Artificial intelligence (AI) powered by deep learning (DL) has shown remarkable progress in image recognition tasks. Over the past decade, AI has proven its feasibility for applications in medical imaging. Various aspects of clinical practice in neuroimaging can be improved with the help of AI. For example, AI can aid in detecting brain metastases, predicting treatment response of brain tumors, generating a parametric map of dynamic contrast-enhanced MRI, and enhancing radiomics research by extracting salient features from input images. In addition, image quality can be improved via AI-based image reconstruction or motion artifact reduction. In this review, we summarize recent clinical applications of DL in various aspects of neuroimaging.

BACKGROUND: The purpose of this study was to qualitatively and quantitatively evaluate the effects of a metal artifact reduction for orthopedic implants (O-MAR) for brain computed tomographic angiography (CTA) in patients with aneurysm clips and coils. METHODS: The study included 36 consecutive patients with 47 intracranial metallic implants (42 aneurysm clips, 5 coils) who underwent brain CTA. The computed tomographic images with and without the O-MAR were independently reviewed both quantitatively and qualitatively by two reviewers. For quantitative analysis, image noises near the metallic implants of non-O-MAR and O-MAR images were compared. For qualitative analysis, image quality improvement and the presence of new streak artifacts were assessed. RESULTS: Image noise was significantly reduced near metallic implants (P < 0.01). Improvement of implant-induced streak artifacts was observed in eight objects (17.0%). However, streak artifacts were aggravated in 11 objects (23.4%), and adjacent vessel depiction was worsened in eight objects (17.0%). In addition, new O-MAR-related streak artifacts were observed in 32 objects (68.1%). New streak artifacts were more prevalent in cases with overlapping metallic implants on the same axial plane than in those without (P = 0.018). Qualitative assessment revealed that the overall image quality was not significantly improved in O-MAR images. CONCLUSION: In conclusion, the use of the O-MAR in patients with metallic implants significantly reduces image noise. However, the degree of the streak artifacts and surrounding vessel depiction were not significantly improved on O-MAR images.
Aneurysm ; Angiography* ; Artifacts* ; Brain* ; Humans ; Noise ; Orthopedics* ; Quality Improvement

OBJECTIVE: To evaluate the diagnostic benefits of 5-mm maximum intensity projection of improved motion-sensitized driven-equilibrium prepared contrast-enhanced 3D T1-weighted turbo-spin echo imaging (MIP iMSDE-TSE) in the detection of brain metastases. The imaging technique was compared with 1-mm images of iMSDE-TSE (non-MIP iMSDE-TSE), 1-mm contrast-enhanced 3D T1-weighted gradient-echo imaging (non-MIP 3D-GRE), and 5-mm MIP 3D-GRE. MATERIALS AND METHODS: From October 2014 to July 2015, 30 patients with 460 enhancing brain metastases (size > 3 mm, n = 150; size ≤ 3 mm, n = 310) were scanned with non-MIP iMSDE-TSE and non-MIP 3D-GRE. We then performed 5-mm MIP reconstruction of these images. Two independent neuroradiologists reviewed these four sequences. Their diagnostic performance was compared using the following parameters: sensitivity, reading time, and figure of merit (FOM) derived by jackknife alternative free-response receiver operating characteristic analysis. Interobserver agreement was also tested. RESULTS: The mean FOM (all lesions, 0.984; lesions ≤ 3 mm, 0.980) and sensitivity ([reader 1: all lesions, 97.3%; lesions ≤ 3 mm, 96.2%], [reader 2: all lesions, 97.0%; lesions ≤ 3 mm, 95.8%]) of MIP iMSDE-TSE was comparable to the mean FOM (0.985, 0.977) and sensitivity ([reader 1: 96.7, 99.0%], [reader 2: 97, 95.3%]) of non-MIP iMSDE-TSE, but they were superior to those of non-MIP and MIP 3D-GREs (all, p < 0.001). The reading time of MIP iMSDE-TSE (reader 1: 47.7 ± 35.9 seconds; reader 2: 44.7 ± 23.6 seconds) was significantly shorter than that of non-MIP iMSDE-TSE (reader 1: 78.8 ± 43.7 seconds, p = 0.01; reader 2: 82.9 ± 39.9 seconds, p < 0.001). Interobserver agreement was excellent (κ> 0.75) for all lesions in both sequences. CONCLUSION: MIP iMSDE-TSE showed high detectability of brain metastases. Its detectability was comparable to that of non-MIP iMSDE-TSE, but it was superior to the detectability of non-MIP/MIP 3D-GREs. With a shorter reading time, the false-positive results of MIP iMSDE-TSE were greater. We suggest that MIP iMSDE-TSE can provide high diagnostic performance and low false-positive rates when combined with 1-mm sequences.
Brain* ; Humans ; Magnetic Resonance Imaging ; Neoplasm Metastasis* ; ROC Curve

OBJECTIVE: To evaluate the utility of high-resolution vessel wall imaging (HR-VWI) of middle cerebral artery (MCA), and to compare HR-VWI findings between striatocapsular infarction (SC-I) and lenticulostriate infarction (LS-I). MATERIALS AND METHODS: This retrospective study was approved by the Institutional Review Board, and informed consent was waived. From July 2009 to February 2012, 145 consecutive patients with deep subcortical infarctions (SC-I, n = 81; LS-I, n = 64) who underwent HR-VWI were included in this study. The degree of MCA stenosis and the characteristics of MCA plaque (presence, eccentricity, location, extent, T2-high signal intensity [T2-HSI], and plaque enhancement) were analyzed, and compared between SC-I and LS-I, using Fisher's exact test. RESULTS: Stenosis was more severe in SC-I than in LS-I (p = 0.040). MCA plaque was more frequent in SC-I than in LS-I (p = 0.028), having larger plaque extent (p = 0.001), more T2-HSI (p = 0.001), and more plaque enhancement (p = 0.002). The eccentricity and location of the plaque showed no significant difference between the two groups. CONCLUSION: Both SC-I and LS-I have similar HR-VWI findings of the MCA plaque, but SC-I had more frequent, larger plaques with greater T2-HSI and enhancement. This suggests that HR-VWI may have a promising role in assisting the differentiation of underlying pathophysiological mechanism between SC-I and LS-I.
Cerebral Infarction* ; Constriction, Pathologic ; Ethics Committees, Research ; Humans ; Infarction ; Informed Consent ; Magnetic Resonance Imaging* ; Middle Cerebral Artery* ; Retrospective Studies ; Stroke

OBJECTIVE: To develop algorithms using convolutional neural networks (CNNs) for automatic segmentation of acute ischemic lesions on diffusion-weighted imaging (DWI) and compare them with conventional algorithms, including a thresholding-based segmentation. MATERIALS AND METHODS: Between September 2005 and August 2015, 429 patients presenting with acute cerebral ischemia (training:validation:test set = 246:89:94) were retrospectively enrolled in this study, which was performed under Institutional Review Board approval. Ground truth segmentations for acute ischemic lesions on DWI were manually drawn under the consensus of two expert radiologists. CNN algorithms were developed using two-dimensional U-Net with squeeze-and-excitation blocks (U-Net) and a DenseNet with squeeze-and-excitation blocks (DenseNet) with squeeze-and-excitation operations for automatic segmentation of acute ischemic lesions on DWI. The CNN algorithms were compared with conventional algorithms based on DWI and the apparent diffusion coefficient (ADC) signal intensity. The performances of the algorithms were assessed using the Dice index with 5-fold cross-validation. The Dice indices were analyzed according to infarct volumes (< 10 mL, ≥ 10 mL), number of infarcts (≤ 5, 6–10, ≥ 11), and b-value of 1000 (b1000) signal intensities (< 50, 50–100, > 100), time intervals to DWI, and DWI protocols. RESULTS: The CNN algorithms were significantly superior to conventional algorithms (p < 0.001). Dice indices for the CNN algorithms were 0.85 for U-Net and DenseNet and 0.86 for an ensemble of U-Net and DenseNet, while the indices were 0.58 for ADC-b1000 and b1000-ADC and 0.52 for the commercial ADC algorithm. The Dice indices for small and large lesions, respectively, were 0.81 and 0.88 with U-Net, 0.80 and 0.88 with DenseNet, and 0.82 and 0.89 with the ensemble of U-Net and DenseNet. The CNN algorithms showed significant differences in Dice indices according to infarct volumes (p < 0.001). CONCLUSION: The CNN algorithm for automatic segmentation of acute ischemic lesions on DWI achieved Dice indices greater than or equal to 0.85 and showed superior performance to conventional algorithms.
Brain Ischemia ; Consensus ; Diffusion ; Ethics Committees, Research ; Humans ; Retrospective Studies

PURPOSE: Although endovascular treatment is currently thought to only be suitable for patients who have pial arterial filling scores >3 as determined by multiphase computed tomography angiography (mpCTA), a cut-off score of 3 was determined by a study, including patients within 12 hours after symptom onset. We aimed to investigate whether a cut-off score of 3 for endovascular treatment within 6 hours of symptom onset is an appropriate predictor of good functional outcome at 3 months. MATERIALS AND METHODS: From April 2015 to January 2016, acute ischemic stroke patients treated with mechanical thrombectomy within 6 hours of symptom onset were enrolled into this study. Pial arterial filling scores were semi-quantitatively assessed using mpCTA, and clinical and radiological parameters were compared between patients with favorable and unfavorable outcomes. Multivariate logistic regression analysis was then performed to investigate the independent association between clinical outcome and pial collateral score, with the predictive power of the latter assessed using C-statistics. RESULTS: Of the 38 patients enrolled, 20 (52.6%) had a favorable outcome and 18 had an unfavorable outcome, with the latter group showing a lower mean pial arterial filling score (3.6±0.8 vs. 2.4±1.2, P=0.002). After adjusting for variables with a P-value of <0.1 in univariate analysis (i.e., age and National Institutes of Health Stroke Scale score at admission), pial arterial filling scores higher than a cut-off of 2 were found to be independently associated with favorable clinical outcomes (P=0.012). C-statistic analysis confirmed that our model had the highest prediction power when pial arterial filling scores were dichotomized at >2 vs. ≤2. CONCLUSION: A pial arterial filling cut-off score of 2 as determined by mpCTA appears to be more suitable for predicting clinical outcomes following endovascular treatment within 6 hours of symptom onset than the cut-off of 3 that had been previously suggested.
Angiography ; Humans ; Logistic Models ; National Institutes of Health (U.S.) ; Stroke ; Thrombectomy

Background: and Purpose Accurate classification of ischemic stroke subtype is important for effective secondary prevention of stroke. We used diffusion-weighted image (DWI) and atrial fibrillation (AF) data to train a deep learning algorithm to classify stroke subtype. Methods: Model development was done in 2,988 patients with ischemic stroke from three centers by using U-net for infarct segmentation and EfficientNetV2 for subtype classification. Experienced neurologists (n=5) determined subtypes for external test datasets, while establishing a consensus for clinical trial datasets. Automatically segmented infarcts were fed into the model (DWI-only algorithm). Subsequently, another model was trained, with AF included as a categorical variable (DWI+AF algorithm). These models were tested: (1) internally against the opinion of the labeling experts, (2) against fresh external DWI data, and (3) against clinical trial dataset. Results: In the training-and-validation datasets, the mean (±standard deviation) age was 68.0±12.5 (61.1% male). In internal testing, compared with the experts, the DWI-only and the DWI+AF algorithms respectively achieved moderate (65.3%) and near-strong (79.1%) agreement. In external testing, both algorithms again showed good agreements (59.3%–60.7% and 73.7%–74.0%, respectively). In the clinical trial dataset, compared with the expert consensus, percentage agreements and Cohen’s kappa were respectively 58.1% and 0.34 for the DWI-only vs. 72.9% and 0.57 for the DWI+AF algorithms. The corresponding values between experts were comparable (76.0% and 0.61) to the DWI+AF algorithm. Conclusion Our model trained on a large dataset of DWI (both with or without AF information) was able to classify ischemic stroke subtypes comparable to a consensus of stroke experts.