Glia Biomarkers and Explainable AI: A New Approach to Precision Classification of Cognitive Impairment

This study integrates glial biomarkers and explainable machine learning (SHAP method). By combining glial cell markers in cerebrospinal fluid (such as GFAP, YKL-40, and Cystatin C) with traditional Alzheimer's disease biomarkers (Aβ42, tau), it significantly improves the accuracy of cognitive impairment classification and enhances the clinical interpretability of the model, providing a new approach for the precision classification of cognitive impairment.

The acceleration of global population aging has made cognitive impairments such as Alzheimer's disease (AD) a public health challenge. Early and accurate diagnosis is crucial, but traditional methods (neuropsychological assessment, cerebrospinal fluid biomarkers, neuroimaging) have limitations: traditional markers (Aβ42, tau) only reflect neurodegenerative changes and cannot fully capture complex pathology. Recent studies have revealed that neuroinflammation plays a key role in the pathogenesis of AD, and glial cells (microglia, astrocytes) are the core of inflammation, raising the question of whether incorporating glial markers into diagnostic models can improve accuracy.

Study Design: A longitudinal observational study involving 333 adults aged over 60 (cognitively normal individuals and patients with varying degrees of cognitive impairment), with data from the Knight Alzheimer's Disease Research Center.

Biomarker Combination:

• Traditional neurodegenerative markers: Aβ42 (amyloid deposition), total tau/phosphorylated tau (neurofibrillary tangles)
• Glial activation markers: GFAP (astrocyte activation), YKL-40 (microglial activation), Cystatin C (neuroinflammation-related protease inhibitor)

Machine Learning Models: Basic models (logistic regression, SVM, random forest), ensemble models (gradient boosting + fusion), deep learning models, hybrid models

Interpretability: SHAP value analysis was used to quantify the contribution of each marker to predictions, enhancing model transparency.

Improvement in Classification Performance:

• Using only traditional markers: AUC=0.868, F1=0.309
• After adding glial markers: AUC increased to 0.959, F1 reached 0.951, showing obvious advantages in distinguishing mild cognitive impairment (MCI) from normal cognition and predicting MCI conversion to dementia.

Contribution of Glial Markers:

• Cystatin C is one of the most influential predictors, with a contribution exceeding some traditional markers
• YKL-40 is key in distinguishing inflammation-related subtypes
• GFAP is associated with disease severity and progression rate

Model Balance: A streamlined model with 7 variables was obtained through feature selection, maintaining high accuracy (AUC=0.912) while reducing the cost of clinical application.

Precision Stratification: Integrating glial markers can identify inflammation-dominant, pure degenerative, and mixed pathological types of cognitive impairment, guiding personalized treatment (such as anti-inflammatory, amyloid/tau-targeted therapies, etc.).

Early Warning: Changes in glial markers precede the deterioration of clinical symptoms, helping to identify high-risk groups early and initiate interventions in the reversible stage.

Drug Development: Provides companion diagnostic tools for neuroinflammation-related AD drugs (such as microglial modulators) to screen patients who will benefit.

Limitations:

• Samples are from a single center with concentrated ethnic and geographical distribution
• Limited longitudinal follow-up time
• Insufficient mechanistic explanation

Future Directions:

• Multi-omics integration (cerebrospinal fluid + blood + genome/proteome)
• Imaging-marker fusion (PET/MRI + body fluids)
• Dynamic monitoring (glial markers + wearables + cognitive tests)
• AI optimization (graph neural networks, federated learning)

This study is an important progress in AI-assisted diagnosis of neurodegenerative diseases. By combining glial markers and explainable AI, it improves classification accuracy and enhances clinical trust. This 'data-driven + mechanistic understanding' strategy accelerates the transformation of cognitive impairment diagnosis and treatment from empirical medicine to evidence-based medicine, and is expected to enter clinical routine practice in the future.