Machine learning predicts the risk of hemorrhagic transformation of acute cerebral infarction and in-hospital death

Risk of bias assessment of post-stroke mortality machine learning predictive models: Systematic review

BACKGROUND

Stroke is a major cause of mortality and permanent disability worldwide. Precise prediction of post-stroke mortality is essential for guiding treatment decisions and rehabilitation planning. The ability of Machine learning models to process large amounts of data, offer a promising alternative for improving mortality prediction in stroke patients. In this review, we aim to evaluate the risk of bias in different machine learning models used for predicting post-stroke mortality.

METHODS

This review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS). Relevant articles were retrieved from Cochrane Library, Scopus, PubMed, and Web of Science databases.

RESULTS

A total of 9 studies were included, with an aggregate patient population of 669,424. Six studies used publicly available datasets, and four used hospital data with a follow up duration ranging from 7 days to 18 months. The range of area under the curve (AUC) for mortality prediction across the studies ranged from 0.81 to 0.95. All studies were determined to have a high overall risk of bias.

CONCLUSION

Machine learning models demonstrated great potential in predicting post-stroke mortality. However, implementation of these models in clinical practice is limited by high risk of bias. Future studies should focus on reducing this bias and enhancing the applicability of these models to improve the reliability of stroke mortality predictions.

Early Predictive Accuracy of Machine Learning for Hemorrhagic Transformation in Acute Ischemic Stroke: Systematic Review and Meta-Analysis

BACKGROUND

Hemorrhagic transformation (HT) is commonly detected in acute ischemic stroke (AIS) and often leads to poor outcomes. Currently, there is no ideal tool for early prediction of HT risk. Recently, machine learning has gained traction in stroke management, prompting the exploration of predictive models for HT. However, systematic evidence on these models is lacking.

OBJECTIVE

In this study, we assessed the predictive capability of machine learning models for HT risk in AIS, aiming to inform the development of HT prediction tools.

METHODS

We conducted a thorough search of medical databases, such as Web of Science, Embase, Cochrane, and PubMed up until March 2025. The risk of bias was determined through the Prediction Model Risk of Bias Assessment Tool (PROBAST). Subgroup analysis was performed based on treatment backgrounds, diagnostic criteria, and types of HT.

RESULTS

A total of 83 eligible articles were included, containing 106 models and 88,197 patients with AIS with 9323 HT cases. There were 104 validation sets with a total c-index of 0.832 (95% CI 0.814-0.849), sensitivity of 0.82 (95% CI 0.79-0.84), and specificity of 0.78 (95% CI 0.74-0.81). Subgroup analysis indicated that the combined model achieved superior prediction accuracy. Moreover, we also analyzed the predictive performance of 6 mature models.

CONCLUSIONS

Currently, although several prediction methods for HT have been developed, their predictive values are not satisfactory. Fortunately, our findings suggest that machine learning methods, particularly those combining clinical features and radiomics, hold promise for improving predictive accuracy. Our meta-analysis may provide evidence-based guidance for the subsequent development of more efficient clinical predictive models for HT.

TRIAL REGISTRATION

PROSPERO CRD42024498997; https://www.crd.york.ac.uk/PROSPERO/view/CRD42024498997.

Enhancing prediction of parenchymal hemorrhage type 2 after endovascular treatment in acute ischemic stroke using dual-phase CTA

OBJECTIVE

To evaluate the effectiveness of dual-phase CT angiography (CTA) in predicting parenchymal hemorrhage type 2 (PH2) following endovascular thrombectomy (EVT) in patients with acute ischemic stroke (AIS).

METHODS

A retrospective analysis was conducted across two centers, including 232 AIS patients who underwent EVT. Three predictive models were developed: a clinical model (Model C), a clinical model incorporating single-phase CTA data (Model CS), and a clinical model incorporating dual-phase CTA data (Model CD). The performance of these models in predicting PH2 occurrence post-EVT was assessed and compared.

RESULTS

The model incorporating dual-phase CTA data (Model CD) demonstrated superior predictive performance, with higher area under the curve (AUC) values in both training and validation datasets, compared to Models C and CS. Calibration and decision curve analyses further confirmed the enhanced accuracy and clinical utility of Model CD.

CONCLUSION

The findings indicate that dual-phase CTA provides a more accurate assessment of collateral circulation compared to single-phase CTA, thereby improving the prediction of PH2 after EVT. This enhanced predictive capability can assist clinicians in making more informed decisions regarding AIS management.

Machine learning based finite element analysis for personalized prediction of pressure injury risk in patients with spinal cord injury

BACKGROUND AND OBJECTIVE

Patients with spinal cord injury (SCI), are prone to pressure injury (PI) in the soft tissues of buttocks. Early prediction of PI holds the potential to reduce the occurrence and progression of PI. This study proposes a machine learning model to predict soft tissue stress/strain and evaluate PI risk in SCI patients.

METHODS

Based on the standard database from parametric models of buttock, the biomechanical response of soft tissues and risk factors affecting PI were analyzed. A comprehensive assessment of multiple machine-learning methods was performed to predict the risk of PI, the selected optimal model is explained locally and globally using Shapley additive explanations (SHAP).

RESULTS

The proposed hybrid model for predicting PI consists of a backpropagation neural network and Extreme Gradient Boosting, performed the coefficient of determination (R2) of 0.977.

CONCLUSION

The model exhibits accurate performance which may be considered as the ideal method for predicting PI. Furthermore, it can be used with other health-monitoring equipment to improve the quality of patients with SCI or other dysfunctional diseases.

Predicting hemorrhagic transformation in acute ischemic stroke: a systematic review, meta-analysis, and methodological quality assessment of CT/MRI-based deep learning and radiomics models

Acute ischemic stroke (AIS) is a major cause of mortality and morbidity, with hemorrhagic transformation (HT) as a severe complication. Accurate prediction of HT is essential for optimizing treatment strategies. This review assesses the accuracy and utility of deep learning (DL) and radiomics in predicting HT through imaging, regarding clinical decision-making for AIS patients. A literature search was conducted across five databases (Pubmed, Scopus, Web of Science, Embase, IEEE) up to January 23, 2025. Studies involving DL or radiomics-based ML models for predicting HT in AIS patients were included. Data from training, validation, and clinical-combined models were extracted and analyzed separately. Pooled sensitivity, specificity, and AUC were calculated with a random-effects bivariate model. For the quality assessment of studies, the Methodological Radiomics Score (METRICS) and QUADAS-2 tool were used. 16 studies consisting of 3,083 individual participants were included in the meta-analysis. The pooled AUC for training cohorts was 0.87, sensitivity 0.80, and specificity 0.85. For validation cohorts, AUC was 0.87, sensitivity 0.81, and specificity 0.86. Clinical-combined models showed an AUC of 0.93, sensitivity 0.84, and specificity 0.89. Moderate to severe heterogeneity was noted and addressed. Deep-learning models outperformed radiomics models, while clinical-combined models outperformed deep learning-only and radiomics-only models. The average METRICS score was 62.85%. No publication bias was detected. DL and radiomics models showed great potential in predicting HT in AIS patients. However, addressing methodological issues-such as inconsistent reference standards and limited external validation-is essential for the clinical implementation of these models.

Traditional and machine learning models for predicting haemorrhagic transformation in ischaemic stroke: a systematic review and meta-analysis

BACKGROUND

Haemorrhagic transformation (HT) is a severe complication after ischaemic stroke, but identifying patients at high risks remains challenging. Although numerous prediction models have been developed for HT following thrombolysis, thrombectomy, or spontaneous occurrence, a comprehensive summary is lacking. This study aimed to review and compare traditional and machine learning-based HT prediction models, focusing on their development, validation, and diagnostic accuracy.

METHODS

PubMed and Ovid-Embase were searched for observational studies or randomised controlled trials related to traditional or machine learning-based models. Data were extracted according to Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) checklist and risk of bias was assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST). Performance data for prediction models that were externally validated at least twice and showed low risk of bias were meta-analysed.

RESULTS

A total of 100 studies were included, with 67 focusing on model development and 33 on model validation. Among 67 model development studies, 44 were traditional model studies involving 47 prediction models (with National Institutes of Health Stroke Scale score being the most frequently used predictor in 35 models), and 23 studies focused on machine learning prediction models (with support vector machines being the most common algorithm, used in 10 models). The 33 validation studies externally validated 34 traditional prediction models. Regarding study quality, 26 studies were assessed as having a low risk of bias, 11 as unclear, and 63 as high risk of bias. Meta-analysis of 15 studies validating eight models showed a pooled area under the receiver operating characteristic curve of approximately 0.70 for predicting HT.

CONCLUSION

While significant progress has been made in developing HT prediction models, both traditional and machine learning-based models still have limitations in methodological rigour, predictive accuracy, and clinical applicability. Future models should undergo more rigorous validation, adhere to standardised reporting frameworks, and prioritise predictors that are both statistically significant and clinically meaningful. Collaborative efforts across research groups are essential for validating these models in diverse populations and improving their broader applicability in clinical practice.

SYSTEMATIC REVIEW REGISTRATION

International Prospective Register of Systematic Reviews (CRD42022332816).

Predicting functional outcomes of patients with spontaneous intracerebral hemorrhage based on explainable machine learning models: a multicenter retrospective study

Background

Spontaneous intracerebral hemorrhage (SICH) is the second most common cause of cerebrovascular disease after ischemic stroke, with high mortality and disability rates, imposing a significant economic burden on families and society. This retrospective study aimed to develop and evaluate an interpretable machine learning model to predict functional outcomes 3 months after SICH.

Methods

A retrospective analysis was conducted on clinical data from 380 patients with SICH who were hospitalized at three different centers between June 2020 and June 2023. Seventy percent of the samples were randomly selected as the training set, while the remaining 30% were used as the validation set. Univariate analysis, Least Absolute Shrinkage and Selection Operator (LASSO) regression, and Pearson correlation analysis were used to screen clinical variables. The selected variables were then incorporated into five machine learning models: complementary naive bayes (CNB), support vector machine (SVM), gaussian naive bayes (GNB), multilayer perceptron (MLP), and extreme gradient boosting (XGB), to assess their performance. Additionally, the area under the curve (AUC) values were evaluated to compare the performance of each algorithmic model, and global and individual interpretive analyses were conducted using importance ranking and Shapley additive explanations (SHAP).

Results

Among the 380 patients, 95 ultimately had poor prognostic outcomes. In the validation set, the AUC values for CNB, SVM, GNB, MLP, and XGB models were 0.899 (0.816-0.979), 0.916 (0.847-0.982), 0.730 (0.602-0.857), 0.913 (0.834-0.986), and 0.969 (0.937-0.998), respectively. Therefore, the XGB model performed the best among the five algorithms. SHAP analysis revealed that the GCS score, hematoma volume, blood pressure changes, platelets, age, bleeding location, and blood glucose levels were the most important variables for poor prognosis.

Conclusion

The XGB model developed in this study can effectively predict the risk of poor prognosis in patients with SICH, helping clinicians make personalized and rational clinical decisions. Prognostic risk in patients with SICH is closely associated with GCS score, hematoma volume, blood pressure changes, platelets, age, bleeding location, and blood glucose levels.

Coronary health index based on immunoglobulin light chains to assess coronary heart disease risk with machine learning: a diagnostic trial

BACKGROUND

Recent studies suggest a connection between immunoglobulin light chains (IgLCs) and coronary heart disease (CHD). However, current diagnostic methods using peripheral blood IgLCs levels or subtype ratios show limited accuracy for CHD, lacking comprehensive assessment and posing challenges in early detection and precise disease severity evaluation. We aim to develop and validate a Coronary Health Index (CHI) incorporating total IgLCs levels and their distribution. Additionally, we aim to evaluate its effectiveness by integrating patient data and using machine learning models through diagnostic trial.

METHODS

The CHI was developed and combined with other clinical data. Nine machine learning models were screened to identify optimal diagnostic performance, with the XGBoost model emerging as the top performer. Performance was assessed based on accuracy, sensitivity, and its ability to identify severe CHD cases characterized by complex lesions (SYNTAX score > 33).

RESULTS

The XGBoost model demonstrated high accuracy and sensitivity in diagnosing CHD, with an area under the curve (AUC) of 0.927. It also accurately identified patients with severe CHD, achieving an AUC of 0.991. An online web tool was introduced for broader external validation, confirming the model's effectiveness.

CONCLUSIONS

Combining the CHI with the XGBoost model offers significant advantages in diagnosing CHD and assessing disease severity. This approach can guide clinical interventions and improve large-scale CHD screening.

Machine learning and deep learning algorithms in stroke medicine: a systematic review of hemorrhagic transformation prediction models

BACKGROUND

Acute ischemic stroke (AIS) is a major cause of morbidity and mortality, with hemorrhagic transformation (HT) further worsening outcomes. Traditional scoring systems have limited predictive accuracy for HT in AIS. Recent research has explored machine learning (ML) and deep learning (DL) algorithms for stroke management. This study evaluates and compares the effectiveness of ML and DL algorithms in predicting HT post-AIS, benchmarking them against conventional models.

METHODS

A systematic search was conducted across PubMed, Embase, Web of Science, Scopus, and IEEE, initially yielding 1421 studies. After screening, 24 studies met the inclusion criteria. The Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to assess the quality of these studies, and a qualitative synthesis was performed due to heterogeneity in the study design.

RESULTS

The included studies featured diverse ML and DL algorithms, with Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) being the most common. Gradient boosting (GB) showed superior performance. Median Area Under the Curve (AUC) values were 0.91 for GB, 0.83 for RF, 0.77 for LR, and 0.76 for SVM. Neural networks had a median AUC of 0.81 and convolutional neural networks (CNNs) had a median AUC of 0.91. ML techniques outperformed conventional models, particularly those integrating clinical and imaging data.

CONCLUSIONS

ML and DL models significantly surpass traditional scoring systems in predicting HT. These advanced models enhance clinical decision-making and improve patient outcomes. Future research should address data expansion, imaging protocol standardization, and model transparency to enhance stroke outcomes further.

Constructing machine learning models based on non-contrast CT radiomics to predict hemorrhagic transformation after stoke: a two-center study

Purpose

Machine learning (ML) models were constructed according to non-contrast computed tomography (NCCT) images as well as clinical and laboratory information to assess risk stratification for the occurrence of hemorrhagic transformation (HT) in acute ischemic stroke (AIS) patients.

Methods

A retrospective cohort was constructed with 180 AIS patients who were diagnosed at two centers between January 2019 and October 2023 and were followed for HT outcomes. Patients were analyzed for clinical risk factors for developing HT, infarct texture features were extracted from NCCT images, and the radiomics score (Rad-score) was calculated. Then, five ML models were established and evaluated, and the optimal ML algorithm was used to construct the clinical, radiomics, and clinical-radiomics models. Receiver operating characteristic (ROC) curves were used to compare the performance of the three models in predicting HT.

Results

Based on the outcomes of the AIS patients, 104 developed HT, and the remaining 76 had no HT. The HT group consisted of 27 hemorrhagic infarction (HI) and 77 parenchymal-hemorrhage (PH). Patients with HT had a greater neutrophil-to-lymphocyte ratio (NLR), baseline National Institutes of Health Stroke Scale (NIHSS) score, infarct volume, and Rad-score and lower Alberta stroke program early CT score (ASPECTS) (all p < 0.01) than patients without HT. The best ML algorithm for building the model was logistic regression. In the training and validation cohorts, the AUC values for the clinical, radiomics, and clinical-radiomics models for predicting HT were 0.829 and 0.876, 0.813 and 0.898, and 0.876 and 0.957, respectively. In subgroup analyses with different treatment modalities, different infarct sizes, and different stroke time windows, the assessment accuracy of the clinical-radiomics model was not statistically meaningful (all p > 0.05), with an overall accuracy of 79.5%. Moreover, this model performed reliably in predicting the PH and HI subcategories, with accuracies of 82.9 and 92.9%, respectively.

Conclusion

ML models based on clinical and NCCT radiomics characteristics can be used for early risk evaluation of HT development in AIS patients and show great potential for clinical precision in treatment and prognostic assessment.

Hemorrhagic transformation in patients with large-artery atherosclerotic stroke is associated with the gut microbiota and lipopolysaccharide

JOURNAL/nrgr/04.03/01300535-202407000-00032/figure1/v/2023-11-20T171125Z/r/image-tiff

Hemorrhagic transformation is a major complication of large-artery atherosclerotic stroke (a major ischemic stroke subtype) that worsens outcomes and increases mortality. Disruption of the gut microbiota is an important feature of stroke, and some specific bacteria and bacterial metabolites may contribute to hemorrhagic transformation pathogenesis. We aimed to investigate the relationship between the gut microbiota and hemorrhagic transformation in large-artery atherosclerotic stroke. An observational retrospective study was conducted. From May 2020 to September 2021, blood and fecal samples were obtained upon admission from 32 patients with first-ever acute ischemic stroke and not undergoing intravenous thrombolysis or endovascular thrombectomy, as well as 16 healthy controls. Patients with stroke who developed hemorrhagic transformation (n = 15) were compared to those who did not develop hemorrhagic transformation (n = 17) and with healthy controls. The gut microbiota was assessed through 16S ribosomal ribonucleic acid sequencing. We also examined key components of the lipopolysaccharide pathway: lipopolysaccharide, lipopolysaccharide-binding protein, and soluble CD14. We observed that bacterial diversity was decreased in both the hemorrhagic transformation and non-hemorrhagic transformation group compared with the healthy controls. The patients with ischemic stroke who developed hemorrhagic transformation exhibited altered gut microbiota composition, in particular an increase in the relative abundance and diversity of members belonging to the Enterobacteriaceae family. Plasma lipopolysaccharide and lipopolysaccharide-binding protein levels were higher in the hemorrhagic transformation group compared with the non-hemorrhagic transformation group. lipopolysaccharide, lipopolysaccharide-binding protein, and soluble CD14 concentrations were associated with increased abundance of Enterobacteriaceae. Next, the role of the gut microbiota in hemorrhagic transformation was evaluated using an experimental stroke rat model. In this model, transplantation of the gut microbiota from hemorrhagic transformation rats into the recipient rats triggered higher plasma levels of lipopolysaccharide, lipopolysaccharide-binding protein, and soluble CD14. Taken together, our findings demonstrate a noticeable change in the gut microbiota and lipopolysaccharide-related inflammatory response in stroke patients with hemorrhagic transformation. This suggests that maintaining a balanced gut microbiota may be an important factor in preventing hemorrhagic transformation after stroke.

Application of interpretable machine learning algorithms to predict acute kidney injury in patients with cerebral infarction in ICU

BACKGROUND

Acute kidney injury (AKI) is not only a complication but also a serious threat to patients with cerebral infarction (CI). This study aimed to explore the application of interpretable machine learning algorithms in predicting AKI in patients with cerebral infarction.

METHODS

The study included 3920 patients with CI admitted to the Intensive Care Unit and Emergency Medicine of the Central Hospital of Lishui City, Zhejiang Province. Nine machine learning techniques, including XGBoost, logistics, LightGBM, random forest (RF), AdaBoost, GaussianNB (GNB), Multi-Layer Perceptron (MLP), support vector machine (SVM), and k-nearest neighbors (KNN) classification, were used to develop a predictive model for AKI in these patients. SHapley Additive exPlanations (SHAP) analysis provided visual explanations for each patient. Finally, model effectiveness was assessed using metrics such as average precision (AP), sensitivity, specificity, accuracy, F1 score, precision-recall (PR) curve, calibration plot, and decision curve analysis (DCA).

RESULTS

The XGBoost model performed better in the internal validation set and the external validation set, with an AUC of 0.940 and 0.887, respectively. The five most important variables in the model were, in order, glomerular filtration rate, low-density lipoprotein, total cholesterol, hemiplegia and serum kalium.

CONCLUSION

This study demonstrates the potential of interpretable machine learning algorithms in predicting CI patients with AKI.