AJNR. American journal of neuroradiology最新文献

Amyloid-targeting therapy has recently become widely available in the U.S. for the treatment of patients with symptomatic mild Alzheimer's disease (AD). At present, there are no biomarkers that have been clinical validated to assess treatment response in routine clinical practice; longitudinal amyloid PET could play a role but is not cost effective. This report presents a case series of six patients with AD, whose amyloid positivity was confirmed by PET or CSF biomarkers, who underwent baseline and longitudinal arterial spin-labeling magnetic resonance imaging (ASL-MR) as part of FDA-mandated, clinical standard-of-care, non-contrast MR monitoring to assess for amyloid-related imaging abnormalities (ARIA). We and others have previously reported that ASL-MR can screen for neurodegenerative disease, as a proxy for FDG-PET, and can be easily added on as a cost-effective, repeatable method to monitor post-therapy changes. This series highlights varied cerebral blood flow (CBF) changes in response to lecanemab therapy. For instance, Cases 1, 3, and 5 showed increased CBF after multiple infusions, with subjective cognitive improvement in Case 1 and improved MoCA scores in Case 3. Case 2 showed improved CBF initially before the 5^th infusion, but this returned to baseline on the subsequent study, with no cognitive improvement over the course of therapy. Cases 4 and 6 have demonstrated no significant changes in regional CBF thus far on therapy, with cognitive decline in Case 4. This case series underscores the potential utility of ASL-MR as an adjunct sequence to current imaging protocols to monitor treatment response to anti-amyloid therapy.ABBREVIATIONS: ASL-MR= arterial spin-labeling magnetic resonance imaging; MRI= magnetic resonance imaging; CBF= cerebral blood flow; AD= Alzheimer's disease; PET= positron emission tomography; CSF= cerebrospinal fluid; FDG= fluorodeoxyglucose.

Background and purpose: To train and evaluate an open-source generative adversarial networks (GANs) to create synthetic lumbar spine MRI STIR volumes from T1 and T2 sequences, providing a proof-of-concept that could allow for faster MRI examinations.

Materials and methods: 1817 MRI examinations with sagittal T1, T2, and STIR sequences were accumulated and randomly divided into training, validation, and test sets. GANs were trained to create synthetic STIR volumes using the T1 and T2 volumes as inputs, optimized using the validation set, then applied to the test set. Acquired and synthetic test set volumes were independently evaluated in a blinded, randomized fashion by three radiologists specializing in musculoskeletal imaging and neuroradiology. Readers assessed image quality, motion artifacts, perceived likelihood of the volume being acquired or synthetic, and presence of 7 pathologies.

Results: The optimal model leveraged a customized loss function that accentuated foreground pixels, achieving a structural similarity imaging metric (SSIM) of 0.842, mean absolute error (MAE) of 0.028, and peak signal to noise ratio (PSNR) of 26.367. Radiologists could distinguish synthetic from acquired volumes; however, the synthetic volumes were of equal or better quality in 77% of test patients and demonstrated equivalent or decreased motion artifacts in 78% of test patients. For common pathologies, the synthetic volumes had high positive predictive value (75-100%) but lower sensitivity (0-67%).

Conclusions: This work links objective computer vision performance metrics and subject clinical evaluation of synthetic spine MRIs using open-source and reproducible methodologies. High-quality synthetic volumes are generated, reproducing many important pathologies, demonstrating a potential means for expediting imaging protocols.

Abbreviations: AI = Artificial Intelligence; GANs = general adversarial networks; aqSTIR = acquired STIR volume; sSTIR = synthetically generated STIR volume; SSIM = structural similarity imaging metric; PSNR = peak signal to noise ratio; MAE = mean absolute error.

Background and purpose: Magnetic Resonance Imaging is widely used to assess disease burden in multiple sclerosis (MS). This study aimed to evaluate the effectiveness of a commercially available k-nearest neighbors (k-NN) software in quantifying white matter lesion (WML) burden in MS. We compared the software's WML quantification to expert radiologists' assessments.

Materials and methods: We retrospectively reviewed brain MRI examinations of adult MS patients and of adult patients without MS and with a normal brain MRI referred from the neurology clinic. MRI images were processed using an AI-powered, cloud-based k-NN software, which generated a DICOM lesion distribution map and a report of WML count and volume in four brain regions (periventricular, deep, juxtacortical, and infratentorial white matter). Two blinded radiologists performed semi-quantitative assessments of WM lesion load and lesion segmentation accuracy. Additionally, four blinded neuroradiologists independently reviewed the data to determine if MRI findings supported an MS diagnosis. Results were considered significant when p < 0.05.

Results: The study included 32 MS patients (35.4 years ± 9.1) and 19 patients without MS (33.5 years ± 12.1). The k-NN software demonstrated 94.1% and 84.3% accuracy in differentiating MS from non-MS subjects based respectively on WML count and WML volume, compared to radiologists' accuracy of 90.2% to 94.1%. Lesion segmentation was more accurate for the deep WM and infratentorial regions than for the juxtacortical region (both p <0.001).

Conclusions: k-NN-derived WML volume and WML count provide valuable quantitative metrics of disease burden in MS. AI-powered post-processing software may enhance the interpretation of brain MRIs in MS patientsABBREVIATIONS: MS = multiple sclerosis; k-NN＝k-Nearest Neighbors; WML＝white matter lesion; MPRAGE = Magnetization-Prepared Rapid Acquisition Gradient Echo; SPACE = Sampling Perfection with Application-optimized Contrasts using a Different Flip Angle Evolution; EDSS = Expanded Disability Status Scale.

Background and purpose: Robustness against input data perturbations is essential for deploying deep-learning models in clinical practice. Adversarial attacks involve subtle, voxel-level manipulations of scans to increase deep-learning models' prediction errors. Testing deep-learning model performance on examples of adversarial images provides a measure of robustness, and including adversarial images in the training set can improve the model's robustness. In this study, we examined adversarial training and input modifications to improve the robustness of deep-learning models in predicting hematoma expansion (HE) from admission head CTs of patients with acute intracerebral hemorrhage (ICH).

Materials and methods: We used a multicenter cohort of n=890 patients for cross-validation/training, and a cohort of n=684 consecutive ICH patients from two stroke centers for independent validation. Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) adversarial attacks were applied for training and testing. We developed and tested four different models to predict ≥3mL, ≥6mL, ≥9mL, and ≥12mL HE in independent validation cohort applying Receiver Operating Characteristics (ROC) Area Under the Curve (AUC). We examined varying mixtures of adversarial and non-perturbed (clean) scans for training as well as including additional input from the hyperparameter-free Otsu multi-threshold segmentation for model.

Results: When deep-learning models trained solely on clean scans were tested with PGD and FGSM adversarial images, the average HE prediction AUC dropped from 0.8 to 0.67 and 0.71, respectively. Overall, the best performing strategy to improve model robustness was training with 5-to-3 mix of clean and PGD adversarial scans and addition of Otsu multi-threshold segmentation to model input, increasing the average AUC to 0.77 against both PGD and FGSM adversarial attacks. Adversarial training with FGSM improved robustness against similar type attack but offered limited cross-attack robustness against PGD-type images.

Conclusions: Adversarial training and inclusion of threshold-based segmentation as an additional input can improve deep-learning model robustness in prediction of HE from admission head CTs in acute ICH.

Abbreviations: ATACH-2= Antihypertensive Treatment of Acute Cerebral Hemorrhage; AUC= Area Under the Curve; Dice=Dice coefficient; CNN= Convolutional Neural Network; FGSM= Fast Gradient Sign Method; ICH= Intracerebral hemorrhage; HD= Hausdorff distance; HE= Hematoma expansion; PGD= Projected Gradient Descent; ROC= Receiver Operating Characteristics; VS= Volume similarity.

Background and purpose: Frontal paraventricular cystic changes have a varied etiology that includes connatal cysts, subependymal pseudocysts, necrosis, and enlarged perivascular spaces. These may be difficult to distinguish by neuroimaging and have a variety of associated prognoses. We aim to refine the neuroimaging definition of frontal horn cysts and correlate it with adverse clinical conditions.

Materials and methods: In this cross-sectional study, the pre-and postnatal neuroimaging database at a quaternary referral children's hospital was searched for all reports containing "frontal horn cysts", "periventricular cysts", or "connatal cysts" after IRB exemption. Frontal paraventricular abnormalities were categorized as either cysts, necroses, enlarged perivascular spaces, caudothalamic groove subependymal pseudocysts, frontal horn asymmetries, intraventricular septations, or ependymal vessels based on location and appearance. Cyst number, size, location, morphology, and signal/density/echotexture were documented, as were additional brain abnormalities. Clinical outcomes were recorded when available. Fisher's exact and Chi-squared tests were used to evaluate categorical data associations, and Kruskall-Wallis tests were employed to compare the medians among groups.

Results: 205 brain imaging exams (148 MRI; 55 US; 2 CT) from 110 distinct subjects (5 fetal: median 29.3, mean 27.5, and range 22.4 to 32.8 gestational weeks; 105 postnatal: mean 2.5 years, median 15 days, range 0 days to 19 years) were included. Seventy-one exams (35%) were initially diagnosed as connatal cysts but, instead, represented necrosis (n=23), enlarged perivascular spaces (n=20), caudothalamic groove germinolytic cysts (n=11), septations/adhesions (n=10), ventricular asymmetries (n=6), and a blood vessel (n=1). These entities differed in size, shape, location, and orientation (p<0.001). Congenital heart disease (p<0.04) and gastrointestinal (p<0.04) disorders were more common in subjects with frontal cysts and necrosis than in subjects with enlarged perivascular spaces; however, the frontal cyst and necrosis groups showed no differences in outcome (p>0.05).

Conclusions: Frontal paraventricular cystic changes represent a common interpretive dilemma. Enlarged perivascular spaces should be distinguished from other frontal cystic changes, which portend a more guarded prognosis, whether necrotic or otherwise.

Abbreviations: CMV= cytomegalovirus; CSPC= caudothalamic groove subependymal pseudocysts; FHCL= frontal horn cystic lesions; GA= gestational age; PVS= perivascular spaces.

Background and purpose: In this study, we aimed to develop and validate a novel nomogram model for predicting 90-day non-favorable clinical outcomes in patients with acute vertebrobasilar artery occlusion after endovascular treatment by integrating clinical and MRI features.

Materials and methods: This multicenter retrospective study analyzed data from 181 patients with vertebrobasilar artery occlusion eligible for endovascular treatment from two Chinese stroke centers. We developed a predictive model for non-favorable clinical outcomes (modified Rankin Scale score >3) using the data of 125 patients from Stroke Center A (2019-2023). The model was constructed using univariate and multivariate logistic regression analyses of clinical and MRI characteristics, with continuous variables dichotomized based on receiver operating characteristic curve analysis. Internal validation employed smooth bootstrapping, while external validation utilized 56 cases from Stroke Center B (2019-2023), ensuring model reliability and generalizability across diverse clinical settings.

Results: Age, NIHSS baseline score, recanalization, novel posterior circulation scores, and MRA-based posterior circulation collateral scores were independent predictors of 90-day prognosis, which were used to create a nomogram model. Internal validation demonstrated excellent discriminative performance of the model (mean area under the curve, 0.92 [95% CI: 0.91-0.93]), while external validation further confirmed its robust generalizability (area under the curve, 0.88). The patients were effectively stratified into the low-risk and high-risk groups using the nomogram model.

Conclusions: The posterior circulation collateral score was an independent predictor of prognosis. Our novel nomogram model, based on clinical and MRI characteristics, effectively predicts 90-day non-favorable clinical outcomes in patients with vertebrobasilar artery occlusion following endovascular treatment.

Abbreviations: AUC = area under the curve; ETO = estimated time of onset; EVT = endovascular treatment; FCO = favorable clinical outcome; mTICI = modified TICI; Novel-PCS = novel posterior circulation score; pc-ASPECTS = posterior circulation Acute Stroke Prognosis Early CT score; PC-CS = posterior circulation collateral score; ROC = receiver operating characteristic; VBAO = vertebrobasilar artery occlusion.

Background and purpose: In patients diagnosed with spontaneous intracranial hypotension (SIH), microspurs are considered the culprit lesion in most ventral dural leaks (type I). The imaging characteristics of discogenic spurs, and their prevalence in the general population has not been reported in the literature.

Materials and methods: This observational case-control study was conducted comparing the prevalence and characteristics of discogenic microspurs between SIH patients with a type I leak treated at a tertiary hospital between 2013 and 2023 and an age-and sex matched cohort of trauma patients.

Results: Each group consisted of 85 patients (mean age 51.6 years ± 11.9 years), 74% (58/85 patients) were female. The prevalence of discogenic microspurs in the control group and SIH group was 31.8% and 90.6%, respectively.The mean length of the culprit microspur responsible for a dural leak was larger compared to the mean length of all co-incidental microspurs from both the SIH and the control group not causing a dural leak (2.6mm versus 1.6mm, p<0.001).Our multivariate logistic regression revealed that an increasing length of a microspur (OR 1.942, CI 1.35-2.80, p<0.001) and a narrower diameter of the spinal canal (OR 0.85, CI 0.76-0.96, p=0.008) were predictive for a dural tear.

Conclusions: A discogenic microspur is a common incidental finding and may be found in almost one third of the general population. The length of the culprit microspur and the diameter of the spinal canal are distinct morphological characteristics for type I associated CSF leaks.

Abbreviations: CI = Confidence interval; CSF = Cerebrospinal fluid; CT = Computed tomography; ED = Emergency department; MRI = Magnetic resonance imaging; OR = Odds ratio; SD = Standard deviation; SIH = Spontaneous intracranial hypotension.

Background and purpose: Leptomeningeal collaterals have been associated with better outcomes in large-vessel stroke, but little is known about how the Circle of Willis (CoW) collaterals affect stroke outcomes. We aimed to determine the relationship between three anatomically distinct CoW subtypes and 90-day outcomes in acute ischemic stroke patients after successful revascularization via endovascular thrombectomy (EVT).

Materials and methods: We performed a retrospective analysis of patients treated with successful EVT for large-vessel occlusion at a comprehensive stroke center between May 2016 and November 2023. The CoW anatomy was trichotomized using baseline computed tomography angiography as follows: (I) complete CoW (C-CoW), (II) non-isolating incomplete CoW (NI-CoW), and (III) isolating incomplete CoW (I-CoW). Chi-squared and logistic regression analyses were utilized to determine the association of the CoW subtype with two co-primary outcomes: the 90-day modified Rankin Scale and 90-day mortality.

Results: A total of 465 patients were included in the analysis. Multivariable logistic regression analysis demonstrated a significant association between I-CoW and 90-day mRS compared to NI-CoW [OR (95% CI), 1.83 (1.08-3.09); p=0.02]. Additionally, I-CoW anatomy was associated with a higher 90-day mortality than C-CoW [OR (95%CI), 2.58 (1.01-6.60); p=0.04] and NI-CoW [OR (95% CI), 1.89 (1.13-3.18); p=0.01].

Conclusions: CoW variants are associated with functional and mortality outcomes in patients treated with endovascular thrombectomy for anterior circulation large vessel occlusion. Further research is needed to determine how CoW vessel anatomy may impact clinical assessment, triage, and treatment in acute ischemic stroke.

Abbreviations: CoW = Circle of Willis; EVT = endovascular thrombectomy; C-CoW = complete Circle of Willis; NI-CoW = non-isolating incomplete Circle of Willis; I-CoW = isolating Circle of Willis; AIS = acute ischemic stroke; LVO = large vessel occlusion; ACom = anterior communicating artery; PCom = posterior communicating artery; Tan CS = Tan collateral scores; ACA = anterior cerebral artery; PCA = posterior cerebral artery.

Background and purpose: Clinically, hemorrhagic transformation (HT) after mechanical thrombectomy (MT) is a common complication. This study aimed to investigate the value of clinical factors, CT signs, and radiomics in the differential diagnosis of high-density areas (HDAs) in the brain after MT in patients with acute ischemic stroke with large-vessel occlusion (AIS-LVO).

Materials and methods: A total of 156 eligible patients with AIS-LVO in Center I from December 2015 to June 2023 were retrospectively enrolled and randomly divided into training (n = 109) and internal validation (n = 47) sets at a ratio of 7:3. The data of 63 patients in Center II were collected as an external validation set. According to the diagnostic criteria, the patients in the 3 data sets were divided into an HT group and a non-HT group. The clinical and imaging data from Centers I and II were used to construct a clinical factor and CT-sign model, a radiomics model, and a combined model by logistic regression. Receiver operating characteristic analysis was used to evaluate the diagnostic efficacy of each model in the 3 data sets.

Results: Clinical blood glucose and the maximum cross-sectional area on CT were associated with the HT or non-HT of the HDA according to multivariate logistic regression analyses (P < .05). Among the 3 models, the combined model had the highest diagnostic efficiency, with area under the curve values of 0.895, 0.882, and 0.820 in the 3 data sets, which were significantly greater than the area under the curve values of the radiomics model (0.887, 0.898, 0.798) and clinical factor and CT-sign model (0.831, 0.744, 0.684).

Conclusions: The combined model based on radiomics had the best performance, indicating that radiomics features can be used as imaging biomarkers to aid in the clinical judgment of the nature of HDA after MT.