The value of machine learning models in differentiating alzheimer's disease from Moderate-to-Severe cerebral small vessel disease.

OBJECTIVE

To establish machine learning models to differentiate Alzheimer’s Disease (AD) from moderate-to-severe cerebral small vessel disease (CSVD) by comparing their imaging features.

METHODS

Imaging data were retrospectively obtained from 50 patients diagnosed with AD and 98 patients presenting moderate-to-severe CSVD. Apparent diffusion coefficients (ADC) within various brain regions, as well as anatomical structural volumes, were quantified using an MRI-based brain segmentation approach. Comparative analyses of imaging characteristics between AD and CSVD patient groups were performed. Subsequently, binary logistic regression analysis was employed to identify significant imaging markers. Predictive models, including logistic regression (Logit), support vector machine (SVM), gradient boosting machine (GBM), and decision tree (DT) classifiers, were established and compared for their ability to distinguish AD from CSVD.

RESULTS

The logistic regression model differentiated AD with a sensitivity of 76.53%, specificity of 79.00%, and an area under the curve (AUC) of 0.778. In comparison, the decision tree model yielded a sensitivity of 80.73%, specificity of 88.76%, and an AUC of 0.844. The SVM model produced a sensitivity of 78.57%, specificity of 83.00%, and an AUC of 0.808. Finally, the GBM model exhibited a sensitivity of 84.69%, specificity of 87.00%, and an AUC of 0.859.

CONCLUSION

The gradient boosting machine model demonstrated the best diagnostic performance for identifying AD (AUC = 0.864), is suitable for clinical screening due to its high sensitivity and specificity.