranial hemorrhage risk was lower in patients with M2-versus-M1 occlusions (4.3% versus 12.2%, P = .03). Successful reperfusion was independently associated with functional independence among patients with M2 occlusions (adjusted OR = 2.84; 95% CI, 1.11-7.29).

CONCLUSIONS

In the late time window, patients with M2 occlusions treated with endovascular thrombectomy achieved better clinical outcomes, similar reperfusion, and lower symptomatic intracranial hemorrhage rates compared with patients with M1 occlusion. These results support the safety and benefit of endovascular thrombectomy in patients with M2 occlusions in the late window.

Prevalence of Intracranial Atherosclerotic Disease in Patients with Low-Risk Transient or Persistent Neurologic Events

BACKGROUND AND PURPOSE

There are limited data on the prevalence and outcome of intracranial atherosclerotic disease in patients with low-risk transient or persistent minor neurologic events. We sought to determine the prevalence and risk factors associated with intracranial atherosclerotic disease in patients with low-risk transient or persistent neurologic events.

MATERIALS AND METHODS

Participants with available intracranial vascular imaging from the Diagnosis of Uncertain-Origin Benign Transient Neurologic Symptoms (DOUBT) study, a large prospective multicenter cohort study, were included in this post hoc analysis. The prevalence of intracranial atherosclerotic disease of ≥50% was determined, and the association with baseline characteristics and DWI lesions was evaluated using logistic regression.

RESULTS

We included 661 patients with a median age of 62 years (interquartile range, 53-70 years), of whom 53% were women. Intracranial atherosclerotic disease was found in 81 (12.3%) patients; asymptomatic intracranial atherosclerotic disease alone, in 65 (9.8%); and symptomatic intracranial atherosclerotic disease, in 16 (2.4%). The most frequent location was in the posterior cerebral artery (29%). Age was the only factor associated with any intracranial atherosclerotic disease (adjusted OR, 1.9 for 10 years increase; 95% CI, 1.6-2.5). Multivariable logistic regression showed a strong association between intracranial atherosclerotic disease and the presence of acute infarct on MR imaging (adjusted OR, 3.47; 95% CI, 1.91-6.25).

CONCLUSIONS

Intracranial atherosclerotic disease is not rare in patients with transient or persistent minor neurologic events and is independently associated with the presence of MR imaging-proved ischemia in this context. Evaluation of the intracranial arteries could be valuable in establishing the etiology of such low-risk events.

Interrater Agreement and Detection Accuracy for Medium-Vessel Occlusions Using Single-Phase and Multiphase CT Angiography

BACKGROUND AND PURPOSE

Accurate and reliable detection of medium-vessel occlusions is important to establish the diagnosis of acute ischemic stroke and initiate appropriate treatment with intravenous thrombolysis or endovascular thrombectomy. However, medium-vessel occlusions are often challenging to detect, especially for unexperienced readers. We aimed to evaluate the accuracy and interrater agreement of the detection of medium-vessel occlusions using single-phase and multiphase CTA.

MATERIALS AND METHODS

Single-phase and multiphase CTA of 120 patients with acute ischemic stroke (20 with no occlusion, 44 with large-vessel occlusion, and 56 with medium-vessel occlusion in the anterior and posterior circulation) were assessed by 3 readers with varying levels of experience (session 1: single-phase CTA; session 2: multiphase CTA). Interrater agreement for occlusion type (large-vessel occlusion versus medium-vessel occlusion versus no occlusion) and for detailed occlusion sites was calculated using the Fleiss κ with 95% confidence intervals. Accuracy for the detection of medium-vessel occlusions was calculated for each reader using classification tables.

RESULTS

Interrater agreement for occlusion type was moderate for single-phase CTA (κ  = 0.58; 95% CI, 0.56-0.62) and almost perfect for multiphase CTA (κ = 0.81; 95% CI, 0.78-0.83). Interrater agreement for detailed occlusion sites was moderate for single-phase CTA (κ = 0.55; 95% CI, 0.53-0.56) and substantial for multiphase CTA (κ = 0.71; 95% CI, 0.67-0.74). On single-phase CTA, readers 1, 2, and 3 classified 33/56 (59%), 34/56 (61%), and 32/56 (57%) correctly as medium-vessel occlusions. On multiphase CTA, 48/56 (86%), 50/56 (89%), and 50/56 (89%) medium-vessel occlusions were classified correctly.

CONCLUSIONS

Interrater agreement for medium-vessel occlusions is moderate when using single-phase CTA and almost perfect with multiphase CTA. Detection accuracy is substantially higher with multiphase CTA compared with single-phase CTA, suggesting that multiphase CTA might be a valuable tool for assessment of medium-vessel occlusion stroke.

Infarct Growth despite Successful Endovascular Reperfusion in Acute Ischemic Stroke: A Meta-analysis

BACKGROUND

Infarct volume inversely correlates with good recovery in stroke. The magnitude and predictors of infarct growth despite successful reperfusion via endovascular treatment are not known.

PURPOSE

We aimed to summarize the extent of infarct growth in patients with acute stroke who achieved successful reperfusion (TICI 2b-3) after endovascular treatment.

DATA SOURCES

We performed a systematic review and meta-analysis by searching MEDLINE and Google Scholar for articles published up to October 31, 2020.

STUDY SELECTION

Studies of >10 patients reporting baseline and post-endovascular treatment infarct volumes on MR imaging were included. Only patients with TICI 2b-3 were included. We calculated infarct growth at a study level as the difference between baseline and follow-up MR imaging infarct volumes.

DATA ANALYSIS

Our search yielded 345 studies, and we included 10 studies reporting on 973 patients having undergone endovascular treatment who achieved successful reperfusion.

DATA SYNTHESIS

The mean baseline infarct volume was 19.5 mL, while the mean final infarct volume was 37.5 mL. A TICI 2b reperfusion grade was achieved in 24% of patients, and TICI 2c or 3 in 76%. The pooled mean infarct growth was 14.8 mL (95% CI, 7.9-21.7 mL). Meta-regression showed higher infarct growth in studies that reported higher baseline infarct volumes, higher rates of incomplete reperfusion (modified TICI 2b), and longer onset-to-reperfusion times.

LIMITATIONS

Significant heterogeneity among studies was noted and might be driven by the difference in infarct growth between early- and late-treatment studies.

CONCLUSIONS

These results suggest considerable infarct growth despite successful endovascular treatment reperfusion and call for a faster workflow and the need for specific therapies to limit infarct growth.

COVID-19 Stroke Apical Lung Examination Study: A Diagnostic and Prognostic Imaging Biomarker in Suspected Acute Stroke

BACKGROUND AND PURPOSE

Diagnosis of coronavirus disease 2019 (COVID-19) relies on clinical features and reverse-transcriptase polymerase chain reaction testing, but the sensitivity is limited. Carotid CTA is a routine acute stroke investigation and includes the lung apices. We evaluated CTA as a potential COVID-19 diagnostic imaging biomarker.

MATERIALS AND METHODS

This was a multicenter, retrospective study (n = 225) including CTAs of patients with suspected acute stroke from 3 hyperacute stroke units (March-April 2020). We evaluated the reliability and accuracy of candidate diagnostic imaging biomarkers. Demographics, clinical features, and risk factors for COVID-19 and stroke were analyzed using univariate and multivariate statistics.

RESULTS

Apical ground-glass opacification was present in 22.2% (50/225) of patients. Ground-glass opacification had high interrater reliability (Fleiss κ = 0.81; 95% CI, 0.68-0.95) and, compared with reverse-transcriptase polymerase chain reaction, had good diagnostic performance (sensitivity, 75% [95% CI, 56-87]; specificity, 81% [95% CI, 71-88]; OR = 11.65 [95% CI, 4.14-32.78]; P < .001) on multivariate analysis. In contrast, all other contemporaneous demographic, clinical, and imaging features available at CTA were not diagnostic for COVID-19. The presence of apical ground-glass opacification was an independent predictor of increased 30-day mortality (18.0% versus 5.7%, P = .017; hazard ratio = 3.51; 95% CI, 1.42-8.66; P = .006).

CONCLUSIONS

We identified a simple, reliable, and accurate COVID-19 diagnostic and prognostic imaging biomarker obtained from CTA lung apices: the presence or absence of ground-glass opacification. Our findings have important implications in the management of patients presenting with suspected stroke through early identification of COVID-19 and the subsequent limitation of disease transmission.

Follow-Up MRI for Small Brain AVMs Treated by Radiosurgery: Is Gadolinium Really Necessary?

BACKGROUND AND PURPOSE

Follow-up MR imaging of brain AVMs currently relies on contrast-enhanced sequences. Noncontrast techniques, including arterial spin-labeling and TOF, may have value in detecting a residual nidus after radiosurgery. The aim of this study was to compare noncontrast with contrast-enhanced MR imaging for the differentiation of residual-versus-obliterated brain AVMs in radiosurgically treated patients.

MATERIALS AND METHODS

Twenty-eight consecutive patients with small brain AVMs (<20 mm) treated by radiosurgery were followed with the same MR imaging protocol. Three neuroradiologists, blinded to the results, independently reviewed the following: 1) postcontrast images alone (4D contrast-enhanced MRA and postcontrast 3D T1 gradient recalled-echo), 2) arterial spin-labeling and TOF images alone, and 3) all MR images combined. The primary end point was the detection of residual brain AVMs using a 5-point scale, with DSA as the reference standard.

RESULTS

The highest interobserver agreement was for arterial spin-labeling/TOF (κ = 0.81; 95% confidence interval, 0.66-0.93). Regarding brain AVM detection, arterial spin-labeling/TOF had higher sensitivity (sensitivity, 85%; specificity, 100%; 95% CI, 62-97) than contrast-enhanced MR imaging (sensitivity, 55%; specificity, 100%; 95% CI, 27-73) and all MR images combined (sensitivity, 75%; specificity, 100%; 95% CI, 51-91) (P = .008). All nidus obliterations on DSA were detected on MR imaging. In 6 patients, a residual brain AVM present on DSA was only detected with arterial spin-labeling/TOF, including 3 based solely on arterial spin-labeling images.

CONCLUSIONS

In this study of radiosurgically treated patients with small brain AVMs, arterial spin-labeling/TOF was found to be superior to gadolinium-enhanced MR imaging in detecting residual AVMs.