Accelerating Prediction of Malignant Cerebral Edema After Ischemic Stroke with Automated Image Analysis and Explainable Neural Networks

Predictive power of artificial intelligence for malignant cerebral edema in stroke patients: a CT-based systematic review and meta-analysis of prevalence and diagnostic performance

Malignant cerebral edema (MCE) is a severe complication of acute ischemic stroke, with high mortality rates. Early and accurate prediction of MCE is critical for initiating timely interventions such as decompressive hemicraniectomy. Artificial intelligence (AI) and radiomics have emerged as promising tools for predicting MCE, offering the potential to transform reactive stroke management into proactive care. However, variability in methodologies and inconsistent reporting limits the widespread adoption of these technologies. A comprehensive search of PubMed, Embase, Web of Science, and Scopus identified studies reporting on the sensitivity, specificity, and area under the curve (AUC) of AI models in MCE prediction. Data were synthesized using random-effects meta-analyses. Subgroup analyses explored the impact of study design, machine learning input type, and other key factors on diagnostic accuracy. Publication bias was assessed using Egger's test and funnel plot analyses. Data from ten studies encompassing 1,594 unique stroke patients were included in the analysis. The pooled sensitivity and specificity of AI models for predicting MCE were 81.1% (95% CI: 73.0-87.2%) and 92.6% (95% CI: 91.2-93.9%), respectively, with an AUC of 0.939. The diagnostic odds ratio was 43.73 (95% CI: 24.78-77.15), demonstrating excellent discriminative ability. Subgroup analyses revealed higher sensitivity and specificity in prospective studies (92.0% and 93.3%) compared to retrospective studies (76.1% and 91.4%). Radiomics-based models showed slightly higher sensitivity (84.2%) compared to non-radiomics models (80.4%), though both input types achieved comparable specificity. Interestingly, patients undergoing revascularization had a higher prevalence of MCE, likely due to their more severe initial presentations. Minimal heterogeneity was observed in specificity across studies, while publication bias was noted for sensitivity estimates. AI models show excellent diagnostic performance for predicting malignant cerebral edema (MCE), offering high sensitivity and specificity. Prospective studies, radiomics integration, and multi-center collaborations enhance their accuracy. However, external validation and standardized methodologies are needed to ensure broader clinical adoption and improve outcomes for stroke patients at risk of MCE. Clinical trial number Not applicable.

Clinically relevant findings on 24-h head CT after acute stroke therapy: The 24-h CT score

BACKGROUND

Routine head computed tomography (CT) is performed 24 h post-acute stroke thrombolysis and thrombectomy, even in patients with stable or improving clinical deficits. Predicting CT results that impact management could help prioritize patients at risk and potentially reduce unnecessary imaging.

METHODS

In this institutional review board (IRB)-approved retrospective study, data from 1461 consecutive acute ischemic stroke patients at our Comprehensive Stroke Center (n = 8943, 2012-2022) who received intravenous thrombolysis or endovascular therapy, exhibited stable or improving 24-h exams, and underwent 24-h follow-up head CT per standard acute stroke care guidelines. CT reports 24 h post-stroke were reviewed for edema, mass effect, herniation, and hemorrhage. The primary outcome was any clinically relevant 24-h CT finding that led to changes in antithrombotic treatment or blood pressure goals, extended intensive care unit (ICU) stays or hospitalizations, neurosurgical interventions, or administration of mannitol or hypertonic saline. Multivariable logistic regression identified independent predictors of clinically meaningful CT abnormalities. A 24-h CT score was developed and cross-validated.

RESULTS

The mean age was 70 years, with 47% women. The median National Institutes of Health Stroke Scale (NIHSS) score at admission was 12 (interquartile range (IQR): 6-18). Stroke-related abnormalities on 24-h CT were present in 325 patients (22.2%), with 183 (12.5%) showing clinically relevant findings. Age, admission NIHSS, and blood glucose levels were independent predictors of clinically relevant 24-h CT findings. The final model C statistic was 0.72 (95% confidence interval (CI): 0.68-0.76) in the derivation cohort and 0.72 (95% CI: 0.67-0.75) in bootstrapping validation. The 24-h CT score was developed using these predictors: NIHSS score 5-15 (+3); NIHSS score ⩾16 (+5); age < 75 years (+1); admission glucose ⩾ 140 mg/dL (+1). The prevalence of clinically relevant CT findings was 4.3% in the low-risk group (24-h CT score ⩽ 4), 11.3% in the medium-risk group (score 5), and 21.4% in the high-risk group (score ⩾ 6). The 24-h CT score demonstrated good calibration.

CONCLUSION

In patients undergoing thrombolysis or thrombectomy who undergo routine 24-h head CT despite remaining clinically stable or improving, only one in eight prove to have 24-h head CT findings that impact management. The 24-h CT score provides risk stratification that may improve resource utilization.

DATA ACCESS STATEMENT

A.S. and B.Z. have full access to the data used in the analysis in this article. Deidentified data will be shared after ethics approval if requested by other investigators for purposes of replicating the results.

Predictive Value of Machine Learning Models for Cerebral Edema Risk in Stroke Patients: A Meta-Analysis

INTRODUCTION

Stroke patients are at high risk of developing cerebral edema, which can have severe consequences. However, there are currently few effective tools for early identification or prediction of this risk. As machine learning (ML) is increasingly used in clinical practice, its effectiveness in predicting cerebral edema risk in stroke patients has been explored. Nonetheless, the lack of systematic evidence on its predictive value challenges the update of simple and user-friendly risk assessment tools. Therefore, we conducted a systematic review to evaluate the predictive utility of ML for cerebral edema in stroke patients.

METHODS

We searched PubMed, Embase, Web of Science, and the Cochrane Database up to February 21, 2024. The risk of bias in selected studies was assessed using a bias assessment tool for predictive models. Meta-analysis synthesized results from validation sets.

RESULTS

We included 22 studies with 25,096 stroke patients and 25 models, which were constructed using common and interpretable clinical features. In the validation cohort, the models achieved a concordance index (c-index) of 0.840 (95% CI: 0.810-0.871) for predicting poststroke cerebral edema, with a sensitivity of 0.76 (95% CI: 0.72-0.79) and a specificity of 0.87 (95% CI: 0.83-0.90).

CONCLUSION

ML models are significant in predicting poststroke cerebral edema, providing clinicians with a powerful prognostic tool. However, radiomics-based research was not included. We anticipate advancements in radiomics research to enhance the predictive power of ML for poststroke cerebral edema.

CSF-Based Volumetric Imaging Biomarkers Highlight Incidence and Risk Factors for Cerebral Edema After Ischemic Stroke

BACKGROUND

Cerebral edema has primarily been studied using midline shift or clinical deterioration as end points, which only captures the severe and delayed manifestations of a process affecting many patients with stroke. Quantitative imaging biomarkers that measure edema severity across the entire spectrum could improve its early detection, as well as identify relevant mediators of this important stroke complication.

METHODS

We applied an automated image analysis pipeline to measure the displacement of cerebrospinal fluid (ΔCSF) and the ratio of lesional versus contralateral hemispheric cerebrospinal fluid (CSF) volume (CSF ratio) in a cohort of 935 patients with hemispheric stroke with follow-up computed tomography scans taken a median of 26 h (interquartile range 24-31) after stroke onset. We determined diagnostic thresholds based on comparison to those without any visible edema. We modeled baseline clinical and radiographic variables against each edema biomarker and assessed how each biomarker was associated with stroke outcome (modified Rankin Scale at 90 days).

RESULTS

The displacement of CSF and CSF ratio were correlated with midline shift (r = 0.52 and - 0.74, p < 0.0001) but exhibited broader ranges. A ΔCSF of greater than 14% or a CSF ratio below 0.90 identified those with visible edema: more than half of the patients with stroke met these criteria, compared with only 14% who had midline shift at 24 h. Predictors of edema across all biomarkers included a higher National Institutes of Health Stroke Scale score, a lower Alberta Stroke Program Early CT score, and lower baseline CSF volume. A history of hypertension and diabetes (but not acute hyperglycemia) predicted greater ΔCSF but not midline shift. Both ΔCSF and a lower CSF ratio were associated with worse outcome, adjusting for age, National Institutes of Health Stroke Scale score, and Alberta Stroke Program Early CT score (odds ratio 1.7, 95% confidence interval 1.3-2.2 per 21% ΔCSF).

CONCLUSIONS

Cerebral edema can be measured in a majority of patients with stroke on follow-up computed tomography using volumetric biomarkers evaluating CSF shifts, including in many without visible midline shift. Edema formation is influenced by clinical and radiographic stroke severity but also by chronic vascular risk factors and contributes to worse stroke outcomes.

Severely Hypoperfused Brain Tissue Correlates with Final Infarct Volume Despite Recanalization in DMVO Stroke

Objectives

We sought to assess whether there were any parameter(s) on baseline computed-tomography-perfusion (CTP) strongly correlating with final-infarct-volume, and infarct volume progression after endovascular recanalization of acute ischemic stroke (AIS) with primary distal, medium vessel occlusion (DMVO).

Materials and Methods

We performed a retrospective analysis of consecutive AIS patients who were successfully recanalized by thrombectomy for DMVO. By comparing baseline CTP and follow-up MRI, we evaluated the correlation between baseline infarct and hypoperfusion volumes, and final infarct volume and infarct volume progression. We also examined their effect on good clinical outcome at 3 months (defined as an mRS score of 0 to 2).

Results

Between January 2018 and January 2021, 38 patients met the inclusion criteria (76% [29/38] female, median age 75 [66-86] years). Median final infarct volume and infarct volume progression were 8.4 mL [IQR: 5.2-44.4] and 7.2 mL [IQR: 4.3-29.1] respectively. TMax > 10 sec volume was strongly correlated with both (r = 0.831 and r = 0.771 respectively, p < 0.0001), as well as with good clinical outcome (-0.5, p = 0.001). A higher baseline TMax > 10 sec volume increased the probability of a higher final-infarct-volume (r2 = 0.690, coefficient = 0.83 [0.64-1.00], p < 0.0001), whereas it decreased the probability of good clinical outcome at 3 months (odds ratio = -0.67 [-1.17 to -0.18], p = 0.008).

Conclusion

TMax > 10 sec volume on baseline CTP correlates strongly with final infarct volume as well as with clinical outcome after mechanical thrombectomy for an AIS with DMVO.

Imaging biomarkers of cerebral edema automatically extracted from routine CT scans of large vessel occlusion strokes

BACKGROUND AND PURPOSE

Volumetric and densitometric biomarkers have been proposed to better quantify cerebral edema after stroke, but their relative performance has not been rigorously evaluated.

METHODS

Patients with large vessel occlusion stroke from three institutions were analyzed. An automated pipeline extracted brain, cerebrospinal fluid (CSF), and infarct volumes from serial CTs. Several biomarkers were measured: change in global CSF volume from baseline (ΔCSF); ratio of CSF volumes between hemispheres (CSF ratio); and relative density of infarct region compared with mirrored contralateral region (net water uptake [NWU]). These were compared to radiographic standards, midline shift and relative hemispheric volume (RHV) and malignant edema, defined as deterioration resulting in need for osmotic therapy, decompressive surgery, or death.

RESULTS

We analyzed 255 patients with 210 baseline CTs, 255 24-hour CTs, and 81 72-hour CTs. Of these, 35 (14%) developed malignant edema and 63 (27%) midline shift. CSF metrics could be calculated for 310 (92%), while NWU could only be obtained from 193 (57%). Peak midline shift was correlated with baseline CSF ratio (ρ = -.22) and with CSF ratio and ΔCSF at 24 hours (ρ = -.55/.63) and 72 hours (ρ = -.66/.69), but not with NWU (ρ = .15/.25). Similarly, CSF ratio was correlated with RHV (ρ = -.69/-.78), while NWU was not. Adjusting for age, National Institutes of Health Stroke Scale, tissue plasminogen activator treatment, and Alberta Stroke Program Early CT Score, CSF ratio (odds ratio [OR]: 1.95 per 0.1, 95% confidence interval [CI]: 1.52-2.59) and ΔCSF at 24 hours (OR: 1.87 per 10%, 95% CI: 1.47-2.49) were associated with malignant edema.

CONCLUSION

CSF volumetric biomarkers can be automatically measured from almost all routine CTs and correlate better with standard edema endpoints than net water uptake.

Application of explainable artificial intelligence for healthcare: A systematic review of the last decade (2011-2022)

BACKGROUND AND OBJECTIVES

Artificial intelligence (AI) has branched out to various applications in healthcare, such as health services management, predictive medicine, clinical decision-making, and patient data and diagnostics. Although AI models have achieved human-like performance, their use is still limited because they are seen as a black box. This lack of trust remains the main reason for their low use in practice, especially in healthcare. Hence, explainable artificial intelligence (XAI) has been introduced as a technique that can provide confidence in the model's prediction by explaining how the prediction is derived, thereby encouraging the use of AI systems in healthcare. The primary goal of this review is to provide areas of healthcare that require more attention from the XAI research community.

METHODS

Multiple journal databases were thoroughly searched using PRISMA guidelines 2020. Studies that do not appear in Q1 journals, which are highly credible, were excluded.

RESULTS

In this review, we surveyed 99 Q1 articles covering the following XAI techniques: SHAP, LIME, GradCAM, LRP, Fuzzy classifier, EBM, CBR, rule-based systems, and others.

CONCLUSION

We discovered that detecting abnormalities in 1D biosignals and identifying key text in clinical notes are areas that require more attention from the XAI research community. We hope this is review will encourage the development of a holistic cloud system for a smart city.

Natural Language Processing of Radiology Reports to Detect Complications of Ischemic Stroke

BACKGROUND

Abstraction of critical data from unstructured radiologic reports using natural language processing (NLP) is a powerful tool to automate the detection of important clinical features and enhance research efforts. We present a set of NLP approaches to identify critical findings in patients with acute ischemic stroke from radiology reports of computed tomography (CT) and magnetic resonance imaging (MRI).

METHODS

We trained machine learning classifiers to identify categorical outcomes of edema, midline shift (MLS), hemorrhagic transformation, and parenchymal hematoma, as well as rule-based systems (RBS) to identify intraventricular hemorrhage (IVH) and continuous MLS measurements within CT/MRI reports. Using a derivation cohort of 2289 reports from 550 individuals with acute middle cerebral artery territory ischemic strokes, we externally validated our models on reports from a separate institution as well as from patients with ischemic strokes in any vascular territory.

RESULTS

In all data sets, a deep neural network with pretrained biomedical word embeddings (BioClinicalBERT) achieved the highest discrimination performance for binary prediction of edema (area under precision recall curve [AUPRC] > 0.94), MLS (AUPRC > 0.98), hemorrhagic conversion (AUPRC > 0.89), and parenchymal hematoma (AUPRC > 0.76). BioClinicalBERT outperformed lasso regression (p < 0.001) for all outcomes except parenchymal hematoma (p = 0.755). Tailored RBS for IVH and continuous MLS outperformed BioClinicalBERT (p < 0.001) and linear regression, respectively (p < 0.001).

CONCLUSIONS

Our study demonstrates robust performance and external validity of a core NLP tool kit for identifying both categorical and continuous outcomes of ischemic stroke from unstructured radiographic text data. Medically tailored NLP methods have multiple important big data applications, including scalable electronic phenotyping, augmentation of clinical risk prediction models, and facilitation of automatic alert systems in the hospital setting.

Evaluation of Mannitol Intervention Effects on Ischemic Cerebral Edema in Mice Using Swept Source Optical Coherence Tomography

Cerebral edema is a serious complication of ischemic cerebrovascular disease and mannitol is a commonly used dehydrating agent for relieving cerebral edema. However, the edema state and surrounding vascular perfusion level during mannitol treatment remains unclear, which affects the clinical application of the medicine. In this paper, we demonstrated the role of swept-source optical coherence tomography (OCT) in the evaluation of mannitol efficacy using mouse models. The OCT-based angiography and attenuation imaging technology were used to obtain the cerebral vascular perfusion level and cerebral edema state at different times. Vascular parameters and edema parameters were quantified and compared. Experimental results show that mannitol can significantly reduce the water content in the central region of edema, effectively inhibiting the rapid growth of the edema area, and restoring cerebral blood flow. On average, the edema area decreased by 33% after 2 h, and the vascular perfusion density increased by 12% after 5 h. This work helps to provide a valuable theoretical basis and research ideas for the clinical treatment of cerebral edema.

Evaluation and Prediction of Post-stroke Cerebral Edema Based on Neuroimaging

Cerebral edema is a common complication of acute ischemic stroke that leads to poorer functional outcomes and substantially increases the mortality rate. Given that its negative effects can be reduced by more intensive monitoring and evidence-based interventions, the early identification of patients with a high risk of severe edema is crucial. Neuroimaging is essential for the assessment and prediction of edema. Simple markers, such as midline shift and hypodensity volume on computed tomography, have been used to evaluate edema in clinical trials; however, advanced techniques can be applied to examine the underlying mechanisms. In this study, we aimed to review current imaging tools in the assessment and prediction of cerebral edema to provide guidance for using these methods in clinical practice.