The assessment of rupture probability is crucial to identifying at risk intracranial aneurysms (IA) in patients harboring multiple aneurysms. We aimed to develop a computer-assisted detection system for small-sized aneurysm ruptures using a convolutional neural network (CNN) based on images of three-dimensional digital subtraction angiography. A retrospective data set, including 368 patients, was used as a training cohort for the CNN using the TensorFlow platform. Aneurysm images in six directions were obtained from each patient and the region-of-interest in each image was extracted. The resulting CNN was prospectively tested in 272 patients and the sensitivity, specificity, overall accuracy, and receiver operating characteristics (ROC) were compared to a human evaluator. Our system showed a sensitivity of 78.76% (95% CI: 72.30%–84.30%), a specificity of 72.15% (95% CI: 60.93%–81.65%), and an overall diagnostic accuracy of 76.84% (95% CI: 71.36%–81.72%) in aneurysm rupture predictions. The area under the ROC (AUROC) in the CNN was 0.755 (95% CI: 0.699%–0.805%), better than that obtained from a human evaluator (AUROC: 0.537; p < 0.001). The CNN-based prediction system was feasible to assess rupture risk in small-sized aneurysms with diagnostic accuracy superior to human evaluators. Additional studies based on a large data set are necessary to enhance diagnostic accuracy and to facilitate clinical application.

Brain magnetic resonance imaging (MRI) is useful for predicting the outcome of patients with acute ischemic stroke (AIS). Although deep learning (DL) using brain MRI with certain image biomarkers has shown satisfactory results in predicting poor outcomes, no study has assessed the usefulness of natural language processing (NLP)-based machine learning (ML) algorithms using brain MRI free-text reports of AIS patients. Therefore, we aimed to assess whether NLP-based ML algorithms using brain MRI text reports could predict poor outcomes in AIS patients. This study included only English text reports of brain MRIs examined during admission of AIS patients. Poor outcome was defined as a modified Rankin Scale score of 3–6, and the data were captured by trained nurses and physicians. We only included MRI text report of the first MRI scan during the admission. The text dataset was randomly divided into a training and test dataset with a 7:3 ratio. Text was vectorized to word, sentence, and document levels. In the word level approach, which did not consider the sequence of words, and the “bag-of-words” model was used to reflect the number of repetitions of text token. The “sent2vec” method was used in the sensation-level approach considering the sequence of words, and the word embedding was used in the document level approach. In addition to conventional ML algorithms, DL algorithms such as the convolutional neural network (CNN), long short-term memory, and multilayer perceptron were used to predict poor outcomes using 5-fold cross-validation and grid search techniques. The performance of each ML classifier was compared with the area under the receiver operating characteristic (AUROC) curve. Among 1840 subjects with AIS, 645 patients (35.1%) had a poor outcome 3 months after the stroke onset. Random forest was the best classifier (0.782 of AUROC) using a word-level approach. Overall, the document-level approach exhibited better performance than did the word- or sentence-level approaches. Among all the ML classifiers, the multi-CNN algorithm demonstrated the best classification performance (0.805), followed by the CNN (0.799) algorithm. When predicting future clinical outcomes using NLP-based ML of radiology free-text reports of brain MRI, DL algorithms showed superior performance over the other ML algorithms. In particular, the prediction of poor outcomes in document-level NLP DL was improved more by multi-CNN and CNN than by recurrent neural network-based algorithms. NLP-based DL algorithms can be used as an important digital marker for unstructured electronic health record data DL prediction.

The optimal optic nerve sheath diameter (ONSD) cut-off for identifying increased intracranial pressure (IICP) remains unclear in adult patients. To validate the diagnostic accuracy of ultrasonographic (US) ONSD > 5.0 mm as a cut-off for detecting IICP by computed tomographic (CT) through a meta-analysis. A systemic literature review was performed of online databases from January 1990 to September 2017. A bivariate random-effects model was used to estimate pooled sensitivity, specificity, and diagnostic odds ratio (DOR) with 95% confidence intervals (CIs). A summary receiver operating characteristic (SROC) graph was used to provide summary points for sensitivity and specificity. Meta-regression tests were performed to estimate the influence of the study characteristics on DOR. Publication bias was assessed using Deeks' funnel plot asymmetry test. Six studies with 352 patients were included in the meta-analysis. US ONSD > 5.0 mm revealed pooled sensitivity of 99% (95% CI = 96–100) and specificity of 73% (95% CI = 65–80) for IICP detection. DOR was 178. The area under the SROC curve was 0.981, indicating a good level of accuracy. Meta-regression studies showed no significant associations between DOR and study characteristics such as probe mode (relative DOR [RDOR] = 0.60; P = 0.78), study quality (RDOR = 0.52; P = 0.67), IICP prevalence (RDOR = 0.04; P = 0.17), or pathology at admission (RDOR = 1.30; P = 0.87). US ONSD > 5.0 mm can be used to rapidly detect IICP in adults in emergency departments and intensive care units. Further meta-analysis based on individual patient-level databases is needed to confirm these results.

Objective : To compare peri-operative any symptomatic stroke after carotid angioplasty and stenting (CAS), based on the application or absence of a cerebral protection device. Methods : A systematic literature review using PubMed, Embase, and the Cochrane Central was done across an online data base from January 1995 to October 2016. Procedures which were performed due to carotid dissection or aneurysm, procedures using covered stents or conducted in an emergency, were excluded. The primary endpoint was perioperative any symptomatic stroke within 30 days after the procedure. A fixed effect model was used in cases of heterogeneity less than 50%. Results : In the 25 articles included in this study, the number of stroke events was 326 (2.0%) in protected CAS and 142 (3.4%) in unprotected CAS. The use of cerebral protection device significantly decreased stroke after CAS (odds ratio [OR] 0.633, 95% confidence interval [CI] 0.479-0.837, p=0.001). In the publication bias analysis, Egger's regression test disclosed that the intercept was -0.317 (95% CI -1.015-0.382, p=0.358). Regarding symptomatic patients (four studies, 539 CAS procedures), the number of stroke was six (1.7%) in protected CAS and 11 (5.7%) in unprotected CAS. The protective effect against stroke events by cerebral protection device did not have a statistical significance (OR 0.455, 95% CI 0.151-1.366, p=0.160). Conclusion : The use of protection device significantly decreased stroke after CAS. However, its efficacy was not demonstrated in symptomatic patients. Routine use of protection device during CAS should be critically assessed before mandatory use.