Fully automated stroke tissue estimation using random forest classifiers (FASTER)

A 25-Year Retrospective of the Use of AI for Diagnosing Acute Stroke: Systematic Review

BACKGROUND

Stroke is a leading cause of death and disability worldwide. Rapid and accurate diagnosis is crucial for minimizing brain damage and optimizing treatment plans.

OBJECTIVE

This review aims to summarize the methods of artificial intelligence (AI)-assisted stroke diagnosis over the past 25 years, providing an overview of performance metrics and algorithm development trends. It also delves into existing issues and future prospects, intending to offer a comprehensive reference for clinical practice.

METHODS

A total of 50 representative articles published between 1999 and 2024 on using AI technology for stroke prevention and diagnosis were systematically selected and analyzed in detail.

RESULTS

AI-assisted stroke diagnosis has made significant advances in stroke lesion segmentation and classification, stroke risk prediction, and stroke prognosis. Before 2012, research mainly focused on segmentation using traditional thresholding and heuristic techniques. From 2012 to 2016, the focus shifted to machine learning (ML)-based approaches. After 2016, the emphasis moved to deep learning (DL), which brought significant improvements in accuracy. In stroke lesion segmentation and classification as well as stroke risk prediction, DL has shown superiority over ML. In stroke prognosis, both DL and ML have shown good performance.

CONCLUSIONS

Over the past 25 years, AI technology has shown promising performance in stroke diagnosis.

Tissue Outcome Prediction in Patients with Proximal Vessel Occlusion and Mechanical Thrombectomy Using Logistic Models

Perfusion CT is established to aid selection of patients with proximal intracranial vessel occlusion for thrombectomy in the extended time window. Selection is mostly based on simple thresholding of perfusion parameter maps, which, however, does not exploit the full information hidden in the high-dimensional perfusion data. We implemented a multiparametric mass-univariate logistic model to predict tissue outcome based on data from 405 stroke patients with acute proximal vessel occlusion in the anterior circulation who underwent mechanical thrombectomy. Input parameters were acute multimodal CT imaging (perfusion, angiography, and non-contrast) as well as basic demographic and clinical parameters. The model was trained with the knowledge of recanalization status and final infarct localization. We found that perfusion parameter maps (CBF, CBV, and Tmax) were sufficient for tissue outcome prediction. Compared with single-parameter thresholding-based models, our logistic model had comparable volumetric accuracy, but was superior with respect to topographical accuracy (AUC of receiver operating characteristic). We also found higher spatial accuracy (Dice index) in an independent internal but not external cross-validation. Our results highlight the value of perfusion data compared with non-contrast CT, CT angiography and clinical information for tissue outcome-prediction. Multiparametric logistic prediction has high potential to outperform the single-parameter thresholding-based approach. In the future, the combination of tissue and functional outcome prediction might provide an individual biomarker for the benefit from mechanical thrombectomy in acute stroke care.

Prediction of tissue outcome in acute ischemic stroke based on single-phase CT angiography at admission

Introduction

In acute ischemic stroke, prediction of the tissue outcome after reperfusion can be used to identify patients that might benefit from mechanical thrombectomy (MT). The aim of this work was to develop a deep learning model that can predict the follow-up infarct location and extent exclusively based on acute single-phase computed tomography angiography (CTA) datasets. In comparison to CT perfusion (CTP), CTA imaging is more widely available, less prone to artifacts, and the established standard of care in acute stroke imaging protocols. Furthermore, recent RCTs have shown that also patients with large established infarctions benefit from MT, which might not have been selected for MT based on CTP core/penumbra mismatch analysis.

Methods

All patients with acute large vessel occlusion of the anterior circulation treated at our institution between 12/2015 and 12/2020 were screened (N = 404) and 238 patients undergoing MT with successful reperfusion were included for final analysis. Ground truth infarct lesions were segmented on 24 h follow-up CT scans. Pre-processed CTA images were used as input for a U-Net-based convolutional neural network trained for lesion prediction, enhanced with a spatial and channel-wise squeeze-and-excitation block. Post-processing was applied to remove small predicted lesion components. The model was evaluated using a 5-fold cross-validation and a separate test set with Dice similarity coefficient (DSC) as the primary metric and average volume error as the secondary metric.

Results

The mean ± standard deviation test set DSC over all folds after post-processing was 0.35 ± 0.2 and the mean test set average volume error was 11.5 mL. The performance was relatively uniform across models with the best model according to the DSC achieved a score of 0.37 ± 0.2 after post-processing and the best model in terms of average volume error yielded 3.9 mL.

Conclusion

24 h follow-up infarct prediction using acute CTA imaging exclusively is feasible with DSC measures comparable to results of CTP-based algorithms reported in other studies. The proposed method might pave the way to a wider acceptance, feasibility, and applicability of follow-up infarct prediction based on artificial intelligence.

Preoperative prediction of early recurrence in resectable pancreatic cancer integrating clinical, radiologic, and CT radiomics features

OBJECTIVES

To use clinical, radiographic, and CT radiomics features to develop and validate a preoperative prediction model for the early recurrence of pancreatic cancer.

METHODS

We retrospectively analyzed 190 patients (150 and 40 in the development and test cohort from different centers) with pancreatic cancer who underwent pancreatectomy between January 2018 and June 2021. Radiomics, clinical-radiologic (CR), and clinical-radiologic-radiomics (CRR) models were developed for the prediction of recurrence within 12 months after surgery. Performance was evaluated using the area under the curve (AUC), Brier score, sensitivity, and specificity.

RESULTS

Early recurrence occurred in 36.7% and 42.5% of the development and test cohorts, respectively (P = 0.62). The features for the CR model included carbohydrate antigen 19-9 > 500 U/mL (odds ratio [OR], 3.60; P = 0.01), abutment to the portal and/or superior mesenteric vein (OR, 2.54; P = 0.054), and adjacent organ invasion (OR, 2.91; P = 0.03). The CRR model demonstrated significantly higher AUCs than the radiomics model in the internal (0.77 vs. 0.73; P = 0.048) and external (0.83 vs. 0.69; P = 0.038) validations. Although we found no significant difference between AUCs of the CR and CRR models (0.83 vs. 0.76; P = 0.17), CRR models showed more balanced sensitivity and specificity (0.65 and 0.87) than CR model (0.41 and 0.91) in the test cohort.

CONCLUSIONS

The CRR model outperformed the radiomics and CR models in predicting the early recurrence of pancreatic cancer, providing valuable information for risk stratification and treatment guidance.

Artificial Intelligence for Clinical Decision Support in Acute Ischemic Stroke: A Systematic Review

BACKGROUND

Established randomized trial-based parameters for acute ischemic stroke group patients into generic treatment groups, leading to attempts using various artificial intelligence (AI) methods to directly correlate patient characteristics to outcomes and thereby provide decision support to stroke clinicians. We review AI-based clinical decision support systems in the development stage, specifically regarding methodological robustness and constraints for clinical implementation.

METHODS

Our systematic review included full-text English language publications proposing a clinical decision support system using AI techniques for direct decision support in acute ischemic stroke cases in adult patients. We (1) describe data and outcomes used in those systems, (2) estimate the systems' benefits compared with traditional stroke diagnosis and treatment, and (3) reported concordance with reporting standards for AI in healthcare.

RESULTS

One hundred twenty-one studies met our inclusion criteria. Sixty-five were included for full extraction. In our sample, utilized data sources, methods, and reporting practices were highly heterogeneous.

CONCLUSIONS

Our results suggest significant validity threats, dissonance in reporting practices, and challenges to clinical translation. We outline practical recommendations for the successful implementation of AI research in acute ischemic stroke treatment and diagnosis.

A Deep Learning-Based Automatic Collateral Assessment in Patients with Acute Ischemic Stroke

This study aimed to develop a supervised deep learning (DL) model for grading collateral status from dynamic susceptibility contrast magnetic resonance perfusion (DSC-MRP) images from patients with large vessel occlusion (LVO) acute ischemic stroke (AIS) and compare its performance against experts' manual grading. Among consecutive LVO-AIS at three medical center sites, DSC-MRP data were processed to generate collateral flow maps consisting of arterial, capillary, and venous phases. With the use of expert readings as a reference, a DL model was developed to analyze collateral status with output classified into good and poor grades. The resulting model was externally validated in a later-collected population from one medical center site. The model was trained on 255 patients and externally validated on 72 patients. In the all-site internal validation population, DL grading of good collateral probability yielded a c statistic of 0.91; in the external validation population, the c statistic was 0.85. In the external validation population, there was moderate agreement between the experts' grades and DL grades (kappa = 0.53, 95% CI = 0.32-0.73, p < 0.0001). Day 7 infarct growth volume was higher in DL-graded poor collateral group than good collateral group patients (median volume [26 mL vs. 6 mL], p = 0.01) in patients with successful reperfusion (modified treatment in cerebral infarction (mTICI) = 2b-3). In all patients with a 90-day modified Rankin Scale (mRS) score, there was a shift to more favorable outcomes in the good collateral group, with a common odds ratio of 2.99 (95% CI = 1.89-4.76, p < 0.0001). The DL-based collateral grading was in good agreement with expert manual grading in both development and validation populations. After exclusion of patients with large infarct volume, early reperfusion is more likely to benefit patients with the poor collateral flow, and the DL method has the potential to aid the assessment of collateral status.

Predicting the tissue outcome of acute ischemic stroke from acute 4D computed tomography perfusion imaging using temporal features and deep learning

Predicting follow-up lesions from baseline CT perfusion (CTP) datasets in acute ischemic stroke patients is important for clinical decision making. Deep convolutional networks (DCNs) are assumed to be the current state-of-the-art for this task. However, many DCN classifiers have not been validated against the methods currently used in research (random decision forests, RDF) and clinical routine (Tmax thresholding). Specialized DCNs have even been designed to extract complex temporal features directly from spatiotemporal CTP data instead of using standard perfusion parameter maps. However, the benefits of applying deep learning to source or deconvolved CTP data compared to perfusion parameter maps have not been formally investigated so far. In this work, a modular UNet-based DCN is proposed that separates temporal feature extraction from tissue outcome prediction, allowing for both model validation using perfusion parameter maps as well as end-to-end learning from spatiotemporal CTP data. 145 retrospective datasets comprising baseline CTP imaging, perfusion parameter maps, and follow-up non-contrast CT with manual lesion segmentations were assembled from acute ischemic stroke patients treated with intravenous thrombolysis alone (IV; n = 43) or intra-arterial mechanical thrombectomy (IA; n = 102) with or without combined IV. Using the perfusion parameter maps as input, the proposed DCN (mean Dice: 0.287) outperformed the RDF (0.262) and simple Tmax-thresholding (0.249). The performance of the proposed DCN was approximately equal using features optimized from the deconvolved residual curves (0.286) compared to perfusion parameter maps (0.287), while using features optimized from the source concentration-time curves (0.296) provided the best tissue outcome predictions.

U-net Models Based on Computed Tomography Perfusion Predict Tissue Outcome in Patients with Different Reperfusion Patterns

Evaluation of cerebral perfusion is important for treatment selection in patients with acute large vessel occlusion (LVO). To assess ischemic core and tissue at risk more accurately, we developed a deep learning model named U-net using computed tomography perfusion (CTP) images. A total of 110 acute ischemic stroke patients undergoing endovascular treatment with major reperfusion (≥ 80%) or minimal reperfusion (≤ 20%) were included. Using baseline CTP, we developed two U-net models: one model in major reperfusion group to identify infarct core; the other in minimal reperfusion group to identify tissue at risk. The performance of fixed-thresholding methods was compared with that of U-net models. In the major reperfusion group, the model estimated infarct core with a Dice score coefficient (DSC) of 0.61 and an area under the curve (AUC) of 0.92, while fixed-thresholding methods had a DSC of 0.52. In the minimal reperfusion group, the model estimated tissue at risk with a DSC of 0.67 and an AUC of 0.93, while fixed-thresholding methods had a DSC of 0.51. In both groups, excellent volumetric consistency (intraclass correlation coefficient was 0.951 in major reperfusion and 0.746 in minimal reperfusion) was achieved between the estimated lesion and the actual lesion volume. Thus, in patients with anterior LVO, the CTP-based U-net models were able to identify infarct core and tissue at risk on baseline CTP superior to fixed-thresholding methods, providing individualized prediction of final lesion in patients with different reperfusion patterns.

Machine learning based analysis of stroke lesions on mouse tissue sections

An unbiased, automated and reliable method for analysis of brain lesions in tissue after ischemic stroke is missing. Manual infarct volumetry or by threshold-based semi-automated approaches is laborious, and biased to human error or biased by many false -positive and -negative data, respectively. Thereby, we developed a novel machine learning, atlas-based method for fully automated stroke analysis in mouse brain slices stained with 2% Triphenyltetrazolium-chloride (2% TTC), named "StrokeAnalyst", which runs on a user-friendly graphical interface. StrokeAnalyst registers subject images on a common spatial domain (a novel mouse TTC- brain atlas of 80 average mathematical images), calculates pixel-based, tissue-intensity statistics (z-scores), applies outlier-detection and machine learning (Random-Forest) models to increase accuracy of lesion detection, and produces volumetry data and detailed neuroanatomical information per lesion. We validated StrokeAnalyst in two separate experimental sets using the filament stroke model. StrokeAnalyst detects stroke lesions in a rater-independent and reproducible way, correctly detects hemispheric volumes even in presence of post-stroke edema and significantly minimizes false-positive errors compared to threshold-based approaches (false-positive rate 1.2-2.3%, p < 0.05). It can process scanner-acquired, and even smartphone-captured or pdf-retrieved images. Overall, StrokeAnalyst surpasses all previous TTC-volumetry approaches and increases quality, reproducibility and reliability of stroke detection in relevant preclinical models.

Performance of Machine Learning for Tissue Outcome Prediction in Acute Ischemic Stroke: A Systematic Review and Meta-Analysis

Machine learning (ML) has been proposed for lesion segmentation in acute ischemic stroke (AIS). This study aimed to provide a systematic review and meta-analysis of the overall performance of current ML algorithms for final infarct prediction from baseline imaging. We made a comprehensive literature search on eligible studies developing ML models for core infarcted tissue estimation on admission CT or MRI in AIS patients. Eleven studies meeting the inclusion criteria were included in the quantitative analysis. Study characteristics, model methodology, and predictive performance of the included studies were extracted. A meta-analysis was conducted on the dice similarity coefficient (DSC) score by using a random-effects model to assess the overall predictive performance. Study heterogeneity was assessed by Cochrane Q and Higgins I² tests. The pooled DSC score of the included ML models was 0.50 (95% CI 0.39–0.61), with high heterogeneity observed across studies (I² 96.5%, p < 0.001). Sensitivity analyses using the one-study removed method showed the adjusted overall DSC score ranged from 0.47 to 0.52. Subgroup analyses indicated that the DL-based models outperformed the conventional ML classifiers with the best performance observed in DL algorithms combined with CT data. Despite the presence of heterogeneity, current ML-based approaches for final infarct prediction showed moderate but promising performance. Before well integrated into clinical stroke workflow, future investigations are suggested to train ML models on large-scale, multi-vendor data, validate on external cohorts and adopt formalized reporting standards for improving model accuracy and robustness.

FeMA: Feature matching auto-encoder for predicting ischaemic stroke evolution and treatment outcome

Although, predicting ischaemic stroke evolution and treatment outcome provide important information one step towards individual treatment planning, classifying functional outcome and modelling the brain tissue evolution remains a challenge due to data complexity and visually subtle changes in the brain. We propose a novel deep learning approach, Feature Matching Auto-encoder (FeMA) that consists of two stages, predicting ischaemic stroke evolution at one week without voxel-wise annotation and predicting ischaemic stroke treatment outcome at 90 days from a baseline scan. In the first stage, we introduce feature similarity and consistency objective, and in the second stage, we show that adding stroke evolution information increase the performance of functional outcome prediction. Comparative experiments demonstrate that our proposed method is more effective to extract representative follow-up features and achieves the best results for functional outcome of stroke treatment.

Deep Learning Applications for Acute Stroke Management

Brain imaging is essential to the clinical care of patients with stroke, a leading cause of disability and death worldwide. Whereas advanced neuroimaging techniques offer opportunities for aiding acute stroke management, several factors, including time delays, inter-clinician variability, and lack of systemic conglomeration of clinical information, hinder their maximal utility. Recent advances in deep machine learning (DL) offer new strategies for harnessing computational medical image analysis to inform decision making in acute stroke. We examine the current state of the field for DL models in stroke triage. First, we provide a brief, clinical practice-focused primer on DL. Next, we examine real-world examples of DL applications in pixel-wise labeling, volumetric lesion segmentation, stroke detection, and prediction of tissue fate postintervention. We evaluate recent deployments of deep neural networks and their ability to automatically select relevant clinical features for acute decision making, reduce inter-rater variability, and boost reliability in rapid neuroimaging assessments, and integrate neuroimaging with electronic medical record (EMR) data in order to support clinicians in routine and triage stroke management. Ultimately, we aim to provide a framework for critically evaluating existing automated approaches, thus equipping clinicians with the ability to understand and potentially apply DL approaches in order to address challenges in clinical practice. ANN NEUROL 2022.

Computational Approaches for Acute Traumatic Brain Injury Image Recognition

In recent years, there have been major advances in deep learning algorithms for image recognition in traumatic brain injury (TBI). Interest in this area has increased due to the potential for greater objectivity, reduced interpretation times and, ultimately, higher accuracy. Triage algorithms that can re-order radiological reading queues have been developed, using classification to prioritize exams with suspected critical findings. Localization models move a step further to capture more granular information such as the location and, in some cases, size and subtype, of intracranial hematomas that could aid in neurosurgical management decisions. In addition to the potential to improve the clinical management of TBI patients, the use of algorithms for the interpretation of medical images may play a transformative role in enabling the integration of medical images into precision medicine. Acute TBI is one practical example that can illustrate the application of deep learning to medical imaging. This review provides an overview of computational approaches that have been proposed for the detection and characterization of acute TBI imaging abnormalities, including intracranial hemorrhage, skull fractures, intracranial mass effect, and stroke.

Detection of Collaterals from Cone-Beam CT Images in Stroke

Collateral vessels play an important role in the restoration of blood flow to the ischemic tissues of stroke patients, and the quality of collateral flow has major impact on reducing treatment delay and increasing the success rate of reperfusion. Due to high spatial resolution and rapid scan time, advance imaging using the cone-beam computed tomography (CBCT) is gaining more attention over the conventional angiography in acute stroke diagnosis. Detecting collateral vessels from CBCT images is a challenging task due to the presence of noises and artifacts, small-size and non-uniform structure of vessels. This paper presents a technique to objectively identify collateral vessels from non-collateral vessels. In our technique, several filters are used on the CBCT images of stroke patients to remove noises and artifacts, then multiscale top-hat transformation method is implemented on the pre-processed images to further enhance the vessels. Next, we applied three types of feature extraction methods which are gray level co-occurrence matrix (GLCM), moment invariant, and shape to explore which feature is best to classify the collateral vessels. These features are then used by the support vector machine (SVM), random forest, decision tree, and K-nearest neighbors (KNN) classifiers to classify vessels. Finally, the performance of these classifiers is evaluated in terms of accuracy, sensitivity, precision, recall, F-Measure, and area under the receiver operating characteristics curve. Our results show that all classifiers achieve promising classification accuracy above 90% and able to detect the collateral and non-collateral vessels from images.

Prediction of Tissue Damage Using a User-Independent Machine Learning Algorithm vs. Tmax Threshold Maps

(1) Background: To test the accuracy of a fully automated stroke tissue estimation algorithm (FASTER) to predict final lesion volumes in an independent dataset in patients with acute stroke; (2) Methods: Tissue-at-risk prediction was performed in 31 stroke patients presenting with a proximal middle cerebral artery occlusion. FDA-cleared perfusion software using the AHA recommendation for the Tmax threshold delay was tested against a prediction algorithm trained on an independent perfusion software using artificial intelligence (FASTER). Following our endovascular strategy to consequently achieve TICI 3 outcome, we compared patients with complete reperfusion (TICI 3) vs. no reperfusion (TICI 0) after mechanical thrombectomy. Final infarct volume was determined on a routine follow-up MRI or CT at 90 days after the stroke; (3) Results: Compared to the reference standard (infarct volume after 90 days), the decision forest algorithm overestimated the final infarct volume in patients without reperfusion. Underestimation was observed if patients were completely reperfused. In cases where the FDA-cleared segmentation was not interpretable due to improper definitions of the arterial input function, the decision forest provided reliable results; (4) Conclusions: The prediction accuracy of automated tissue estimation depends on (i) success of reperfusion, (ii) infarct size, and (iii) software-related factors introduced by the training sample. A principal advantage of machine learning algorithms is their improved robustness to artifacts in comparison to solely threshold-based model-dependent software. Validation on independent datasets remains a crucial condition for clinical implementations of decision support systems in stroke imaging.

Tissue outcome prediction in hyperacute ischemic stroke: Comparison of machine learning models

Machine Learning (ML) has been proposed for tissue fate prediction after acute ischemic stroke (AIS), with the aim to help treatment decision and patient management. We compared three different ML models to the clinical method based on diffusion-perfusion thresholding for the voxel-based prediction of final infarct, using a large MRI dataset obtained in a cohort of AIS patients prior to recanalization treatment. Baseline MRI (MRI₀), including diffusion-weighted sequence (DWI) and Tmax maps from perfusion-weighted sequence, and 24-hr follow-up MRI (MRI_24h) were retrospectively collected in consecutive 394 patients AIS patients (median age = 70 years; final infarct volume = 28mL). Manually segmented DWI_24h lesion was considered the final infarct. Gradient Boosting, Random Forests and U-Net were trained using DWI, apparent diffusion coefficient (ADC) and Tmax maps on MRI₀ as inputs to predict final infarct. Tissue outcome predictions were compared to final infarct using Dice score. Gradient Boosting had significantly better predictive performance (median [IQR] Dice Score as for median age, maybe you can replace the comma with an equal sign for consistency 0.53 [0.29-0.68]) than U-Net (0.48 [0.18-0.68]), Random Forests (0.51 [0.27-0.66]), and clinical thresholding method (0.45 [0.25-0.62]) (P < 0.001). In this benchmark of ML models for tissue outcome prediction in AIS, Gradient Boosting outperformed other ML models and clinical thresholding method and is thus promising for future decision-making.

autoTICI: Automatic Brain Tissue Reperfusion Scoring on 2D DSA Images of Acute Ischemic Stroke Patients

The Thrombolysis in Cerebral Infarction (TICI) score is an important metric for reperfusion therapy assessment in acute ischemic stroke. It is commonly used as a technical outcome measure after endovascular treatment (EVT). Existing TICI scores are defined in coarse ordinal grades based on visual inspection, leading to inter- and intra-observer variation. In this work, we present autoTICI, an automatic and quantitative TICI scoring method. First, each digital subtraction angiography (DSA) acquisition is separated into four phases (non-contrast, arterial, parenchymal and venous phase) using a multi-path convolutional neural network (CNN), which exploits spatio-temporal features. The network also incorporates sequence level label dependencies in the form of a state-transition matrix. Next, a minimum intensity map (MINIP) is computed using the motion corrected arterial and parenchymal frames. On the MINIP image, vessel, perfusion and background pixels are segmented. Finally, we quantify the autoTICI score as the ratio of reperfused pixels after EVT. On a routinely acquired multi-center dataset, the proposed autoTICI shows good correlation with the extended TICI (eTICI) reference with an average area under the curve (AUC) score of 0.81. The AUC score is 0.90 with respect to the dichotomized eTICI. In terms of clinical outcome prediction, we demonstrate that autoTICI is overall comparable to eTICI.

Emerging role of artificial intelligence in stroke imaging

Introduction: The recognition and therapy of patients with stroke is becoming progressively intricate as additional treatment choices become accessible and new associations between disease characteristics and treatment response are incessantly uncovered. Therefore, clinicians must regularly learn new skill, stay up to date with the literature and integrate advances into daily practice. The application of artificial intelligence (AI) to assist clinical decision making could diminish inter-rater variation in routine clinical practice and accelerate the mining of vital data that could expand recognition of patients with stroke, forecast of treatment responses and patient outcomes.Areas covered: In this review, the authors provide an up-to-date review of AI in stroke, analyzing the latest papers on this subject. These have been divided in two main groups: stroke diagnosis and outcome prediction.Expert opinion: The highest value of AI is its capability to merge, select and condense a large amount of clinical and imaging features of a single patient and to associate these with fitted models that have gone through robust assessment and optimization with large cohorts of data to support clinical decision making.

Predictive modelling of hypoxic ischaemic encephalopathy risk following perinatal asphyxia

Hypoxic Ischemic Encephalopathy (HIE) remains a major cause of neurological disability. Early intervention with therapeutic hypothermia improves outcome, but prediction of HIE is difficult and no single clinical marker is reliable. Machine learning algorithms may allow identification of patterns in clinical data to improve prognostic power. Here we examine the use of a Random Forest machine learning algorithm and five-fold cross-validation to predict the occurrence of HIE in a prospective cohort of infants with perinatal asphyxia. Infants with perinatal asphyxia were recruited at birth and neonatal course was followed for the development of HIE. Clinical variables were recorded for each infant including maternal demographics, delivery details and infant's condition at birth. We found that the strongest predictors of HIE were the infant's condition at birth (as expressed by Apgar score), need for resuscitation, and the first postnatal measures of pH, lactate, and base deficit. Random Forest models combining features including Apgar score, most intensive resuscitation, maternal age and infant birth weight both with and without biochemical markers of pH, lactate, and base deficit resulted in a sensitivity of 56-100% and a specificity of 78-99%. This study presents a dynamic method of rapid classification that has the potential to be easily adapted and implemented in a clinical setting, with and without the availability of blood gas analysis. Our results demonstrate that applying machine learning algorithms to readily available clinical data may support clinicians in the early and accurate identification of infants who will develop HIE. We anticipate our models to be a starting point for the development of a more sophisticated clinical decision support system to help identify which infants will benefit from early therapeutic hypothermia.

Multi-view iterative random walker for automated salvageable tissue delineation in ischemic stroke from multi-sequence MRI

BACKGROUND AND OBJECTIVE

Non-invasive and robust identification of salvageable tissue (penumbra) is crucial for interventional stroke therapy. Besides identifying stroke injury as a whole, the ability to automatically differentiate core and penumbra tissues, using both diffusion and perfusion magnetic resonance imaging (MRI) sequences is essential for ischemic stroke treatment.

METHOD

A fully automated and novel one-shot multi-view iterative random walker (MIRW) method with an automated injury seed point detection is developed for lesion delineation. MIRW utilizes the heirarchical decomposition of multi-sequence MRI physical properties of the underlying tissue within the lesion to maximize the inter-class variations of the volumetric histogram to estimate the probable seed points. These estimates are further utilized to conglomerate the lesion estimations iteratively from axial, coronal and sagittal MRI volumes for a computationally efficient segmentation and quantification of salvageable and necrotic tissues from multi-sequence MRI.

RESULTS

Comprehensive experimental analysis of MIRW is performed on three challenging adult(sub-)acute ischemic stroke datasets using performance measures like precision, sensitivity, specificity and Dice similarity score (DSC), which are computed with respect to the manual ground-truth.

COMPARISON WITH EXISTING METHODS

MIRW method resulted in a high DSC of 83.5% in a very less computational time of 98.23 s/volume, which is a significant improvement on the ISLES benchmark dataset for penumbra detection, compared to the state-of-the-art techniques.

CONCLUSION

Quantitative measures demonstrate the promising potential of MIRW for computational analysis of adult stroke and quantifying penumbra in stroke patients which is essential for selecting the good candidates for recanalization.

Segmentation of MRI head anatomy using deep volumetric networks and multiple spatial priors

Purpose: Conventional automated segmentation of the head anatomy in magnetic resonance images distinguishes different brain and nonbrain tissues based on image intensities and prior tissue probability maps (TPMs). This works well for normal head anatomies but fails in the presence of unexpected lesions. Deep convolutional neural networks (CNNs) leverage instead spatial patterns and can learn to segment lesions but often ignore prior probabilities. Approach: We add three sources of prior information to a three-dimensional (3D) convolutional network, namely, spatial priors with a TPM, morphological priors with conditional random fields, and spatial context with a wider field-of-view at lower resolution. We train and test these networks on 3D images of 43 stroke patients and 4 healthy individuals which have been manually segmented. Results: We demonstrate the benefits of each source of prior information, and we show that the new architecture, which we call Multiprior network, improves the performance of existing segmentation software, such as SPM, FSL, and DeepMedic for abnormal anatomies. The relevance of the different priors was compared, and the TPM was found to be most beneficial. The benefit of adding a TPM is generic in that it can boost the performance of established segmentation networks such as the DeepMedic and a UNet. We also provide an out-of-sample validation and clinical application of the approach on an additional 47 patients with disorders of consciousness. We make the code and trained networks freely available. Conclusions: Biomedical images follow imaging protocols that can be leveraged as prior information into deep CNNs to improve performance. The network segmentations match human manual corrections performed in 3D and are comparable in performance to human segmentations obtained from scratch in 2D for abnormal brain anatomies.

Tissue at Risk and Ischemic Core Estimation Using Deep Learning in Acute Stroke

BACKGROUND AND PURPOSE

In acute stroke patients with large vessel occlusions, it would be helpful to be able to predict the difference in the size and location of the final infarct based on the outcome of reperfusion therapy. Our aim was to demonstrate the value of deep learning-based tissue at risk and ischemic core estimation. We trained deep learning models using a baseline MR image in 3 multicenter trials.

MATERIALS AND METHODS

Patients with acute ischemic stroke from 3 multicenter trials were identified and grouped into minimal (≤20%), partial (20%-80%), and major (≥80%) reperfusion status based on 4- to 24-hour follow-up MR imaging if available or into unknown status if not. Attention-gated convolutional neural networks were trained with admission imaging as input and the final infarct as ground truth. We explored 3 approaches: 1) separate: train 2 independent models with patients with minimal and major reperfusion; 2) pretraining: develop a single model using patients with partial and unknown reperfusion, then fine-tune it to create 2 separate models for minimal and major reperfusion; and 3) thresholding: use the current clinical method relying on apparent diffusion coefficient and time-to-maximum of the residue function maps. Models were evaluated using area under the curve, the Dice score coefficient, and lesion volume difference.

RESULTS

Two hundred thirty-seven patients were included (minimal, major, partial, and unknown reperfusion: n = 52, 80, 57, and 48, respectively). The pretraining approach achieved the highest median Dice score coefficient (tissue at risk = 0.60, interquartile range, 0.43-0.70; core = 0.57, interquartile range, 0.30-0.69). This was higher than the separate approach (tissue at risk = 0.55; interquartile range, 0.41-0.69; P = .01; core = 0.49; interquartile range, 0.35-0.66; P = .04) or thresholding (tissue at risk = 0.56; interquartile range, 0.42-0.65; P = .008; core = 0.46; interquartile range, 0.16-0.54; P < .001).

CONCLUSIONS

Deep learning models with fine-tuning lead to better performance for predicting tissue at risk and ischemic core, outperforming conventional thresholding methods.

Artificial intelligence in clinical decision support and outcome prediction - applications in stroke

Artificial intelligence (AI) is making a profound impact in healthcare, with the number of AI applications in medicine increasing substantially over the past five years. In acute stroke, it is playing an increasingly important role in clinical decision-making. Contemporary advances have increased the amount of information - both clinical and radiological - which clinicians must consider when managing patients. In the time-critical setting of acute stroke, AI offers the tools to rapidly evaluate and consolidate available information, extracting specific predictions from rich, noisy data. It has been applied to the automatic detection of stroke lesions on imaging and can guide treatment decisions through the prediction of tissue outcomes and long-term functional outcomes. This review examines the current state of AI applications in stroke, exploring their potential to reform stroke care through clinical decision support, as well as the challenges and limitations which must be addressed to facilitate their acceptance and adoption for clinical use.

Random forest-based prediction of stroke outcome

We research into the clinical, biochemical and neuroimaging factors associated with the outcome of stroke patients to generate a predictive model using machine learning techniques for prediction of mortality and morbidity 3-months after admission. The dataset consisted of patients with ischemic stroke (IS) and non-traumatic intracerebral hemorrhage (ICH) admitted to Stroke Unit of a European Tertiary Hospital prospectively registered. We identified the main variables for machine learning Random Forest (RF), generating a predictive model that can estimate patient mortality/morbidity according to the following groups: (1) IS + ICH, (2) IS, and (3) ICH. A total of 6022 patients were included: 4922 (mean age 71.9 ± 13.8 years) with IS and 1100 (mean age 73.3 ± 13.1 years) with ICH. NIHSS at 24, 48 h and axillary temperature at admission were the most important variables to consider for evolution of patients at 3-months. IS + ICH group was the most stable for mortality prediction [0.904 ± 0.025 of area under the receiver operating characteristics curve (AUC)]. IS group presented similar results, although variability between experiments was slightly higher (0.909 ± 0.032 of AUC). ICH group was the one in which RF had more problems to make adequate predictions (0.9837 vs. 0.7104 of AUC). There were no major differences between IS and IS + ICH groups according to morbidity prediction (0.738 and 0.755 of AUC) but, after checking normality with a Shapiro Wilk test with the null hypothesis that the data follow a normal distribution, it was rejected with W = 0.93546 (p-value < 2.2e-16). Conditions required for a parametric test do not hold, and we performed a paired Wilcoxon Test assuming the null hypothesis that all the groups have the same performance. The null hypothesis was rejected with a value < 2.2e-16, so there are statistical differences between IS and ICH groups. In conclusion, machine learning algorithms RF can be effectively used in stroke patients for long-term outcome prediction of mortality and morbidity.

Advances in imaging acute ischemic stroke: evaluation before thrombectomy

Recent advances in neuroimaging have demonstrated significant assessment benefits and appropriate triage of patients based on specific clinical and radiological features in the acute stroke setting. Endovascular thrombectomy is arguably the most important aspect of acute stroke management with an extended time window. Imaging-based physiological information may potentially shift the treatment paradigm from a rigid time-based model to a more flexible and individualized, tissue-based approach, increasing the proportion of patients amenable to treatment. Various imaging modalities are routinely used in the diagnosis and management of acute ischemic stroke, including multimodal computed tomography (CT) and magnetic resonance imaging (MRI). Therefore, these imaging methods should provide information beyond the presence or absence of intracranial hemorrhage as well as the presence and extent of the ischemic core, collateral circulation and penumbra in patients with neurological symptoms. Target mismatch may optimize selection of patients with late or unknown symptom onset who would potentially be eligible for revascularization therapy. The purpose of this study was to provide a comprehensive review of the current evidence about efficacy and theoretical basis of present imaging modalities, and explores future directions for imaging in the management of acute ischemic stroke.

Radiomics in Stroke Neuroimaging: Techniques, Applications, and Challenges

Stroke is a leading cause of disability and mortality worldwide, resulting in substantial economic costs for post-stroke care each year. Neuroimaging, such as cranial computed tomography or magnetic resonance imaging, is the backbone of stroke management strategies, which can guide treatment decision-making (thrombolysis or hemostasis) at an early stage. With advances in computational technologies, particularly in machine learning, visual image information can now be converted into numerous quantitative features in an objective, repeatable, and high-throughput manner, in a process known as radiomics. Radiomics is mainly used in the field of oncology, which remains an area of active research. Over the past few years, investigators have attempted to apply radiomics to stroke in the hope of gaining benefits similar to those obtained in cancer management, i.e., in promoting the development of personalized precision medicine. Currently, radiomic analysis has shown promise for a variety of applications in stroke, including the diagnosis of stroke lesions, early prediction of outcomes, and evaluation for long-term prognosis. In this article, we elaborate the contributions of radiomics to stroke, as well as the subprocesses and techniques involved in radiomics studies. We also discuss the potential challenges facing its widespread implementation in routine practice and the directions for future research.

Automatic brain lesion segmentation on standard magnetic resonance images: a scoping review

OBJECTIVES

Medical image analysis practices face challenges that can potentially be addressed with algorithm-based segmentation tools. In this study, we map the field of automatic MR brain lesion segmentation to understand the clinical applicability of prevalent methods and study designs, as well as challenges and limitations in the field.

DESIGN

Scoping review.

SETTING

Three databases (PubMed, IEEE Xplore and Scopus) were searched with tailored queries. Studies were included based on predefined criteria. Emerging themes during consecutive title, abstract, methods and whole-text screening were identified. The full-text analysis focused on materials, preprocessing, performance evaluation and comparison.

RESULTS

Out of 2990 unique articles identified through the search, 441 articles met the eligibility criteria, with an estimated growth rate of 10% per year. We present a general overview and trends in the field with regard to publication sources, segmentation principles used and types of lesions. Algorithms are predominantly evaluated by measuring the agreement of segmentation results with a trusted reference. Few articles describe measures of clinical validity.

CONCLUSIONS

The observed reporting practices leave room for improvement with a view to studying replication, method comparison and clinical applicability. To promote this improvement, we propose a list of recommendations for future studies in the field.

Machine learning identifies stroke features between species

Identification and localization of ischemic stroke (IS) lesions is routinely performed to confirm diagnosis, assess stroke severity, predict disability and plan rehabilitation strategies using magnetic resonance imaging (MRI). In basic research, stroke lesion segmentation is necessary to study complex peri-infarction tissue changes. Moreover, final stroke volume is a critical outcome evaluated in clinical and preclinical experiments to determine therapy or intervention success. Manual segmentations are performed but they require a specialized skill set, are prone to inter-observer variation, are not entirely objective and are often not supported by histology. The task is even more challenging when dealing with large multi-center datasets, multiple experimenters or large animal cohorts. On the other hand, current automatized segmentation approaches often lack histological validation, are not entirely user independent, are often based on single parameters, or in the case of complex machine learning methods, require vast training datasets and are prone to a lack of model interpretation.

Methods: We induced IS using the middle cerebral artery occlusion model on two rat cohorts. We acquired apparent diffusion coefficient (ADC) and T2-weighted (T2W) images at 24 h and 1-week after IS induction. Subsets of the animals at 24 h and 1-week post IS were evaluated using histology and immunohistochemistry. Using a Gaussian mixture model, we segmented voxel-wise interactions between ADC and T2W parameters at 24 h using one of the rat cohorts. We then used these segmentation results to train a random forest classifier, which we applied to the second rat cohort. The algorithms' stroke segmentations were compared to manual stroke delineations, T2W and ADC thresholding methods and the final stroke segmentation at 1-week. Volume correlations to histology were also performed for every segmentation method. Metrics of success were calculated with respect to the final stroke volume. Finally, the trained random forest classifier was tested on a human dataset with a similar temporal stroke on-set. Manual segmentations, ADC and T2W thresholds were again used to evaluate and perform comparisons with the proposed algorithms' output.

Results: In preclinical rat data our framework significantly outperformed commonly applied automatized thresholding approaches and segmented stroke regions similarly to manual delineation. The framework predicted the localization of final stroke regions in 1-week post-stroke MRI with a median Dice similarity coefficient of 0.86, Matthew's correlation coefficient of 0.80 and false positive rate of 0.04. The predicted stroke volumes also strongly correlated with final histological stroke regions (Pearson correlation = 0.88, P < 0.0001). Lastly, the stroke region characteristics identified by our framework in rats also identified stroke lesions in human brains, largely outperforming thresholding approaches in stroke volume prediction (P<0.01).

Conclusion: Our findings reveal that the segmentation produced by our proposed framework using 24 h MRI rat data strongly correlated with the final stroke volume, denoting a predictive effect. In addition, we show for the first time that the stroke imaging features can be directly translated between species, allowing identification of acute stroke in humans using the model trained on animal data. This discovery reduces the gap between the clinical and preclinical fields, unveiling a novel approach to directly co-analyze clinical and preclinical data. Such methods can provide further biological insights into human stroke and highlight the differences between species in order to help improve the experimental setups and animal models of the disease.

Impact of the reperfusion status for predicting the final stroke infarct using deep learning

BACKGROUND

Predictive maps of the final infarct may help therapeutic decisions in acute ischemic stroke patients. Our objectives were to assess whether integrating the reperfusion status into deep learning models would improve their performance, and to compare them to current clinical prediction methods.

METHODS

We trained and tested convolutional neural networks (CNNs) to predict the final infarct in acute ischemic stroke patients treated by thrombectomy in our center. When training the CNNs, non-reperfused patients from a non-thrombectomized cohort were added to the training set to increase the size of this group. Baseline diffusion and perfusion-weighted magnetic resonance imaging (MRI) were used as inputs, and the lesion segmented on day-6 MRI served as the ground truth for the final infarct. The cohort was dichotomized into two subsets, reperfused and non-reperfused patients, from which reperfusion status specific CNNs were developed and compared to one another, and to the clinically-used perfusion-diffusion mismatch model. Evaluation metrics included the Dice similarity coefficient (DSC), precision, recall, volumetric similarity, Hausdorff distance and area-under-the-curve (AUC).

RESULTS

We analyzed 109 patients, including 35 without reperfusion. The highest DSC were achieved in both reperfused and non-reperfused patients (DSC = 0.44 ± 0.25 and 0.47 ± 0.17, respectively) when using the corresponding reperfusion status-specific CNN. CNN-based models achieved higher DSC and AUC values compared to those of perfusion-diffusion mismatch models (reperfused patients: AUC = 0.87 ± 0.13 vs 0.79 ± 0.17, P < 0.001; non-reperfused patients: AUC = 0.81 ± 0.13 vs 0.73 ± 0.14, P < 0.01, in CNN vs perfusion-diffusion mismatch models, respectively).

CONCLUSION

The performance of deep learning models improved when the reperfusion status was incorporated in their training. CNN-based models outperformed the clinically-used perfusion-diffusion mismatch model. Comparing the predicted infarct in case of successful vs failed reperfusion may help in estimating the treatment effect and guiding therapeutic decisions in selected patients.

Combining unsupervised and supervised learning for predicting the final stroke lesion

Predicting the final ischaemic stroke lesion provides crucial information regarding the volume of salvageable hypoperfused tissue, which helps physicians in the difficult decision-making process of treatment planning and intervention. Treatment selection is influenced by clinical diagnosis, which requires delineating the stroke lesion, as well as characterising cerebral blood flow dynamics using neuroimaging acquisitions. Nonetheless, predicting the final stroke lesion is an intricate task, due to the variability in lesion size, shape, location and the underlying cerebral haemodynamic processes that occur after the ischaemic stroke takes place. Moreover, since elapsed time between stroke and treatment is related to the loss of brain tissue, assessing and predicting the final stroke lesion needs to be performed in a short period of time, which makes the task even more complex. Therefore, there is a need for automatic methods that predict the final stroke lesion and support physicians in the treatment decision process. We propose a fully automatic deep learning method based on unsupervised and supervised learning to predict the final stroke lesion after 90 days. Our aim is to predict the final stroke lesion location and extent, taking into account the underlying cerebral blood flow dynamics that can influence the prediction. To achieve this, we propose a two-branch Restricted Boltzmann Machine, which provides specialized data-driven features from different sets of standard parametric Magnetic Resonance Imaging maps. These data-driven feature maps are then combined with the parametric Magnetic Resonance Imaging maps, and fed to a Convolutional and Recurrent Neural Network architecture. We evaluated our proposal on the publicly available ISLES 2017 testing dataset, reaching a Dice score of 0.38, Hausdorff Distance of 29.21 mm, and Average Symmetric Surface Distance of 5.52 mm.

Fueling Clinical and Translational Research in Appalachia: Informatics Platform Approach

BACKGROUND

The Appalachian population is distinct, not just culturally and geographically but also in its health care needs, facing the most health care disparities in the United States. To meet these unique demands, Appalachian medical centers need an arsenal of analytics and data science tools with the foundation of a centralized data warehouse to transform health care data into actionable clinical interventions. However, this is an especially challenging task given the fragmented state of medical data within Appalachia and the need for integration of other types of data such as environmental, social, and economic with medical data.

OBJECTIVE

This paper aims to present the structure and process of the development of an integrated platform at a midlevel Appalachian academic medical center along with its initial uses.

METHODS

The Appalachian Informatics Platform was developed by the Appalachian Clinical and Translational Science Institute's Division of Clinical Informatics and consists of 4 major components: a centralized clinical data warehouse, modeling (statistical and machine learning), visualization, and model evaluation. Data from different clinical systems, billing systems, and state- or national-level data sets were integrated into a centralized data warehouse. The platform supports research efforts by enabling curation and analysis of data using the different components, as appropriate.

RESULTS

The Appalachian Informatics Platform is functional and has supported several research efforts since its implementation for a variety of purposes, such as increasing knowledge of the pathophysiology of diseases, risk identification, risk prediction, and health care resource utilization research and estimation of the economic impact of diseases.

CONCLUSIONS

The platform provides an inexpensive yet seamless way to translate clinical and translational research ideas into clinical applications for regions similar to Appalachia that have limited resources and a largely rural population.

Thrombolysis in Cerebral Infarction 2b Reperfusions: To Treat or to Stop?

In patients undergoing mechanical thrombectomy, achieving complete (Thrombolysis in Cerebral Infarction 3) rather than incomplete successful reperfusion (Thrombolysis in Cerebral Infarction 2b) is associated with better functional outcome. Despite technical improvements, incomplete reperfusion remains the final angiographic result in 40% of patients according to recent trials. As most incomplete reperfusions are caused by distal vessel occlusions, they are potentially amenable to rescue strategies. While observational data suggest a net benefit of up to 20% in functional independence of incomplete versus complete reperfusions, the net benefit of secondary improvement from Thrombolysis in Cerebral Infarction 2b to 3 reperfusion might differ due to lengthier procedures and delayed reperfusion. Current strategies to tackle distal vessel occlusions consist of distal (microcatheter) aspiration, small adjustable stent retrievers, and administration of intra-arterial thrombolytics. While there are promising reports evaluating those techniques, all available studies show relevant limitations in terms of selection bias, single-center design, or nonconsecutive patient inclusion. Besides an assessment of risks associated with rescue maneuvers, we advocate that the decision-making process should also include a consideration of potential outcomes if complete reperfusion would successfully be achieved. These include (1) a futile angiographic improvement (hypoperfused territory is already infarcted), (2) an unnecessary angiographic improvement (the patient would not have developed infarction if no rescue maneuver was performed), and (3) a successful rescue maneuver with clinical benefit. Currently there is paucity of data on how these scenarios can be predicted and the decision whether to treat or to stop in a patient with incomplete reperfusion involves many unknowns. To advance the status quo, we outline current knowledge gaps and avenues of potential research regarding this clinically important question.

Artificial intelligence in stroke imaging: Current and future perspectives

Artificial intelligence (AI) is a fast-growing research area in computer science that aims to mimic cognitive processes through a number of techniques. Supervised machine learning, a subfield of AI, includes methods that can identify patterns in high-dimensional data using labeled 'ground truth' data and apply these learnt patterns to analyze, interpret, or make predictions on new datasets. Supervised machine learning has become a significant area of interest within the medical community. Radiology and neuroradiology in particular are especially well suited for application of machine learning due to the vast amount of data that is generated. One devastating disease for which neuroimaging plays a significant role in the clinical management is stroke. Within this context, AI techniques can play pivotal roles for image-based diagnosis and management of stroke. This overview focuses on the recent advances of artificial intelligence methods - particularly supervised machine learning and deep learning - with respect to workflow, image acquisition and reconstruction, and image interpretation in patients with acute stroke, while also discussing potential pitfalls and future applications.

Association of initial imaging modality and futile recanalization after thrombectomy

Objective

To test the hypothesis that selection by initial imaging modality (MRI vs CT) is associated with rate of futile recanalizations (FRs) after mechanical thrombectomy (MT), we assessed this association in a multicenter, retrospective observational registry (BEYOND-SWIFT [Registry for Evaluating Outcome of Acute Ischemic Stroke Patients Treated With Mechanical Thrombectomy], NCT03496064).

Methods

In 2,011 patients (49.7% female, median age 73 years [61–81]) included between 2009 and 2017, we performed univariate and multivariate analyses regarding the occurrence of FR. FRs were defined as 90-day modified Rankin Scale (mRS) score 4–6 despite successful recanalization in patients selected by MRI (n = 690) and CT (n = 1,321) with a sensitivity analysis considering only patients with mRS 5–6 as futile.

Results

MRI as compared to CT resulted in similar rates of subsequent MT (adjusted odds ratio [aOR] 1.048, 95% confidence interval [CI] 0.677–1.624). Rates of FR were as follows: 571/1,489 (38%) FR mRS 4–6 including 393/1,489 (26%) FR mRS 5–6. CT-based selection was associated with increased rates of FRs compared to MRI (44% [41%–47%] vs 29% [25%–32%], p < 0.001; aOR 1.77 [95% CI 1.25–2.51]). These findings were robust in sensitivity analysis. MRI-selected patients had a delay of approximately 30 minutes in workflow metrics in real-world university comprehensive stroke centers. However, functional outcome and mortality were more favorable in patients selected by MRI compared to patients selected with CT.

Conclusions

CT selection for MT was associated with an increased risk of FRs as compared to MRI selection. Efforts are needed to shorten workflow delays in MRI patients. Further research is needed to clarify the role of the initial imaging modality on FR occurrence and to develop a reliable FR prediction algorithm.

Novel Estimation of Penumbra Zone Based on Infarct Growth Using Machine Learning Techniques in Acute Ischemic Stroke

While the penumbra zone is traditionally assessed based on perfusion-diffusion mismatch, it can be assessed based on machine learning (ML) prediction of infarct growth. The purpose of this work was to develop and validate an ML method for the prediction of infarct growth distribution and volume, in cases of successful (SR) and unsuccessful recanalization (UR). Pre-treatment perfusion-weighted, diffusion-weighted imaging (DWI) data, and final infarct lesions annotated from day-7 DWI from patients with middle cerebral artery occlusion were utilized to develop and validate two ML models for prediction of tissue fate. SR and UR models were developed from data in patients with modified treatment in cerebral infarction (mTICI) scores of 2b-3 and 0-2a, respectively. When compared to manual infarct annotation, ML-based infarct volume predictions resulted in an intraclass correlation coefficient (ICC) of 0.73 (95% CI = 0.31-0.91, p < 0.01) for UR, and an ICC of 0.87 (95% CI = 0.73-0.94, p < 0.001) for SR. Favorable outcomes for mismatch presence and absence in SR were 50% and 36%, respectively, while they were 61%, 56%, and 25%, respectively, for the low, intermediate, and high infarct growth groups. The presented method can offer novel and alternative insights into selecting patients for recanalization therapy and predicting functional outcome.

Use of Deep Learning to Predict Final Ischemic Stroke Lesions From Initial Magnetic Resonance Imaging

IMPORTANCE

Predicting infarct size and location is important for decision-making and prognosis in patients with acute stroke.

OBJECTIVES

To determine whether a deep learning model can predict final infarct lesions using magnetic resonance images (MRIs) acquired at initial presentation (baseline) and to compare the model with current clinical prediction methods.

DESIGN, SETTING, AND PARTICIPANTS

In this multicenter prognostic study, a specific type of neural network for image segmentation (U-net) was trained, validated, and tested using patients from the Imaging Collaterals in Acute Stroke (iCAS) study from April 14, 2014, to April 15, 2018, and the Diffusion Weighted Imaging Evaluation for Understanding Stroke Evolution Study-2 (DEFUSE-2) study from July 14, 2008, to September 17, 2011 (reported in October 2012). Patients underwent baseline perfusion-weighted and diffusion-weighted imaging and MRI at 3 to 7 days after baseline. Patients were grouped into unknown, minimal, partial, and major reperfusion status based on 24-hour imaging results. Baseline images acquired at presentation were inputs, and the final true infarct lesion at 3 to 7 days was considered the ground truth for the model. The model calculated the probability of infarction for every voxel, which can be thresholded to produce a prediction. Data were analyzed from July 1, 2018, to March 7, 2019.

MAIN OUTCOMES AND MEASURES

Area under the curve, Dice score coefficient (DSC) (a metric from 0-1 indicating the extent of overlap between the prediction and the ground truth; a DSC of ≥0.5 represents significant overlap), and volume error. Current clinical methods were compared with model performance in subgroups of patients with minimal or major reperfusion.

RESULTS

Among the 182 patients included in the model (97 women [53.3%]; mean [SD] age, 65 [16] years), the deep learning model achieved a median area under the curve of 0.92 (interquartile range [IQR], 0.87-0.96), DSC of 0.53 (IQR, 0.31-0.68), and volume error of 9 (IQR, -14 to 29) mL. In subgroups with minimal (DSC, 0.58 [IQR, 0.31-0.67] vs 0.55 [IQR, 0.40-0.65]; P = .37) or major (DSC, 0.48 [IQR, 0.29-0.65] vs 0.45 [IQR, 0.15-0.54]; P = .002) reperfusion for which comparison with existing clinical methods was possible, the deep learning model had comparable or better performance.

CONCLUSIONS AND RELEVANCE

The deep learning model appears to have successfully predicted infarct lesions from baseline imaging without reperfusion information and achieved comparable performance to existing clinical methods. Predicting the subacute infarct lesion may help clinicians prepare for decompression treatment and aid in patient selection for neuroprotective clinical trials.

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.

Machine Learning Principles for Radiology Investigators

Artificial intelligence and deep learning are areas of high interest for radiology investigators at present. However, the field of machine learning encompasses multiple statistics-based techniques useful for investigators, which may be complementary to deep learning approaches. After a refresher in basic statistical concepts, relevant considerations for machine learning practitioners are reviewed: regression, classification, decision boundaries, and bias-variance tradeoff. Regularization, ground truth, and populations are discussed along with compute and data management principles. Advanced statistical machine learning techniques including bootstrapping, bagging, boosting, decision trees, random forest, XGboost, and support vector machines are reviewed along with relevant examples from the radiology literature.

Technical considerations of multi-parametric tissue outcome prediction methods in acute ischemic stroke patients

Decisions regarding acute stroke treatment rely heavily on imaging, but interpretation can be difficult for physicians. Machine learning methods can assist clinicians by providing tissue outcome predictions for different treatment approaches based on acute multi-parametric imaging. To produce such clinically viable machine learning models, factors such as classifier choice, data normalization, and data balancing must be considered. This study gives comprehensive consideration to these factors by comparing the agreement of voxel-based tissue outcome predictions using acute imaging and clinical parameters with manual lesion segmentations derived from follow-up imaging. This study considers random decision forest, generalized linear model, and k-nearest-neighbor machine learning classifiers in conjunction with three data normalization approaches (non-normalized, relative to contralateral hemisphere, and relative to contralateral VOI), and two data balancing strategies (full dataset and stratified subsampling). These classifier settings were evaluated based on 90 MRI datasets from acute ischemic stroke patients. Distinction was made between patients recanalized using intraarterial and intravenous methods, as well as those without successful recanalization. For primary quantitative comparison, the Dice metric was computed for each voxel-based tissue outcome prediction and its corresponding follow-up lesion segmentation. It was found that the random forest classifier outperformed the generalized linear model and the k-nearest-neighbor classifier, that normalization did not improve the Dice score of the lesion outcome predictions, and that the models generated lesion outcome predictions with higher Dice scores when trained with balanced datasets. No significant difference was found between the treatment groups (intraarterial vs intravenous) regarding the Dice score of the tissue outcome predictions.

Comparison of classification methods for tissue outcome after ischaemic stroke

In acute ischaemic stroke, identifying brain tissue at high risk of infarction is important for clinical decision-making. This tissue may be identified with suitable classification methods from magnetic resonance imaging data. The aim of the present study was to assess and compare the performance of five popular classification methods (adaptive boosting, logistic regression, artificial neural networks, random forest and support vector machine) in identifying tissue at high risk of infarction on human voxel-based brain imaging data. The classification methods were used with eight MRI parameters, including diffusion-weighted imaging and perfusion-weighted imaging obtained in 55 patients. The five criteria used to assess the performance of the methods were the area under the receiver operating curve (AUC_roc ), the area under the precision-recall curve (AUC_pr ), sensitivity, specificity and the Dice coefficient. The methods performed equally in terms of sensitivity and specificity, while the results of AUC_roc and the Dice coefficient were significantly better for adaptive boosting, logistic regression, artificial neural networks and random forest. However, there was no statistically significant difference between the performances of these five classification methods regarding AUC_pr , which was the main comparison metric. Machine learning methods can provide valuable prognostic information using multimodal imaging data in acute ischaemic stroke, which in turn can assist in developing personalized treatment decision for clinicians after a thorough validation of methods with an independent data set.

Neural Network-derived Perfusion Maps for the Assessment of Lesions in Patients with Acute Ischemic Stroke

Purpose

To perform a proof-of-concept study to investigate the clinical utility of perfusion maps derived from convolutional neural networks (CNNs) for the workup of patients with acute ischemic stroke presenting with a large vessel occlusion.

Materials and Methods

Data on endovascularly treated patients with acute ischemic stroke (n = 151; median age, 68 years [interquartile range, 59-75 years]; 82 of 151 [54.3%] women) were retrospectively extracted from a single-center institutional prospective registry (between January 2011 and December 2015). Dynamic susceptibility perfusion imaging data were processed by applying a commercially available reference method and in parallel by a recently proposed CNN method to automatically infer time to maximum of the tissue residue function (Tmax) perfusion maps. The outputs were compared by using quantitative markers of tissue at risk derived from manual segmentations of perfusion lesions from two expert raters.

Results

Strong correlations of lesion volumes (Tmax > 4 seconds, > 6 seconds, and > 8 seconds; R = 0.865-0.914; P < .001) and good spatial overlap of respective lesion segmentations (Dice coefficients, 0.70-0.85) between the CNN method and reference output were observed. Eligibility for late-window reperfusion treatment was feasible with use of the CNN method, with complete interrater agreement for the CNN method (Cohen κ = 1; P < .001), although slight discrepancies compared with the reference-based output were observed (Cohen κ = 0.609-0.64; P < .001). The CNN method tended to underestimate smaller lesion volumes, leading to a disagreement between the CNN and reference method in five of 45 patients (9%).

Conclusion

Compared with standard deconvolution-based processing of raw perfusion data, automatic CNN-derived Tmax perfusion maps can be applied to patients who have acute ischemic large vessel occlusion stroke, with similar clinical utility.© RSNA, 2019Supplemental material is available for this article.

Automatic Neuroimage Processing and Analysis in Stroke-A Systematic Review

This article presents a systematic review of the current computational technologies applied to medical images for the detection, segmentation, and classification of strokes. Besides, analyzing and evaluating the technological advances, the challenges to be overcome and the future trends are discussed. The principal approaches make use of artificial intelligence, digital image processing and analysis, and various other technologies to develop computer-aided diagnosis (CAD) systems to improve the accuracy in the diagnostic process, as well as the interpretation consistency of medical images. However, there are some points that require greater attention such as low sensitivity, optimization of the algorithm, a reduction of false positives, and improvement in the identification and segmentation processes of different sizes and shapes. Also, there is a need to improve the classification steps of different stroke types and subtypes. Furthermore, there is an additional need for further research to improve the current techniques and develop new algorithms to overcome disadvantages identified here. The main focus of this research is to analyze the applied technologies for the development of CAD systems and verify how effective they are for stroke detection, segmentation, and classification. The main contributions of this review are that it analyzes only up-to-date studies, mainly from 2015 to 2018, as well as organizing the various studies in the area according to the research proposal, i.e., detection, segmentation, and classification of the types of stroke and the respective techniques used. Thus, the review has great relevance for future research, since it presents an ample comparison of the most recent works in the area, clearly showing the existing difficulties and the models that have been proposed to overcome such difficulties.

The Current Research Landscape of the Application of Artificial Intelligence in Managing Cerebrovascular and Heart Diseases: A Bibliometric and Content Analysis

The applications of artificial intelligence (AI) in aiding clinical decision-making and management of stroke and heart diseases have become increasingly common in recent years, thanks in part to technological advancements and the heightened interest of the research and medical community. This study aims to provide a comprehensive picture of global trends and developments of AI applications relating to stroke and heart diseases, identifying research gaps and suggesting future directions for research and policy-making. A novel analysis approach that combined bibliometrics analysis with a more complex analysis of abstract content using exploratory factor analysis and Latent Dirichlet allocation, which uncovered emerging research domains and topics, was adopted. Data were extracted from the Web of Science database. Results showed topics with the most compelling growth to be AI for big data analysis, robotic prosthesis, robotics-assisted stroke rehabilitation, and minimally invasive surgery. The study also found an emerging landscape of research that was centered on population-specific and early detection of stroke and heart disease. Application of AI in health behavior tracking and improvement as well as the use of robotics in medical diagnostics and prognostication have also been found to attract significant research attention. In light of these findings, it is suggested that the currently under-researched issues of data management, AI model reliability, as well as validation of its clinical utility, need to be further explored in future research and policy decisions to maximize the benefits of AI applications in stroke and heart diseases.

Characterization of clot composition in acute cerebral infarct using machine learning techniques

Objective

Clot characteristics can provide information on the cause of cerebral artery occlusion and may guide acute revascularization and secondary prevention strategies. We developed a rapid automated clot analysis system using machine learning (ML) and validated its accuracy in patients undergoing endovascular treatment.

Methods

Pre‐endovascular treatment gradient echo (GRE) images from consecutive patients with middle cerebral artery occlusion were utilized to develop and validate an ML system to predict whether atrial fibrillation (AF) was the underlying cause of ischemic stroke. The accuracy of the ML algorithm was compared with that of visual inspection by neuroimaging specialists for the presence of blooming artifact. Endovascular procedures and outcomes were compared in patients with and without AF.

Results

Of 67 patients, 29 (43.3%) had AF. Of these, 13 had known AF and 16 were newly diagnosed with cardiac monitoring. By visual inspection, interrater correlation for blooming artifact was 0.73 and sensitivity and specificity for AF were 0.79 and 0.63, respectively. For AF classification, the ML algorithms yielded an average accuracy of > 75.4% in fivefold cross‐validation with clot signal profiles obtained from 52 patients and an area under the curve >0.87 for the average AF probability from five signal profiles in external validation (n = 15). Analysis with an in‐house interface took approximately 3 min per patient. Absence of AF was associated with increased number of passes by stentriever, high reocclusion frequency, and additional use of rescue stenting and/or glycogen IIb/IIIa blocker for recanalization.

Interpretation

ML‐based rapid clot analysis is feasible and can identify AF with high accuracy, enabling selection of endovascular treatment strategy.

Stroke Lesion Outcome Prediction Based on MRI Imaging Combined With Clinical Information

In developed countries, the second leading cause of death is stroke, which has the ischemic stroke as the most common type. The preferred diagnosis procedure involves the acquisition of multi-modal Magnetic Resonance Imaging. Besides detecting and locating the stroke lesion, Magnetic Resonance Imaging captures blood flow dynamics that guides the physician in evaluating the risks and benefits of the reperfusion procedure. However, the decision process is an intricate task due to the variability of lesion size, shape, and location, as well as the complexity of the underlying cerebral hemodynamic process. Therefore, an automatic method that predicts the stroke lesion outcome, at a 3-month follow-up, would provide an important support to the physicians' decision process. In this work, we propose an automatic deep learning-based method for stroke lesion outcome prediction. Our main contribution resides in the combination of multi-modal Magnetic Resonance Imaging maps with non-imaging clinical meta-data: the thrombolysis in cerebral infarction scale, which categorizes the success of recanalization, achieved through mechanical thrombectomy. In our proposal, this clinical information is considered at two levels. First, at a population level by embedding the clinical information in a custom loss function used during training of our deep learning architecture. Second, at a patient-level through an extra input channel of the neural network used at testing time for a given patient case. By merging imaging with non-imaging clinical information, we aim to obtain a model aware of the principal and collateral blood flow dynamics for cases where there is no perfusion beyond the point of occlusion and for cases where the perfusion is complete after the occlusion point.

ISLES 2016 and 2017-Benchmarking Ischemic Stroke Lesion Outcome Prediction Based on Multispectral MRI

Performance of models highly depend not only on the used algorithm but also the data set it was applied to. This makes the comparison of newly developed tools to previously published approaches difficult. Either researchers need to implement others' algorithms first, to establish an adequate benchmark on their data, or a direct comparison of new and old techniques is infeasible. The Ischemic Stroke Lesion Segmentation (ISLES) challenge, which has ran now consecutively for 3 years, aims to address this problem of comparability. ISLES 2016 and 2017 focused on lesion outcome prediction after ischemic stroke: By providing a uniformly pre-processed data set, researchers from all over the world could apply their algorithm directly. A total of nine teams participated in ISLES 2015, and 15 teams participated in ISLES 2016. Their performance was evaluated in a fair and transparent way to identify the state-of-the-art among all submissions. Top ranked teams almost always employed deep learning tools, which were predominately convolutional neural networks (CNNs). Despite the great efforts, lesion outcome prediction persists challenging. The annotated data set remains publicly available and new approaches can be compared directly via the online evaluation system, serving as a continuing benchmark (www.isles-challenge.org).

Relating Acute Lesion Loads to Chronic Outcome in Ischemic Stroke-An Exploratory Comparison of Mismatch Patterns and Predictive Modeling

Objectives: To investigate the relationship between imaging features derived from lesion loads and 3 month clinical assessments in ischemic stroke patients. To support clinically implementable predictive modeling with information from lesion-load features. Methods: A retrospective cohort of ischemic stroke patients was studied. The dataset was dichotomized based on revascularization treatment outcome (TICI score). Three lesion delineations were derived from magnetic resonance imaging in each group: two clinically implementable (threshold based and fully automatic prediction) and 90-day follow-up as final groundtruth. Lesion load imaging features were created through overlay of the lesion delineations on a histological brain atlas, and were correlated with the clinical assessment (NIHSS). Significance of the correlations was assessed by constructing confidence intervals using bootstrap sampling. Results: Overall, high correlations between lesion loads and clinical score were observed (up to 0.859). Delineations derived from acute imaging yielded on average somewhat lower correlations than delineations derived from 90-day follow-up imaging. Correlations suggest that both total lesion volume and corticospinal tract lesion load are associated with functional outcome, and in addition highlight other potential areas associated with poor clinical outcome, including the primary somatosensory cortex BA3a. Fully automatic prediction was comparable to ADC threshold-based delineation on the successfully treated cohort and superior to the Tmax threshold-based delineation in the unsuccessfully treated cohort. Conclusions: The confirmation of established predictors for stroke outcome (e.g., corticospinal tract integrity and total lesion volume) gives support to the proposed methodology-relating acute lesion loads to 3 month outcome assessments by way of correlation. Furthermore, the preliminary results indicate an association of further brain regions and structures with three month NIHSS outcome assessments. Hence, prediction models might observe an increased accuracy when incorporating regional (instead of global) lesion loads. Also, the results lend support to the clinical utilization of the automatically predicted volumes from FASTER, rather than the simpler DWI and PWI lesion delineations.