Machine learning-based prediction of insufficient contrast enhancement in coronary computed tomography angiography.

OBJECTIVES: Patient-tailored contrast delivery protocols strongly reduce the total iodine load and in general improve image quality in CT coronary angiography (CTCA). We aim to use machine learning to predict cases with insufficient contrast enhancement and to identify parameters with the highest predictive value. METHODS: Machine learning models were developed using data from 1,447 CTs. We included patient features, imaging settings, and test bolus features. The models were trained to predict CTCA images with a mean attenuation value in the ascending aorta below 400 HU. The accuracy was assessed by the area under the receiver operating characteristic (AUROC) and precision-recall curves (AUPRC). Shapley Additive exPlanations was used to assess the impact of features on the prediction of insufficient contrast enhancement. RESULTS: A total of 399 out of 1,447 scans revealed attenuation values in the ascending aorta below 400 HU. The best model trained using only patient features and CT settings achieved an AUROC of 0.78 (95% CI: 0.73-0.83) and AUPRC of 0.65 (95% CI: 0.58-0.71). With the inclusion of the test bolus features, it achieved an AUROC of 0.84 (95% CI: 0.81-0.87), an AUPRC of 0.71 (95% CI: 0.66-0.76), and a sensitivity of 0.66 and specificity of 0.88. The test bolus' peak height was the feature that impacted low attenuation prediction most. CONCLUSION: Prediction of insufficient contrast enhancement in CT coronary angiography scans can be achieved using machine learning models. Our experiments suggest that test bolus features are strongly predictive of low attenuation values and can be used to further improve patient-specific contrast delivery protocols. KEY POINTS: • Prediction of insufficient contrast enhancement in CT coronary angiography scans can be achieved using machine learning models. • The peak height of the test bolus curve is the most impacting feature for the best performing model.

Follow-up infarct volume as a mediator of endovascular treatment effect on functional outcome in ischaemic stroke.

OBJECTIVE: The putative mechanism for the favourable effect of endovascular treatment (EVT) on functional outcome after acute ischaemic stroke is preventing follow-up infarct volume (FIV) progression. We aimed to assess to what extent difference in FIV explains the effect of EVT on functional outcome in a randomised trial of EVT versus no EVT (MR CLEAN). METHODS: FIV was assessed on non-contrast CT scan 5-7 days after stroke. Functional outcome was the score on the modified Rankin Scale at 3 months. We tested the causal pathway from intervention, via FIV to functional outcome with a mediation model, using linear and ordinal regression, adjusted for relevant baseline covariates, including stroke severity. Explained effect was assessed by taking the ratio of the log odds ratios of treatment with and without adjustment for FIV. RESULTS: Of the 500 patients included in MR CLEAN, 60 died and four patients underwent hemicraniectomy before FIV was assessed, leaving 436 patients for analysis. Patients in the intervention group had better functional outcomes (adjusted common odds ratio (acOR) 2.30 (95% CI 1.62-3.26) than controls and smaller FIV (median 53 vs. 81 ml) (difference 28 ml; 95% CI 13-41). Smaller FIV was associated with better outcome (acOR per 10 ml 0.60, 95% CI 0.52-0.68). After adjustment for FIV the effect of intervention on functional outcome decreased but remained substantial (acOR 2.05, 95% CI 1.44-2.91). This implies that preventing FIV progression explains 14% (95% CI 0-34) of the beneficial effect of EVT on outcome. CONCLUSION: The effect of EVT on FIV explains only part of the treatment effect on functional outcome. KEY POINTS: • Endovascular treatment in acute ischaemic stroke patients prevents progression of follow-up infarct volume on non-contrast CT at 5-7 days. • Follow-up infarct volume was related to functional outcome, but only explained a modest part of the effect of intervention on functional outcome. • A large proportion of treatment effect on functional outcome remains unexplained, suggesting FIV alone cannot be used as an early surrogate imaging marker of functional outcome.

Value of machine learning in predicting TAVI outcomes.

BACKGROUND: Transcatheter aortic valve implantation (TAVI) has become a commonly applied procedure for high-risk aortic valve stenosis patients. However, for some patients, this procedure does not result in the expected benefits. Previous studies indicated that it is difficult to predict the beneficial effects for specific patients. We aim to study the accuracy of various traditional machine learning (ML) algorithms in the prediction of TAVI outcomes. METHODS AND RESULTS: Clinical and laboratory data from 1,478 TAVI patients from a single centre were collected. The outcome measures were improvement of dyspnoea and mortality. Three experiments were performed using (1) screening data, (2) laboratory data, and (3) the combination of both. Five well-established ML techniques were implemented, and the models were evaluated based on the area under the curve (AUC). Random forest classifier achieved the highest AUC (0.70) for predicting mortality. Logistic regression had the highest AUC (0.56) in predicting improvement of dyspnoea. CONCLUSIONS: In our single-centre TAVI population, the tree-based models were slightly more accurate than others in predicting mortality. However, ML models performed poorly in predicting improvement of dyspnoea.

Ruptured middle cerebral artery aneurysms with a concomitant intraparenchymal hematoma: the role of hematoma volume.

PURPOSE: To study whether clinical outcome data from our patient cohort could give support to the new recommendation in the AHA/ASA guidelines for the management of aneurysmal subarachnoid hemorrhage that states "that microsurgical clipping may receive increased consideration in patients with ruptured middle cerebral artery (MCA) aneurysms and large (>50 mL) intraparenchymal hematomas", while clinical outcome data supporting this recommendation are sparse. METHODS: We reviewed the clinical and radiological data of 81 consecutive patients with MCA aneurysms and concomitant hematomas admitted between January 2006 and December 2015. The relation between (semi-automatically quantified) hematoma volume (< or > 50 ml), neurological condition on admission (poor: GCS < 8 or non-reactive pupils), treatment strategies (no treatment, coiling, or clipping with or without decompression and/or clot removal), and outcome (favorable: mRS score 0-3) was evaluated. RESULTS: Clinical outcome data were available for 76 patients. A significant difference in favorable outcome (17 vs 68%) was seen when comparing patients with poor and good neurological condition on admission (p < 0.01). Patients with hematomas > 50 ml had similar outcomes for coiling and clipping, all underwent decompression. Patients with hematomas < 50 ml did not show differences in favorable outcome when comparing coiling and clipping with (33 and 31%) or without decompression (90 and 88%). CONCLUSION: Poor neurological condition on admission, and not large intraparenchymal hematoma volume, was associated with poor clinical outcome. Therefore, even in patients with large hematomas, the neurological condition on admission and the aneurysm configuration seem to be equally important factors to determine the most appropriate treatment strategy.

Added value of multiphase CTA imaging for thrombus perviousness assessment.

PURPOSE: Thrombus perviousness has been associated with favorable functional outcome in acute ischemic stroke (AIS) patients. Measuring thrombus perviousness on CTA may be suboptimal due to potential delay in contrast agent arrival in occluded arteries at the moment of imaging. Dynamic sequences acquired over time can potentially overcome this issue. We investigate if dynamic CTA has added value in assessing thrombus perviousness. METHODS: Prospectively collected image data of AIS patients with proven occlusion of the anterior or posterior circulation with thin-slice multi-phase CTA (MCTA) and non-contrast CT were co-registered (n = 221). Thrombus attenuation increase (TAI; a perviousness measure) was measured for the arterial, venous, and delayed phase of the MCTA and time-invariant CTAs (TiCTA). Associations with favorable clinical outcome (90-day mRS ≤ 2) were assessed using univariate and multivariable regressions and calculating areas under receiver operating curves (AUC). RESULTS: TAI determined from the arterial phase CTA was superior in the association with favorable outcome with OR = 1.21 per 10 HU increase (95%CI 1.04-1.41, AUC 0.62, p = 0.014) compared to any other phase (venous 1.14(95%CI 1.01-1.30, AUC 0.58, p = 0.033), delayed 1.046(95%CI 0.919-1.19, AUC 0.53, p = 0.50)), and TiCTA (1.15(95%CI 1.02-1.30, AUC 0.60, p = 0.022). In the multivariable model, only TAI on arterial phase was significantly associated with favorable outcome (aOR 1.59, 95%CI 1.04-2.43, p = 0.032). CONCLUSION: Association between TAI with functional outcome was optimal on arterial-phase CTA such that dynamic CTA imaging has no additional benefits in current thrombus perviousness assessment, thereby suggesting that the delay of contrast arrival at the clot is a key variable for patient functional outcome.

Non-contrast enhanced navigator-gated balanced steady state free precession magnetic resonance angiography as a preferred magnetic resonance technique for assessment of the thoracic aorta.

AIM: To compare the objective and subjective image quality of non-contrast three-dimensional (3D) navigator-gated balanced steady state free precession magnetic resonance angiography (NC-MRA) and contrast-enhanced magnetic resonance angiography (CE-MRA) along the entire thoracic aorta. MATERIALS AND METHODS: Fifty consecutive patients with thoracic aortic disease underwent NC-MRA and CE-MRA using a 1.5 T MRI system. Vessel sharpness was assessed using signal intensity profiles at five predefined levels of the thoracic aorta. Two readers scored subjective quality. Manual diameter measurements of both readers were used for calculation of interobserver variation. RESULTS: NC-MRA resulted in significantly sharper delineation of the aortic root, ascending aorta, and distal descending aorta compared to CE-MRA. Sharpness was comparable at the level of the arch and proximal descending aorta. NC-MRA resulted in significantly better subjective image quality. Interobserver agreement for diameter measurements was excellent for both techniques. CONCLUSION: NC-MRA resulted in superior image quality for assessment of the thoracic aorta and in better vessel sharpness for assessment of the aortic root and ascending aorta, when compared to CE-MRA. NC-MRA can be considered the MRA technique of choice for the assessment of the thoracic aorta diameters in clinical practice.

Biomechanical Imaging Markers as Predictors of Abdominal Aortic Aneurysm Growth or Rupture: A Systematic Review.

OBJECTIVES: Biomechanical characteristics, such as wall stress, are important in the pathogenesis of abdominal aortic aneurysms (AAA) and can be visualised and quantified using imaging techniques. This systematic review aims to present an overview of all biomechanical imaging markers that have been studied in relation to AAA growth and rupture. METHODS: This systematic review followed the PRISMA guidelines. A search in Medline, Embase, and the Cochrane Library identified 1503 potentially relevant articles. Studies were included if they assessed biomechanical imaging markers and their potential association with growth or rupture. RESULTS: Twenty-seven articles comprising 1730 patients met the inclusion criteria. Eighteen studies performed wall stress analysis using finite element analysis (FEA), 13 of which used peak wall stress (PWS) to quantify wall stress. Ten of 13 case control FEA studies reported a significantly higher PWS for symptomatic or ruptured AAAs than for intact AAAs. However, in some studies there was confounding bias because of baseline differences in aneurysm diameter between groups. Clinical heterogeneity in methodology obstructed a meaningful meta-analysis of PWS. Three of five FEA studies reported a significant positive association between several wall stress markers, such as PWS and 99th percentile stress, and growth. One study reported a significant negative association and one other study reported no significant association. Studies assessing wall compliance, the augmentation index and wall stress analysis using Laplace's law, computational fluid dynamics and fluid structure interaction were also included in this systematic review. CONCLUSIONS: Although PWS is significantly higher in symptomatic or ruptured AAAs in most FEA studies, confounding bias, clinical heterogeneity, and lack of standardisation limit the interpretation and generalisability of the results. Also, there is conflicting evidence on whether increased wall stress is associated with growth.

High performance clean versus artifact dry electrode EEG data classification using Convolutional Neural Network transfer learning.

Objective: Convolutional Neural Networks (CNNs) are promising for artifact detection in electroencephalography (EEG) data, but require large amounts of data. Despite increasing use of dry electrodes for EEG data acquisition, dry electrode EEG datasets are sparse. We aim to develop an algorithm for clean versus artifact dry electrode EEG data classification using transfer learning. Methods: Dry electrode EEG data were acquired in 13 subjects while physiological and technical artifacts were induced. Data were per 2-second segment labeled as clean or artifact and split in an 80% train and 20% test set. With the train set, we fine-tuned a pre-trained CNN for clean versus artifact wet electrode EEG data classification using 3-fold cross validation. The three fine-tuned CNNs were combined in one final clean versus artifact classification algorithm, in which the majority vote was used for classification. We calculated accuracy, F1-score, precision, and recall of the pre-trained CNN and fine-tuned algorithm when applied to unseen test data. Results: The algorithm was trained on 0.40 million and tested on 0.17 million overlapping EEG segments. The pre-trained CNN had a test accuracy of 65.6%. The fine-tuned clean versus artifact classification algorithm had an improved test accuracy of 90.7%, F1-score of 90.2%, precision of 89.1% and recall of 91.2%. Conclusions: Despite a relatively small dry electrode EEG dataset, transfer learning enabled development of a high performing CNN-based algorithm for clean versus artifact classification. Significance: Development of CNNs for classification of dry electrode EEG data is challenging as dry electrode EEG datasets are sparse. Here, we show that transfer learning can be used to overcome this problem.

Correlation between EEG spectral power and cerebral perfusion in patients with acute ischemic stroke.

Dry electrode electroencephalography (EEG) has the potential to diagnose ischemic stroke in the acute phase. In the current study we determined the correlation between EEG spectral power and ischemic stroke size and location as determined by computed tomography perfusion (CTP). Dry electrode EEG recordings were performed in patients with acute ischemic stroke in the emergency room. CTP preceded the EEG recordings as part of standard imaging protocol. Infarct core volume, total hypoperfused volume and local cerebral blood flow (CBF) were estimated with CTP. Additionally, global and local EEG spectral power were determined. We used Spearman's correlation coefficients to evaluate the correlation between variables. We included 27 patients (median age 72 [IQR:69-80] years, 15/27 [56%] men). Median CTP-to-EEG time was 32 (range:8-138) minutes. Hypoperfused volumes were estimated for 12/27 (44%) patients. Infarct core volume correlated best with global delta power (ρ = 0.76, p < 0.01), total hypoperfused volume with global alpha power (ρ = -0.58, p = 0.05), and local CBF with local alpha power (ρ = 0.43, p < 0.01). We conclude that dry electrode EEG signals slow down with increasing hypoperfused volume, which could potentially be used to discriminate between small and large ischemic strokes.

Detection of patent foramen ovale in patients with ischemic stroke on prospective ECG-gated cardiac CT compared to transthoracic echocardiography.

BACKGROUND: Cardiac CT acquired during the acute stroke imaging protocol is an emerging alternative to transthoracic echocardiography (TTE) to screen for sources of cardioembolism. Currently, its diagnostic accuracy to detect patent foramen ovale (PFO) is unclear. METHODS: This was a substudy of Mind the Heart, a prospective cohort in which consecutive adult patients with acute ischemic stroke underwent prospective ECG-gated cardiac CT during the initial stroke imaging protocol. Patients also underwent TTE. We included patients < 60 years who underwent TTE with agitated saline contrast (cTTE) and assessed sensitivity, specificity, negative and positive predictive value of cardiac CT for the detection of PFO using cTTE as the reference standard. RESULTS: Of 452 patients in Mind the Heart, 92 were younger than 60 years. Of these, 59 (64%) patients underwent both cardiac CT and cTTE and were included. Median age was 54 (IQR 49-57) years and 41/59 (70%) were male. Cardiac CT detected a PFO in 5/59 (8%) patients, 3 of which were confirmed on cTTE. cTTE detected a PFO in 12/59 (20%) patients. Sensitivity and specificity of cardiac CT were 25% (95% CI 5-57%) and 96% (95% CI 85-99%), respectively. Positive and negative predictive values were 59% (95% CI 14-95) and 84% (95% CI 71-92). CONCLUSION: Prospective ECG-gated cardiac CT acquired during the acute stroke imaging protocol does not appear to be a suitable screening method for PFO due to its low sensitivity. Our data suggest that if cardiac CT is used as a first-line screening method for cardioembolism, additional echocardiography remains indicated in young patients with cryptogenic stroke, in whom PFO detection would have therapeutic consequences. These results need to be confirmed in larger cohorts.

Infarct Evolution in Patients with Anterior Circulation Large-Vessel Occlusion Randomized to IV Alteplase and Endovascular Treatment versus Endovascular Treatment Alone.

BACKGROUND AND PURPOSE: Infarct evolution after endovascular treatment varies widely among patients with stroke and may be affected by baseline characteristics and procedural outcomes. Moreover, IV alteplase and endovascular treatment may influence the relationship of these factors to infarct evolution. We aimed to assess whether the infarct evolution between baseline and follow-up imaging was different for patients who received IVT and EVT versus EVT alone. MATERIALS AND METHODS: We included patients from the Multicenter Randomized Clinical Trial of Endovascular Treatment for Acute Ischemic Stroke in the Netherlands (MR CLEAN)-NO IV trial with baseline CTP and follow-up imaging. Follow-up infarct volume was segmented on 24-hour or 1-week follow-up DWI or NCCT. Infarct evolution was defined as the follow-up lesion volume: CTP core volume. Substantial infarct growth was defined as an increase in follow-up infarct volume of >10 mL. We assessed whether infarct evolution was different for patients with IV alteplase and endovascular treatment versus endovascular treatment alone and evaluated the association of baseline characteristics and procedural outcomes with infarct evolution using multivariable regression. RESULTS: From 228 patients with CTP results available, 145 patients had follow-up imaging and were included in our analysis. For patients with IV alteplase and endovascular treatment versus endovascular treatment alone, the baseline median CTP core volume was 17 (interquartile range = 4-35) mL versus 11 (interquartile range = 6-24) mL. The median follow-up infarct volume was 13 (interquartile range, 4-48) mL versus 17 (interquartile range = 4-50) mL. Collateral status and occlusion location were negatively associated with substantial infarct growth in patients with and without IV alteplase before endovascular treatment. CONCLUSIONS: No statistically significant difference in infarct evolution was found in directly admitted patients who received IV alteplase and endovascular treatment within 4.5 hours of symptom onset versus patients who underwent endovascular treatment alone. Collateral status and occlusion location may be useful predictors of infarct evolution prognosis in patients eligible for IV alteplase who underwent endovascular treatment.

Cardiac thrombus dissolution in acute ischemic stroke: A substudy of Mind the Heart.

Background: Cardiac thrombi are an important cause of ischemic stroke but are infrequently detected on cardiac imaging. We hypothesized that this might be explained by early dissolution of these cardiac thrombi after stroke occurrence. Methods: We performed a single-center observational pilot study between November 2019 and November 2020, embedded in the larger "Mind-the-Heart" study. We included patients with AIS and a cardiac thrombus in the left atrium or ventricle (filling defect <100 Hounsfield Units) diagnosed on cardiac CT that was acquired during the initial stroke imaging protocol. We repeated cardiac CT within one week to determine if the thrombus had dissolved. Results: Five patients (four men, median age 52 years, three with atrial fibrillation and one with anticoagulation therapy at baseline) were included. Median time from symptom onset to first cardiac CT was 383 (range 42-852) minutes and median time from first to second cardiac CT was three days (range 1-7). Two patients received intravenous thrombolysis (IVT). In total, six thrombi were seen on initial CT imaging (one in the left ventricle, four in the left atrial appendage, one in the left atrium). The left atrium thrombus and one left atrial appendage thrombus had dissolved on follow-up cardiac CT, one of which was in a patient with IVT treatment. Conclusion: This pilot study illustrates that cardiac thrombi can dissolve within days of stroke occurrence both with and without IVT treatment.

Complex flow patterns in a real-size intracranial aneurysm phantom: phase contrast MRI compared with particle image velocimetry and computational fluid dynamics.

The aim of this study was to validate the flow patterns measured by high-resolution, time-resolved, three-dimensional phase contrast MRI in a real-size intracranial aneurysm phantom. Retrospectively gated three-dimensional phase contrast MRI was performed in an intracranial aneurysm phantom at a resolution of 0.2 × 0.2 × 0.3 mm(3) in a solenoid rat coil. Both steady and pulsatile flows were applied. The phase contrast MRI measurements were compared with particle image velocimetry measurements and computational fluid dynamics simulations. A quantitative comparison was performed by calculating the differences between the magnitude of the velocity vectors and angles between the velocity vectors in corresponding voxels. Qualitative analysis of the results was executed by visual inspection and comparison of the flow patterns. The root-mean-square errors of the velocity magnitude in the comparison between phase contrast MRI and computational fluid dynamics were 5% and 4% of the maximum phase contrast MRI velocity, and the medians of the angle distribution between corresponding velocity vectors were 16° and 14° for the steady and pulsatile measurements, respectively. In the phase contrast MRI and particle image velocimetry comparison, the root-mean-square errors were 12% and 10% of the maximum phase contrast MRI velocity, and the medians of the angle distribution between corresponding velocity vectors were 19° and 15° for the steady and pulsatile measurements, respectively. Good agreement was found in the qualitative comparison of flow patterns between the phase contrast MRI measurements and both particle image velocimetry measurements and computational fluid dynamics simulations. High-resolution, time-resolved, three-dimensional phase contrast MRI can accurately measure complex flow patterns in an intracranial aneurysm phantom.

Robustness of radiomics to variations in segmentation methods in multimodal brain MRI.

Radiomics in neuroimaging uses fully automatic segmentation to delineate the anatomical areas for which radiomic features are computed. However, differences among these segmentation methods affect radiomic features to an unknown extent. A scan-rescan dataset (n = 46) of T1-weighted and diffusion tensor images was used. Subjects were split into a sleep-deprivation and a control group. Scans were segmented using four segmentation methods from which radiomic features were computed. First, we measured segmentation agreement using the Dice-coefficient. Second, robustness and reproducibility of radiomic features were measured using the intraclass correlation coefficient (ICC). Last, difference in predictive power was assessed using the Friedman-test on performance in a radiomics-based sleep deprivation classification application. Segmentation agreement was generally high (interquartile range = 0.77-0.90) and median feature robustness to segmentation method variation was higher (ICC > 0.7) than scan-rescan reproducibility (ICC 0.3-0.8). However, classification performance differed significantly among segmentation methods (p < 0.001) ranging from 77 to 84%. Accuracy was higher for more recent deep learning-based segmentation methods. Despite high agreement among segmentation methods, subtle differences significantly affected radiomic features and their predictive power. Consequently, the effect of differences in segmentation methods should be taken into account when designing and evaluating radiomics-based research methods.

Unsupervised Deep Learning for Stroke Lesion Segmentation on Follow-up CT Based on Generative Adversarial Networks.

BACKGROUND AND PURPOSE: Supervised deep learning is the state-of-the-art method for stroke lesion segmentation on NCCT. Supervised methods require manual lesion annotations for model development, while unsupervised deep learning methods such as generative adversarial networks do not. The aim of this study was to develop and evaluate a generative adversarial network to segment infarct and hemorrhagic stroke lesions on follow-up NCCT scans. MATERIALS AND METHODS: Training data consisted of 820 patients with baseline and follow-up NCCT from 3 Dutch acute ischemic stroke trials. A generative adversarial network was optimized to transform a follow-up scan with a lesion to a generated baseline scan without a lesion by generating a difference map that was subtracted from the follow-up scan. The generated difference map was used to automatically extract lesion segmentations. Segmentation of primary hemorrhagic lesions, hemorrhagic transformation of ischemic stroke, and 24-hour and 1-week follow-up infarct lesions were evaluated relative to expert annotations with the Dice similarity coefficient, Bland-Altman analysis, and intraclass correlation coefficient. RESULTS: The median Dice similarity coefficient was 0.31 (interquartile range, 0.08-0.59) and 0.59 (interquartile range, 0.29-0.74) for the 24-hour and 1-week infarct lesions, respectively. A much lower Dice similarity coefficient was measured for hemorrhagic transformation (median, 0.02; interquartile range, 0-0.14) and primary hemorrhage lesions (median, 0.08; interquartile range, 0.01-0.35). Predicted lesion volume and the intraclass correlation coefficient were good for the 24-hour (bias, 3 mL; limits of agreement, -64-59 mL; intraclass correlation coefficient, 0.83; 95% CI, 0.78-0.88) and excellent for the 1-week (bias, -4 m; limits of agreement,-66-58 mL; intraclass correlation coefficient, 0.90; 95% CI, 0.83-0.93) follow-up infarct lesions. CONCLUSIONS: An unsupervised generative adversarial network can be used to obtain automated infarct lesion segmentations with a moderate Dice similarity coefficient and good volumetric correspondence.

Quantitative analysis of EEG reactivity for neurological prognostication after cardiac arrest.

OBJECTIVE: To test whether 1) quantitative analysis of EEG reactivity (EEG-R) using machine learning (ML) is superior to visual analysis, and 2) combining quantitative analyses of EEG-R and EEG background pattern increases prognostic value for prediction of poor outcome after cardiac arrest (CA). METHODS: Several types of ML models were trained with twelve quantitative features derived from EEG-R and EEG background data of 134 adult CA patients. Poor outcome was a Cerebral Performance Category score of 3-5 within 6 months. RESULTS: The Random Forest (RF) trained on EEG-R showed the highest AUC of 83% (95-CI 80-86) of tested ML classifiers, predicting poor outcome with 46% sensitivity (95%-CI 40-51) and 89% specificity (95%-CI 86-92). Visual analysis of EEG-R had 80% sensitivity and 65% specificity. The RF was also the best classifier for EEG background (AUC 85%, 95%-CI 83-88) at 24 h after CA, with 62% sensitivity (95%-CI 57-67) and 84% specificity (95%-CI 79-88). Combining EEG-R and EEG background RF classifiers reduced the number of false positives. CONCLUSIONS: Quantitative EEG-R using ML predicts poor outcome with higher specificity, but lower sensitivity compared to visual analysis of EEG-R, and is of some additional value to ML on EEG background data. SIGNIFICANCE: Quantitative EEG-R using ML is a promising alternative to visual analysis and of some added value to ML on EEG background data.

Evolutionary algorithms and decision trees for predicting poor outcome after endovascular treatment for acute ischemic stroke.

Despite the large overall beneficial effects of endovascular treatment in patients with acute ischemic stroke, severe disability or death still occurs in almost one-third of patients. These patients, who might not benefit from treatment, have been previously identified with traditional logistic regression models, which may oversimplify relations between characteristics and outcome, or machine learning techniques, which may be difficult to interpret. We developed and evaluated a novel evolutionary algorithm for fuzzy decision trees to accurately identify patients with poor outcome after endovascular treatment, which was defined as having a modified Rankin Scale score (mRS) higher or equal to 5. The created decision trees have the benefit of being comprehensible, easily interpretable models, making its predictions easy to explain to patients and practitioners. Insights in the reason for the predicted outcome can encourage acceptance and adaptation in practice and help manage expectations after treatment. We compared our proposed method to CART, the benchmark decision tree algorithm, on classification accuracy and interpretability. The fuzzy decision tree significantly outperformed CART: using 5-fold cross-validation with on average 1090 patients in the training set and 273 patients in the test set, the fuzzy decision tree misclassified on average 77 (standard deviation of 7) patients compared to 83 (±7) using CART. The mean number of nodes (decision and leaf nodes) in the fuzzy decision tree was 11 (±2) compared to 26 (±1) for CART decision trees. With an average accuracy of 72% and much fewer nodes than CART, the developed evolutionary algorithm for fuzzy decision trees might be used to gain insights into the predictive value of patient characteristics and can contribute to the development of more accurate medical outcome prediction methods with improved clarity for practitioners and patients.

Model-based measurements of the diameter of the internal carotid artery in CT angiography images.

PURPOSE: Computed tomography angiography (CTA) is often used to determine the degree of stenosis in patients that suffer from carotid artery occlusive disease. Accurate and precise measurements of the diameter of the stenosed internal carotid artery are required to make decisions on treatment of the patient. However, the inherent blurring of images hampers a straightforward measurement, especially for smaller vessels. The authors propose a model-based approach to perform diameter measurements in which explicit allowance is made for the blurring of structures in the images. Three features of the authors' approach are the use of prior knowledge in the fitting of the model at the site of the stenosis, the applicability to vessels both with circular and noncircular cross-section, and the ability to deal with additional structures close to the arteries such as calcifications. METHODS: Noncircular cross-sections of vessels were modeled with elliptic Fourier descriptors. When calcifications or other high-intensity structures are adjacent to the lumen, both the lumen and the high-intensity structures were modeled in order to improve the diameter estimates of the vessel. Measurements were performed in CT scans of a phantom mimicking stenosed carotids and in CTA scans of two patients with an internal carotid stenosis. In an attempt to validate the measurements in CTA images, measurements were also performed in three-dimensional rotational angiography (3DRA) images of the same patients. RESULTS: The validity of the approach for diameter measurements of cylindrical arteries in CTA images is evident from phantom measurements. When prior knowledge about the enhancement and the blurring parameter was used, accurate and precise diameter estimates were obtained down to a diameter of 0.4 mm. The potential of the presented approach, both with respect to the extension to noncircular cross-sections and the modeling of adjacent calcifications, appears from the patient data. The accuracy of the size estimates in the patient images could not be unambiguously established because no gold standard was available and the quality of the 3DRA images was often suboptimal. CONCLUSIONS: The authors have shown that the inclusion of a priori information results in accurate and precise diameter measurements of arteries with a small diameter. Furthermore, in patient data, the assumption of a circular cross-section often appears to be too simple. The extension to noncircular cross-sections and adjacent calcifications paves the way to realistic modeling of the carotid artery.

Prediction of Outcome Using Quantified Blood Volume in Aneurysmal SAH.

BACKGROUND AND PURPOSE: In patients with SAH, the amount of blood is strongly associated with clinical outcome. However, it is commonly estimated with a coarse grading scale, potentially limiting its predictive value. Therefore, we aimed to develop and externally validate prediction models for clinical outcome, including quantified blood volumes, as candidate predictors. MATERIALS AND METHODS: Clinical and radiologic candidate predictors were included in a logistic regression model. Unfavorable outcome was defined as a modified Rankin Scale score of 4-6. An automatic hemorrhage-quantification algorithm calculated the total blood volume. Blood was manually classified as cisternal, intraventricular, or intraparenchymal. The model was selected with bootstrapped backward selection and validated with the R 2, C-statistic, and calibration plots. If total blood volume remained in the final model, its performance was compared with models including location-specific blood volumes or the modified Fisher scale. RESULTS: The total blood volume, neurologic condition, age, aneurysm size, and history of cardiovascular disease remained in the final models after selection. The externally validated predictive accuracy and discriminative power were high (R 2 = 56% ± 1.8%; mean C-statistic = 0.89 ± 0.01). The location-specific volume models showed a similar performance (R 2 = 56% ± 1%, P = .8; mean C-statistic = 0.89 ± 0.00, P = .4). The modified Fisher models were significantly less accurate (R 2 = 45% ± 3%, P < .001; mean C-statistic = 0.85 ± 0.01, P = .03). CONCLUSIONS: The total blood volume-based prediction model for clinical outcome in patients with SAH showed a high predictive accuracy, higher than a prediction model including the commonly used modified Fisher scale.

Comparing Morphology and Hemodynamics of Stable-versus-Growing and Grown Intracranial Aneurysms.

BACKGROUND AND PURPOSE: Aneurysm growth has been related to higher rupture risk. A better understanding of the characteristics related to growth may assist in the treatment decisions of unruptured intracranial aneurysms. This study aimed to identify morphologic and hemodynamic characteristics associated with aneurysm growth and to determine whether these characteristics deviate further from those of stable aneurysms after growth. MATERIALS AND METHODS: We included 81 stable and 56 growing aneurysms. 3D vascular models were segmented on CTA, MRA, or 3D rotational angiographic images. With these models, we performed computational fluid dynamics simulations. Morphologic (size, size ratios, and shape) and hemodynamic (inflow, vorticity, shear stress, oscillatory shear index, flow instability) characteristics were automatically calculated. We compared the characteristics between aneurysms that were stable and those that had grown at baseline and final imaging. The significance level after Bonferroni correction was P < .002. RESULTS: At baseline, no significant differences between aneurysms that were stable and those that had grown were detected (P > .002). Significant differences between aneurysms that were stable and those that had grown were seen at the final imaging for shear rate, aneurysm velocity, vorticity, and mean wall shear stress (P < .002). The latter was 11.5 (interquartile range, 5.4-18.8 dyne/cm2) compared with 17.5 (interquartile range, 11.2-29.9 dyne/cm2) in stable aneurysms (P = .001). Additionally, a trend toward lower area weighted average Gaussian curvature in aneurysms that had grown was observed with a median of 6.0 (interquartile range, 3.2-10.7 cm-2) compared with 10.4 (interquartile range, 5.0-21.2 cm-2) in stable aneurysms (P = .004). CONCLUSIONS: Morphologic and hemodynamic characteristics at baseline were not associated with aneurysm growth in our population. After growth, almost all indices increase toward values associated with higher rupture risks. Therefore, we stress the importance of longitudinal imaging and repeat risk assessment in unruptured aneurysms.