Using Domain Adaptation and Inductive Transfer Learning to Improve Patient Outcome Prediction in the Intensive Care Unit: Retrospective Observational Study.

BACKGROUND: Accurate patient outcome prediction in the intensive care unit (ICU) can potentially lead to more effective and efficient patient care. Deep learning models are capable of learning from data to accurately predict patient outcomes, but they typically require large amounts of data and computational resources. Transfer learning (TL) can help in scenarios where data and computational resources are scarce by leveraging pretrained models. While TL has been widely used in medical imaging and natural language processing, it has been rare in electronic health record (EHR) analysis. Furthermore, domain adaptation (DA) has been the most common TL method in general, whereas inductive transfer learning (ITL) has been rare. To the best of our knowledge, DA and ITL have never been studied in-depth in the context of EHR-based ICU patient outcome prediction. OBJECTIVE: This study investigated DA, as well as rarely researched ITL, in EHR-based ICU patient outcome prediction under simulated, varying levels of data scarcity. METHODS: Two patient cohorts were used in this study: (1) eCritical, a multicenter ICU data from 55,689 unique admission records from 48,672 unique patients admitted to 15 medical-surgical ICUs in Alberta, Canada, between March 2013 and December 2019, and (2) Medical Information Mart for Intensive Care III, a single-center, publicly available ICU data set from Boston, Massachusetts, acquired between 2001 and 2012 containing 61,532 admission records from 46,476 patients. We compared DA and ITL models with baseline models (without TL) of fully connected neural networks, logistic regression, and lasso regression in the prediction of 30-day mortality, acute kidney injury, ICU length of stay, and hospital length of stay. Random subsets of training data, ranging from 1% to 75%, as well as the full data set, were used to compare the performances of DA and ITL with the baseline models at various levels of data scarcity. RESULTS: Overall, the ITL models outperformed the baseline models in 55 of 56 comparisons (all P values <.001). The DA models outperformed the baseline models in 45 of 56 comparisons (all P values <.001). ITL resulted in better performance than DA in terms of the number of times and the margin with which it outperformed the baseline models. In 11 of 16 cases (8 of 8 for ITL and 3 of 8 for DA), TL models outperformed baseline models when trained using 1% data subset. CONCLUSIONS: TL-based ICU patient outcome prediction models are useful in data-scarce scenarios. The results of this study can be used to estimate ICU outcome prediction performance at different levels of data scarcity, with and without TL. The publicly available pretrained models from this study can serve as building blocks in further research for the development and validation of models in other ICU cohorts and outcomes.

Automatic classification of major depression disorder using arterial spin labeling MRI perfusion measurements.

AIM: Neuroimaging-based multivariate pattern-recognition methods have been successfully used to develop diagnostic algorithms to distinguish patients with major depressive disorder (MDD) from healthy controls (HC). We developed and evaluated the accuracy of a multivariate classification method for the differentiation of MDD and HC using cerebral blood flow (CBF) features measured by non-invasive arterial spin labeling (ASL) MRI. METHODS: Twenty-two medication-free patients with the diagnosis of MDD based on DSM-IV criteria and 22 HC underwent pseudo-continuous 3-D-ASL imaging to assess CBF. Using an atlas-based approach, regional CBF was determined in various brain regions and used together with sex and age as classification features. A linear kernel support vector machine was used for feature ranking and selection as well as for the classification of patients with MDD and HC. Permutation testing was used to test for significance of the classification results. RESULTS: The automatic classifier based on CBF features showed a statistically significant accuracy of 77.3% (P = 0.004) with a specificity of 80% and sensitivity of 75% for classification of MDD versus HC. The features that contributed to the classification were sex and regional CBF of the cortical, limbic, and paralimbic regions. CONCLUSION: Machine-learning models based on CBF measurements are capable of differentiating MDD from HC with high accuracy. The use of larger study cohorts and inclusion of other imaging measures may improve the performance of the classifier to achieve the accuracy required for clinical application.

Influence of stroke infarct location on functional outcome measured by the modified rankin scale.

BACKGROUND AND PURPOSE: In the early days after ischemic stroke, information on structural brain damage from MRI supports prognosis of functional outcome. It is rated widely by the modified Rankin Scale that correlates only moderately with lesion volume. We therefore aimed to elucidate the influence of lesion location from early MRI (days 2-3) on functional outcome after 1 month using voxel-based lesion symptom mapping. METHODS: We analyzed clinical and MRI data of patients from a prospective European multicenter stroke imaging study (I-KNOW). Lesions were delineated on fluid-attenuated inversion recovery images on days 2 to 3 after stroke onset. We generated statistic maps of lesion contribution related to clinical outcome (modified Rankin Scale) after 1 month using voxel-based lesion symptom mapping. RESULTS: Lesion maps of 101 patients with middle cerebral artery infarctions were included for analysis (right-sided stroke, 47%). Mean age was 67 years, median admission National Institutes of Health Stroke Scale was 11. Mean infarct volumes were comparable between both sides (left, 37.5 mL; right, 43.7 mL). Voxel-based lesion symptom mapping revealed areas with high influence on higher modified Rankin Scale in regions involving the corona radiata, internal capsule, and insula. In addition, asymmetrically distributed impact patterns were found involving the right inferior temporal gyrus and left superior temporal gyrus. CONCLUSIONS: In this group of patients with stroke, characteristic lesion patterns in areas of motor control and areas involved in lateralized brain functions on early MRI were found to influence functional outcome. Our data provide a novel map of the impact of lesion localization on functional stroke outcome as measured by the modified Rankin Scale.

Spatial normalization for voxel-based lesion symptom mapping: impact of registration approaches.

Background: Voxel-based lesion symptom mapping (VLSM) assesses the relation of lesion location at a voxel level with a specific clinical or functional outcome measure at a population level. Spatial normalization, that is, mapping the patient images into an atlas coordinate system, is an essential pre-processing step of VLSM. However, no consensus exists on the optimal registration approach to compute the transformation nor are downstream effects on VLSM statistics explored. In this work, we evaluate four registration approaches commonly used in VLSM pipelines: affine (AR), nonlinear (NLR), nonlinear with cost function masking (CFM), and enantiomorphic registration (ENR). The evaluation is based on a standard VLSM scenario: the analysis of statistical relations of brain voxels and regions in imaging data acquired early after stroke onset with follow-up modified Rankin Scale (mRS) values. Materials and methods: Fluid-attenuated inversion recovery (FLAIR) MRI data from 122 acute ischemic stroke patients acquired between 2 and 3 days after stroke onset and corresponding lesion segmentations, and 30 days mRS values from a European multicenter stroke imaging study (I-KNOW) were available and used in this study. The relation of the voxel location with follow-up mRS was assessed by uni- as well as multi-variate statistical testing based on the lesion segmentations registered using the four different methods (AR, NLR, CFM, ENR; implementation based on the ANTs toolkit). Results: The brain areas evaluated as important for follow-up mRS were largely consistent across the registration approaches. However, NLR, CFM, and ENR led to distortions in the patient images after the corresponding nonlinear transformations were applied. In addition, local structures (for instance the lateral ventricles) and adjacent brain areas remained insufficiently aligned with corresponding atlas structures even after nonlinear registration. Conclusions: For VLSM study designs and imaging data similar to the present work, an additional benefit of nonlinear registration variants for spatial normalization seems questionable. Related distortions in the normalized images lead to uncertainties in the VLSM analyses and may offset the theoretical benefits of nonlinear registration.

Risk Factors for Perinatal Arterial Ischemic Stroke: A Machine Learning Approach.

BACKGROUND AND OBJECTIVES: Perinatal arterial ischemic stroke (PAIS) is a focal vascular brain injury presumed to occur between the fetal period and the first 28 days of life. It is the leading cause of hemiparetic cerebral palsy. Multiple maternal, intrapartum, delivery, and fetal factors have been associated with PAIS, but studies are limited by modest sample sizes and complex interactions between factors. Machine learning approaches use large and complex data sets to enable unbiased identification of clinical predictors but have not yet been applied to PAIS. We combined large PAIS data sets and used machine learning methods to identify clinical PAIS factors and compare this data-driven approach with previously described literature-driven clinical prediction models. METHODS: Common data elements from 3 registries with patients with PAIS, the Alberta Perinatal Stroke Project, Canadian Cerebral Palsy Registry, International Pediatric Stroke Study, and a longitudinal cohort of healthy controls (Alberta Pregnancy Outcomes and Nutrition Study), were used to identify potential predictors of PAIS. Inclusion criteria were term birth and idiopathic PAIS (absence of primary causative medical condition). Data including maternal/pregnancy, intrapartum, and neonatal factors were collected between January 2003 and March 2020. Common data elements were entered into a validated random forest machine learning pipeline to identify the highest predictive features and develop a predictive model. Univariable analyses were completed post hoc to assess the relationship between each predictor and outcome. RESULTS: A machine learning model was developed using data from 2,571 neonates, including 527 cases (20%) and 2,044 controls (80%). With a mean of 21 features selected, the random forest machine learning approach predicted the outcome with approximately 86.5% balanced accuracy. Factors that were selected a priori through literature-driven variable selection that were also identified as most important by the machine learning model were maternal age, recreational substance exposure, tobacco exposure, intrapartum maternal fever, and low Apgar score at 5 minutes. Additional variables identified through machine learning included in utero alcohol exposure, infertility, miscarriage, primigravida, meconium, spontaneous vaginal delivery, neonatal head circumference, and 1-minute Apgar score. Overall, the machine learning model performed better (area under the curve [AUC] 0.93) than the literature-driven model (AUC 0.73). DISCUSSION: Machine learning may be an alternative, unbiased method to identify clinical predictors associated with PAIS. Identification of previously suggested and novel clinical factors requires cautious interpretation but supports the multifactorial nature of PAIS pathophysiology. Our results suggest that identification of neonates at risk of PAIS is possible.

Persistent hemodynamic changes in ruptured brain arteriovenous malformations.

BACKGROUND AND PURPOSE: Hemodynamic properties of brain arteriovenous malformations (AVMs) with risk factors for a future hemorrhage are essentially unknown. We hypothesized that AVMs with anatomic properties, which are associated with an increased rupture risk, exhibit different hemodynamic characteristics than those without these properties. METHODS: Seventy-two consecutive patients with AVMs diagnosed by conventional angiography underwent MRI examination, including time-resolved 3-dimensional MR angiography. Signal-intensity curves derived from the time-resolved 3-dimensional MR angiography datasets were used to calculate relative blood flow transit times through the AVM nidus based on the time-to-peak parameter. For identification of characteristics associated with altered transit times, a multiple normal regression model was fitted with stepwise selection of the following regressors: intracranial hemorrhage, deep nidus location, infratentorial location, deep drainage, associated aneurysm, nidus size, draining venous stenosis, and number of draining veins. RESULTS: A previous intracranial hemorrhage is the only characteristic that was associated with a significant alteration of the relative transit time, leading to an increase of 2.4 seconds (95% CI, 1.2-3.6 seconds;, P<0.001) without adjustment and 2.1 seconds (95% CI, 0.6-3.6 seconds; P=0.007) with adjustment for all other regressors considered. The association was independent of the bleeding age. CONCLUSIONS: Hemodynamic parameters do not seem useful for risk assessment of an AVM-related hemorrhage because only a previous AVM rupture leads to a significant and permanent alteration of the hemodynamic situation.

4D blood flow visualization fusing 3D and 4D MRA image sequences.

PURPOSE: To present and evaluate the feasibility of a novel automatic method for generating 4D blood flow visualizations fusing high spatial resolution 3D and time-resolved (4D) magnetic resonance angiography (MRA) datasets. MATERIALS AND METHODS: In a first step, the cerebrovascular system is segmented in the 3D MRA dataset and a surface model is computed. The hemodynamic information is extracted from the 4D MRA dataset and transferred to the surface model using rigid registration where it can be visualized color-coded or dynamically over time. The presented method was evaluated using software phantoms and 20 clinical datasets from patients with an arteriovenous malformation. Clinical evaluation was performed by comparison of Spetzler-Martin scores determined from the 4D blood flow visualizations and corresponding digital subtraction angiographies. RESULTS: The performed software phantom validation showed that the presented method is capable of producing reliable visualization results for vessels with a minimum diameter of 2 mm for which a mean temporal error of 0.27 seconds was achieved. The clinical evaluation based on 20 datasets comparing the 4D visualization to DSA images revealed an excellent interrater reliability. CONCLUSION: The presented method enables an improved combined representation of blood flow and anatomy while reducing the time needed for clinical rating.

Rigid 3D-3D registration of TOF MRA integrating vessel segmentation for quantification of recurrence volumes after coiling cerebral aneurysm.

INTRODUCTION: A fast and reproducible quantification of the recurrence volume of coiled aneurysms is required to enable a more timely evaluation of new coils. This paper presents two registration schemes for the semi-automatic quantification of aneurysm recurrence volumes based on baseline and follow-up 3D MRA TOF datasets. METHODS: The quantification of shape changes requires a previous definition of corresponding structures in both datasets. For this, two different rigid registration methods have been developed and evaluated. Besides a state-of-the-art rigid registration method, a second approach integrating vessel segmentations is presented. After registration, the aneurysm recurrence volume can be calculated based on the difference image. The computed volumes were compared to manually extracted volumes. RESULTS: An evaluation based on 20 TOF MRA datasets (baseline and follow-up) of ten patients showed that both registration schemes are generally capable of providing sufficient registration results. Regarding the quantification of aneurysm recurrence volumes, the results suggest that the second segmentation-based registration method yields better results, while a reduction of the computation and interaction time is achieved at the same time. CONCLUSION: The proposed registration scheme incorporating vessel segmentation enables an improved quantification of recurrence volumes of coiled aneurysms with reduced computation and interaction time.

Generation of a probabilistic arterial cerebrovascular atlas derived from 700 time-of-flight MRA datasets.

The cerebral vasculature is a complex vessel network with high variations among human subjects. Although the coarse structure and spatial relationships of the main cerebrovascular branches are well known, not much knowledge about inter-individual vessel variability of humans at a finer level is available. The aim of this work is to present a probabilistic atlas of cerebral arterial vascular structures derived from 700 Time-of-Flight (TOF) magnetic resonance angiography (MRA) datasets of healthy subjects. Therefore, the cerebrovascular system was automatically segmented in each TOF datasets. In a following step, each TOF dataset and corresponding segmentation was registered to the MNI brain atlas. The registered datasets were then used for generation of a probabilistic cerebrovascular atlas. The generated atlas was evaluated with respect to three possible applications. The results suggest that the atlas is especially helpful to obtain knowledge about the cerebrovascular anatomy and its variations in terms of vessel occurrence probability. Furthermore, it appears useful for initialization of automatic cerebrovascular segmentation methods while an application for detection of vessel pathologies seems only feasible for large malformations.

Lesion-symptom mapping with NIHSS sub-scores in ischemic stroke patients.

BACKGROUND: Lesion-symptom mapping (LSM) is a statistical technique to investigate the population-specific relationship between structural integrity and post-stroke clinical outcome. In clinical practice, patients are commonly evaluated using the National Institutes of Health Stroke Scale (NIHSS), an 11-domain clinical score to quantitate neurological deficits due to stroke. So far, LSM studies have mostly used the total NIHSS score for analysis, which might not uncover subtle structure-function relationships associated with the specific sub-domains of the NIHSS evaluation. Thus, the aim of this work was to investigate the feasibility to perform LSM analyses with sub-score information to reveal category-specific structure-function relationships that a total score may not reveal. METHODS: Employing a multivariate technique, LSM analyses were conducted using a sample of 180 patients with NIHSS assessment at 48-hour post-stroke from the ESCAPE trial. The NIHSS domains were grouped into six categories using two schemes. LSM was conducted for each category of the two groupings and the total NIHSS score. RESULTS: Sub-score LSMs not only identify most of the brain regions that are identified as critical by the total NIHSS score but also reveal additional brain regions critical to each function category of the NIHSS assessment without requiring extensive, specialised assessments. CONCLUSION: These findings show that widely available sub-scores of clinical outcome assessments can be used to investigate more specific structure-function relationships, which may improve predictive modelling of stroke outcomes in the context of modern clinical stroke assessments and neuroimaging. TRIAL REGISTRATION NUMBER: NCT01778335.

Reference-based linear curve fitting for bolus arrival time estimation in 4D MRA and MR perfusion-weighted image sequences.

The bolus arrival time (BAT) based on an indicator dilution curve is an important hemodynamic parameter. As the direct estimation of this parameter is generally problematic, various parametric models have been proposed that describe typical physiological shapes of indicator dilution curves, but it remains unclear which model describes the real physiological background. This article presents a method that indirectly incorporates physiological information derived from the data available. For this, a patient-specific hemodynamic reference curve is extracted, and the corresponding reference BAT is determined. To estimate a BAT for a given signal curve, the reference curve is fitted linearly to the signal curve. The parameters of the fitting process are then used to transfer the reference BAT to the signal curve. The validation of the method proposed based on Monte Carlo simulations showed that the approach presented is capable of improving the BAT estimation precision compared with standard BAT estimation methods by up to 59% while at the same time reduces the computation time. A major benefit of the method proposed is that no assumption about the underlying distribution of indicator dilution has to be made, as it is implicitly modeled in the reference curve.

Quantification of recurrence volumes after endovascular treatment of cerebral aneurysm as surrogate endpoint for treatment stability.

INTRODUCTION: New coils with unproven clinical benefit enlarge the armamentarium for endovascular aneurysm treatment continuously. Large patient numbers needed to detect benefits of such new techniques prevent timely evaluation of efficacy. We propose measuring the volume of aneurysm recurrences as surrogate endpoint for coil stability. We hypothesize that this method allows detecting effects of new materials with reduced sample sizes in comparison to conventional studies with dichotomous endpoints. METHODS: Institutional review board approval and informed consent were obtained. Fifty-nine patients with decreasing aneurysm size and at least two available follow-up time-of-flight magnetic resonance angiographies (ToF-MRAs) were included. Newly developed software for measuring aneurysm volume differences based on ToF-MRA images was used. Based on the observed recurrence volumes and retreatment rates, the sample size for future studies comparing standard versus "new coils" were calculated. RESULTS: Mean recurrence volume was 38.92 µl (SD110.85 µl). To show a 50% reduction of retreatment rate to control (p = 0.05; power 80%) in a regular study (dichotomous endpoint), the required sample size would be n = 356 compared with n = 78 if using the continuous surrogate endpoint "recurrence volume". When extrapolating our data to data given in the literature, sample sizes could be reduced from n = 948 to n = 74 without loss of statistical power. CONCLUSION: Further studies on new materials using volumetric analysis based on ToF-MRA as surrogate endpoint could substantially decrease sample size and allow a more timely assessment of possible benefit of new materials with a fraction of the cost.

Image-based classification of parkinsonian syndromes using T2'-atlases.

Parkinsonian syndromes (PS) are genetically and pathologically heterogeneous neurodegenerative disorders. Clinical distinction between different PS can be difficult, particularly in early disease stages. This paper describes an automatic method for the distinction between classical Parkinson's disease (PD) and progressive supranuclear palsy (PSP) using T2' atlases. This procedure is based on the assumption that regional brain iron content differs between PD and PSP, which can be selectively measured using T2' MR imaging. The proposed method was developed and validated based on 33 PD patients, 10 PSP patients, and 24 healthy controls. The first step of the proposed procedure comprises T2' atlas generation for each group using affine and following non-linear registration. For classification, a T2' dataset is registered to the atlases and compared to each one of them using the mean sum of squared differences metric. The dataset is assigned to the group for which the corresponding atlas yields the lowest value. The evaluation using leave-one-out validation revealed that the proposed method achieves a classification accuracy of 91%. The presented method might serve as the basis for an improved automatic classification of PS in the future.

Integrating expert knowledge for dementia risk prediction in individuals with mild cognitive impairment (MCI): a study protocol.

INTRODUCTION: To date, there is no broadly accepted dementia risk score for use in individuals with mild cognitive impairment (MCI), partly because there are few large datasets available for model development. When evidence is limited, the knowledge and experience of experts becomes more crucial for risk stratification and providing MCI patients with prognosis. Structured expert elicitation (SEE) includes formal methods to quantify experts' beliefs and help experts to express their beliefs in a quantitative form, reducing biases in the process. This study proposes to (1) assess experts' beliefs about important predictors for 3-year dementia risk in persons with MCI through SEE methodology and (2) to integrate expert knowledge and patient data to derive dementia risk scores in persons with MCI using a Bayesian approach. METHODS AND ANALYSIS: This study will use a combination of SEE methodology, prospectively collected clinical data, and statistical modelling to derive a dementia risk score in persons with MCI . Clinical expert knowledge will be quantified using SEE methodology that involves the selection and training of the experts, administration of questionnaire for eliciting expert knowledge, discussion meetings and results aggregation. Patient data from the Prospective Registry for Persons with Memory Symptoms of the Cognitive Neurosciences Clinic at the University of Calgary; the Alzheimer's Disease Neuroimaging Initiative; and the National Alzheimer's Coordinating Center's Uniform Data Set will be used for model training and validation. Bayesian Cox models will be used to incorporate patient data and elicited data to predict 3-year dementia risk. DISCUSSION: This study will develop a robust dementia risk score that incorporates clinician expert knowledge with patient data for accurate risk stratification, prognosis and management of dementia.

Automatic analysis of the anatomy of arteriovenous malformations using 3D and 4D MRA image sequences.

The cerebral arteriovenous malformation (AVM) is an abnormal connection between arteries and veins without capillaries in between, leading to increased blood pressure which might result in a rupture and acute bleeding. Exact knowledge about the patient's individual anatomy of the AVM is needed for improved therapy planning. This paper describes a method for automatic extraction of the AVM and automatic recognition of its feeders and draining veins and en passage vessels based on 3D and 4D MRA image sequences. After registration of the MRA datasets the AVM is segmented using a support vector machine based on blood velocity information, a vesselness measure and the bolus arrival time. The extracted hemodynamic information is then used to detect feeders and draining veins of the AVM. The segmentation of the AVM was validated based on manual segmentations for five patient datasets, whereas a mean Dice value of 0.74 was achieved. The presented hemodynamic characterization was able to detect feeders and draining veins with an accuracy of 100%. In summary the presented approach can improve presurgical planning of AVM surgeries.

Evaluation of methods for bolus arrival time determination using a four-dimensional MRA flow phantom.

In this paper an evaluation of methods determining the bolus arrival time (BAT) using a four-dimensional flow phantom to simulate 4D MR angiography is presented. Spatiotemporal 4D MRA images were acquired for analyzing the hemodynamic characteristics of cerebral vessel anomalies. Model-independent and model-dependent methods for BAT extraction are published. Generally, for the evaluation no gold standard exists and datasets with known BAT values are required. Here, a 4D flow phantom is generated based on a synthetic 3D MRA dataset with BAT values defining the time point of blood inflow for each voxel. Then, voxel-by-voxel concentration-time curves based on the gamma-variate function were computed leading to a simulated 4D MRA dataset. Additionally, partial volume effects and Gaussian noise were integrated. The simulated 4D MRA was visually inspected and regarded as similar to clinical data. Finally, phantom datasets with different vessel diameter and signal-to-noise ratio are computed. Three state-of-the-art methods were used to extract BAT values. Computed and known values were compared. The results suggest that model-dependent approaches perform better than the model-independent method.

Segmentation-Based Blood Flow Parameter Refinement in Cerebrovascular Structures Using 4-D Arterial Spin Labeling MRA.

OBJECTIVE: Cerebrovascular diseases are one of the main global causes of death and disability in the adult population. The preferred imaging modality for the diagnostic routine is digital subtraction angiography, an invasive modality. Time-resolved three-dimensional arterial spin labeling magnetic resonance angiography (4D ASL MRA) is an alternative non-invasive modality, which captures morphological and blood flow data of the cerebrovascular system, with high spatial and temporal resolution. This work proposes advanced medical image processing methods that extract the anatomical and hemodynamic information contained in 4D ASL MRA datasets. METHODS: A previously published segmentation method, which uses blood flow data to improve its accuracy, is extended to estimate blood flow parameters by fitting a mathematical model to the measured vascular signal. The estimated values are then refined using regression techniques within the cerebrovascular segmentation. The proposed method was evaluated using fifteen 4D ASL MRA phantoms, with ground-truth morphological and hemodynamic data, fifteen 4D ASL MRA datasets acquired from healthy volunteers, and two 4D ASL MRA datasets from patients with a stenosis. RESULTS: The proposed method reached an average Dice similarity coefficient of 0.957 and 0.938 in the phantom and real dataset segmentation evaluations, respectively. The estimated blood flow parameter values are more similar to the ground-truth values after the refinement step, when using phantoms. A qualitative analysis showed that the refined blood flow estimation is more realistic compared to the raw hemodynamic parameters. CONCLUSION: The proposed method can provide accurate segmentations and blood flow parameter estimations in the cerebrovascular system using 4D ASL MRA datasets. SIGNIFICANCE: The information obtained with the proposed method can help clinicians and researchers to study the cerebrovascular system non-invasively.

Correction: Improved multi-parametric prediction of tissue outcome in acute ischemic stroke patients using spatial features.

[This corrects the article DOI: 10.1371/journal.pone.0228113.].

Improved multi-parametric prediction of tissue outcome in acute ischemic stroke patients using spatial features.

INTRODUCTION: In recent years, numerous methods have been proposed to predict tissue outcome in acute stroke patients using machine learning methods incorporating multiparametric imaging data. Most methods include diffusion and perfusion parameters as image-based parameters but do not include any spatial information although these parameters are spatially dependent, e.g. different perfusion properties in white and gray brain matter. This study aims to investigate if including spatial features improves the accuracy of multi-parametric tissue outcome prediction. MATERIALS AND METHODS: Acute and follow-up multi-center MRI datasets of 99 patients were available for this study. Logistic regression, random forest, and XGBoost machine learning models were trained and tested using acute MR diffusion and perfusion features and known follow-up lesions. Different combinations of atlas coordinates and lesion probability maps were included as spatial information. The stroke lesion predictions were compared to the true tissue outcomes using the area under the receiver operating characteristic curve (ROC AUC) and the Dice metric. RESULTS: The statistical analysis revealed that including spatial features significantly improves the tissue outcome prediction. Overall, the XGBoost and random forest models performed best in every setting and achieved state-of-the-art results regarding both metrics with similar improvements achieved including Montreal Neurological Institute (MNI) reference space coordinates or voxel-wise lesion probabilities. CONCLUSION: Spatial features should be integrated to improve lesion outcome prediction using machine learning models.

Fuzzy-based extraction of vascular structures from time-of-flight MR images.

In this paper an automatic fuzzy based method for the extraction of the cerebrovascular system from 3D Time-of-Flight (TOF) MRA image sequences is presented. In order to exclude non-brain tissue an automatic skull stripping method is applied in a preprocessing step. Based on the TOF images vesselness and maximum parameter images are computed first. These parameter images are then combined with the TOF sequence using a fuzzy inference. The resulting fuzzy image offers an improved enhancement of small as well as malformed vessels against the remaining brain. Finally, the fuzzy-connectedness approach is used to extract the vascular system. A first evaluation showed that the fuzzy-based method proposed performs better than a state of the art method and yields results in the range of the inter-observer variation.