Improved perfusion pattern score association with type 2 diabetes severity using machine learning pipeline: Pilot...

Area of Science:

• Neuroimaging
• Medical Machine Learning
• Cerebrovascular Health

Background:

• Type 2 diabetes mellitus (T2DM) impacts brain health, affecting the blood-brain barrier, neurons, and arterial stiffness, which influences cerebral metabolism and blood flow.
• A need exists for advanced methods, like machine learning, to detect T2DM-related cerebral perfusion patterns and their links to cognitive function and disease severity.

Purpose of the Study:

• To develop and validate a machine-learning pipeline for identifying T2DM-related cerebral perfusion patterns.
• To assess the pipeline's ability to differentiate T2DM patients from controls.
• To explore associations between the identified pattern and cognitive performance or disease severity.

Main Methods:

• Utilized a machine-learning pipeline involving principal component analysis, feature selection, and logistic regression on 3T pseudocontinuous arterial spin-labeling MRI data.
• Included cross-sectional and 2-year longitudinal data from 73 subjects at baseline and 42 at follow-up.
• Employed linear regression analysis, controlling for gray matter volume and education years.

Main Results:

• The machine-learning approach demonstrated superior discriminative value compared to univariate methods, with significant improvements in accuracy, AUC, and predictive values.
• Individual pattern scores correlated with diabetes severity, mobility, and cognitive performance at baseline.
• Longitudinal changes in individual pattern scores were significantly associated with changes in HbA1c and baseline cholesterol levels.

Conclusions:

• The developed machine-learning pipeline effectively identifies a T2DM-related perfusion pattern.
• The individual perfusion pattern score shows potential as a biomarker for monitoring T2DM progression.
• Further validation in larger cohorts is recommended.