Multimodal deep learning outperforms clinical and brain region models in predicting stroke-associated pneumonia: an explainable AI study

 

Predictions are useless!  Solve the correct problem!  You've known of the need to prevent pneumonia for years! I'd have you all fired for incompetence!

• 11% Stroke-associated pneumonia (21 posts to October 2020)

Multimodal deep learning outperforms clinical and brain region models in predicting stroke-associated pneumonia: an explainable AI study

• 

  Haoyang He 1†

• 

  Xu Zhang 1†

• 

  Lijuan Gu 2*

• 

  Zhihong Jian 1*

• 

  Xiaoxing Xiong 1*

• 1. Department of Neurosurgery, Renmin Hospital of Wuhan University, Wuhan, China

• 2. Central Laboratory, Renmin Hospital of Wuhan University, Wuhan, China

Abstract

Background: 

Stroke-associated pneumonia (SAP) is a frequent complication after acute ischemic stroke (AIS) and is associated with poor outcomes. This study aimed to develop an interpretable multimodal deep learning model integrating MRI, lesion-related brain regions, and clinical variables for early SAP prediction.

Methods: 

A total of 426 AIS patients were retrospectively enrolled, including 71 patients with SAP. Multimodal MRI data (DWI, T1WI, and T2-FLAIR) were processed using standardized registration and lesion segmentation. A 3D convolutional neural network was used to extract imaging representations, which were fused with clinical variables and AAL3-based brain-region features. Model performance was assessed using stratified five-fold cross-validation and nested cross-validation when applicable, with further evaluation based on receiver operating characteristic (ROC) analysis, calibration analysis, and decision curve analysis. Grad-CAM was applied for model interpretation.

Results: 

The multimodal fusion model achieved the best performance for SAP prediction, with an AUC of 0.782 (95% CI: 0.712–0.839), compared with the clinical model based on conventional clinical variables (AUC = 0.756), the imaging model based on 3D CNN representations (AUC = 0.693), and the brain-region model based on AAL3-derived lesion location features (AUC = 0.501). The fusion model showed superior clinical utility and favorable calibration. Grad-CAM visualization demonstrated that model predictions were mainly driven by lesion-related cortical and subcortical regions.

Conclusion: 

A multimodal deep learning framework integrating MRI, brain-region information, and clinical characteristics improved SAP prediction after AIS and provided an interpretable approach for individualized risk stratification.

Construction and multi-center validation of a model based on blending ensemble learning for predicting stroke-associated pneumonia following acute ischemic stroke

 

Predictions are useless!  Solve the correct problem!  You've known of the need to prevent pneumonia for years! I'd have you all fired for incompetence!

• 11% Stroke-associated pneumonia (21 posts to October 2020)

Construction and multi-center validation of a model based on blending ensemble learning for predicting stroke-associated pneumonia following acute ischemic stroke

• 

  Xinying Liu 1,2†

• Q

  Qiang Fu 1,2†

• S

  Shuo Yang 1,2†

• 

  Lin Zhang 1,2

• 

  Boxuan Feng 1,2

• 

  Ying Gao 1,2

• 

  Chengboya Zhao 1,2

• D

  Di Gao 1,2

• G

• 

• X

  Xiaokun Yang 1,2*

• 1. First Teaching Hospital of Tianjin University of Traditional Chinese Medicine, Tianjin, China

• 2. National Clinical Research Center for Chinese Medicine, Tianjin, China

Abstract

Background: 

Stroke-associated pneumonia (SAP) is a common and serious infectious complication following acute ischemic stroke (AIS), leading to worsened clinical outcomes. We aimed to develop and validate a novel predictive model based on Blending Ensemble Learning, to enable early, precise screening and individualized risk stratification for SAP.

Methods: 

In this multi-center study, a derivation cohort was initially established using 1,123 patients admitted between 2022 and 2024. Subsequently, 199 patients were prospectively enrolled in early 2025 as a temporal validation set, and 287 patients were extracted from a regional health big data platform (covering 12 hospitals) to serve as an external validation set. High-value clinical and biomarker features were identified via least absolute shrinkage and selection operator (LASSO) regression. To overcome the limitations of single classifiers, a Blending Ensemble Learning strategy was adopted to construct the final predictive model. The model underwent comprehensive evaluation for discrimination, calibration, and clinical utility, and was benchmarked against single random forest (RF) and extreme gradient boosting (XGBoost) models.

Results: 

Seventeen predictors were identified, including systemic immune-inflammation index (SII), neutrophil-to-lymphocyte ratio (NLR), the National Institutes of Health Stroke Scale (NIHSS), and D-dimer. The Blending ensemble model demonstrated superior discrimination in the validation cohort, achieving an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.898 (95% CI: 0.845–0.951), which significantly outperformed the optimal single model. Furthermore, SHapley Additive exPlanations (SHAP) analysis highlighted hyperinflammation, immunothrombosis, and nutritional depletion as pivotal mechanisms driving SAP pathogenesis.

Conclusion: 

The proposed Blending ensemble model exhibits excellent generalizability and high sensitivity. The accompanying web-based tool provides robust decision support for identifying high-risk patients and implementing early precision interventions.

Identifying key predictors of post-stroke depression and cognitive impairment in acute stroke survivors

 Predictions are useless! Both of these are solved by EXACT 100% RECOVERY PROTOCOLS! Solve the correct problem! I'd have you all fired for incompetence!

And the commentors on this need to be fired also for missing the real problem; lack of 100% recovery protocols!

Commentary: Identifying key predictors of post-stroke depression and cognitive impairment in acute stroke survivors 

The latest here:

Identifying key predictors of post-stroke depression and cognitive impairment in acute stroke survivors

• 

  Zhiwen Yan 1†

• H

  Huan Zhao 1†

• 

  Jianjun Chen 2

• F

  Fang Liu 1

• L

  Lei Gong 1

• 

  Yingli Li 1

• J

  Jie Zhang 1

• M

  Mi Xiao 1

• 

  Jun Mu 1*

• 1. Department of Neurology, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China

• 2. Chongqing Key Laboratory of Neurobiology, Institute of Neuroscience, Chongqing Medical University, Chongqing, China

Abstract

Background: 

Post-stroke depression (PSD) and post-stroke cognitive impairment (PSCI) are prevalent complications in aging stroke survivors and are often overlooked due to the lack of early diagnostic indicators, leading to poor prognosis. Identifying reliable predictors is crucial for timely intervention.

Methods: 

This prospective cohort study followed 78 acute stroke survivors for 6 months. A composite neuropsychological outcome—defined as the development of PSD and/or PSCI—was determined using the Diagnostic and Statistical Manual of Mental Disorders-5th Edition (DSM-5) and NINDS-CSN criteria. To account for the limited sample size, multivariable Firth’s penalized logistic regression was employed to identify independent predictors, generating robust odds ratios (ORs) and 95% confidence intervals (CIs). An exploratory classification and regression tree (CART) analysis was also conducted for hypothesis generation.

Results: 

The final cohort comprised 78 acute ischemic stroke survivors with a median age of 62 years (IQR 51–71). Among these participants, 26.0% were women, and the median admission score on the National Institutes of Health Stroke Scale (NIHSS) was 3 (IQR 1–5). Within 6 months, 56 patients (71.8%) developed the composite outcome (13 experienced PSCI alone, 24 had PSD alone, and 19 had both conditions). A multivariable analysis revealed that right hemisphere lesions (OR = 9.019, 95% CI: 1.329–61.213, p = 0.016), greater baseline emotional distress (higher 9-item Patient Health Questionnaire (PHQ-9) scores; OR = 5.157, 95% CI: 1.835–14.494, p < 0.001), and pre-existing cognitive vulnerability (lower Mini–Mental State Examination (MMSE) scores; OR = 0.714, 95% CI: 0.517–0.984, p = 0.023) were independent predictors of poor neuropsychological outcomes. Advanced age (p = 0.094) and elevated urea levels (p = 0.095) showed only marginal trends. Exploratory CART modeling highlighted the hierarchical interaction of these baseline clinical scores for risk stratification.

Conclusion: 

Right hemisphere lesions, early emotional distress, and baseline cognitive vulnerability independently predicted a high risk of composite neuropsychological impairment at 6 months post-stroke. Rather than serving merely as novel biomarkers, high baseline PHQ-9 scores and low MMSE scores reflected the persistence of early distress and poor cognitive reserve, respectively. These highly accessible clinical parameters facilitate early risk stratification, emphasizing the absolute need for immediate psychological triage and integrated, long-term cognitive-emotional monitoring.

Association between complete blood count-derived hematological inflammatory ratios and nutritional risk in elderly patients with acute ischemic stroke

 

'Associations' don't get you recovered, or are you too fucking stupid to see that? What prevents nutritional risk is the needed research, not this crapola! Like EXACT DIET PROTOCOLS, starting in the hospital!

• 49% malnutrition post-stroke (1 posts to April 2025)

You've known of the risk for over a year! 

SOLVE THE GODDAMN PROBLEM!

Association between complete blood count-derived hematological inflammatory ratios and nutritional risk in elderly patients with acute ischemic stroke

• 

  Liang’e Xu 1

• 

  Wenjing Wu 2

• 

  Chao Chen 3*

• 1. Department of Neurology, The Third Affiliated Hospital of Wenzhou Medical University, Wenzhou, China

• 2. Department of Neurology, The Second Affiliated Hospital of Zhejiang Chinese Medical University, Hangzhou, China

Abstract

Background and aims: 

Mounting evidence suggests CBC-derived hematological inflammatory ratios correlate with nutritional status, yet few studies explore their cross-sectional associations with GNRI-defined concurrent nutritional risk in elderly patients with acute ischemic stroke (AIS). This single-center, cross-sectional, observational study was designed to evaluate the correlation between routine admission hematological markers and nutritional risk in elderly patients with AIS. Twelve CBC-derived indices were analyzed: lymphocyte ratio (LR), red blood cell-to-lymphocyte ratio (RLR), hemoglobin-to-lymphocyte ratio (HLR), monocyte-to-lymphocyte ratio (MLR), monocyte-to-neutrophil ratio (MNR), neutrophil-to-lymphocyte ratio (NLR), NLR-to-platelet ratio (NLR/PLT100), MLR-to-platelet ratio (MLR/PLT), platelet-to-neutrophil ratio (PNR), platelet-to-lymphocyte ratio (PLR), mean platelet volume-to-lymphocyte ratio (MPVLR), serum albumin-to-lymphocyte ratio (ALBLR).

Methods: 

Between January 2022 and January 2024, 540 elderly AIS patients (≥60 years) were enrolled. All participants underwent admission GNRI assessment and were split into GNRI-defined nutritional risk group (GNRI≤98, n = 245, 45.4%) and non-nutritional risk group (GNRI>98, n = 295). Venous blood was collected within 24 h on admission. Univariate and multivariate logistic regression were applied to explore cross-sectional correlations, and an exploratory combined statistical model was built. ROC curve, DeLong’s test, likelihood ratio test and AIC were used to evaluate model discrimination and fit.

Results: 

Multiple indicators including age, hypertension, ALT, UA, TG, LR, RLR, MLR, NLR, NLR/PLT100, MLR/PLT, PLR and MPVLR showed univariate correlations with nutritional risk. Multivariate regression identified MLR and NLR/PLT100 as independent correlates within this cohort; MLR showed the strongest association in this dataset instead of robust correlative ability. Adding MLR alone or combining MLR + NLR/PLT100 to base clinical variables significantly improved C-statistic. The combined model reached an AUC of 0.735, representing only moderate discriminative capacity, and it was merely the superior exploratory model limited to the present sample without external generalizability.

Conclusion: 

Within this single-center cohort, MLR exhibited the strongest cross-sectional correlation with GNRI-defined nutritional risk among all tested CBC-derived hematological inflammatory ratios. The composite model combining baseline clinical indicators, MLR and NLR/PLT*100 demonstrated moderate discriminative ability within the study sample, but is not yet robust enough to serve as an effective tool for routine clinical risk stratification.