MRI-based machine learning model: A potential modality for predicting cognitive dysfunction in patients with type 2...

Area of Science:

• Neuroimaging
• Machine Learning
• Diabetology

Background:

• Type 2 Diabetes Mellitus (T2DM) is a significant risk factor for cognitive impairment.
• Current neuropsychiatric screening tests for cognitive function in T2DM patients may lack repeatability.
• Machine learning (ML) shows potential in assessing cognitive impairment, particularly in Alzheimer's disease (AD).

Purpose of the Study:

• To develop and evaluate MRI-based ML models for predicting cognitive dysfunction in T2DM patients.
• To compare the performance of different ML classifiers in identifying cognitive states within the T2DM population.

Main Methods:

• Utilized Fluid Attenuated Inversion Recovery (FLAIR) MRI data from 122 T2DM patients.
• Assessed cognitive function using the Chinese version of the Montréal Cognitive Assessment Scale-B (MoCA-B), categorizing patients into Dementia (DM), Mild Cognitive Impairment (MCI), and Normal (N) groups.
• Extracted radiomics features, applied feature selection (variance threshold, SelectKBest, LASSO), and constructed ML models (KNN, SVM, LR) for classification.

Main Results:

• 1,409 radiomics features were reduced to 13 optimal discriminators.
• The Logistic Regression (LR) classifier achieved the highest predictive performance in the validation set.
• LR demonstrated AUCs of 0.831 (DM), 0.883 (MCI), and 0.904 (N), outperforming SVM and KNN.

Conclusions:

• MRI-based ML models show significant potential for predicting cognitive dysfunction in T2DM.
• The Logistic Regression algorithm exhibited superior performance compared to SVM and KNN for this predictive task.
• These findings suggest a novel, potentially more repeatable approach for cognitive assessment in T2DM patients.