An international consortium within the DEMON Network systematically reviews 75 studies applying machine learning to cerebral small vessel disease markers in MRI, achieving pooled AUCs of 0.88 for Alzheimer’s dementia and 0.84 for cognitive impairment.
Key points
- Meta-analysis of 16 studies shows pooled AUCs of 0.88 for Alzheimer’s dementia and 0.84 for cognitive impairment.
- ML algorithms—SVM, logistic regression, random forests, CNNs—use CSVD markers (WMH, lacunes, microbleeds) from MRI for classification.
- Only 5/75 studies performed external dataset validation, underscoring the need for broader generalisability testing.
Why it matters: Demonstrating high diagnostic performance of ML on vascular MRI markers highlights a new avenue to integrate cerebrovascular features into AI-driven dementia screening and personalized care.
Q&A
- What are CSVD markers?
- Why use area under the ROC curve (AUC)?
- Why is external validation crucial?
- How do ML models process vascular MRI data?
