Classifying cognitive impairment based on FDG-PET and combined T1-MRI and rs-fMRI: An ADNI study

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• Mild cognitive impairment (MCI) is a precursor to dementia, necessitating advanced diagnostic tools.
• Heterogeneity of MCI requires novel approaches for accurate classification and prediction.
• Fluorodeoxyglucose positron emission tomography (FDG-PET) offers robust Alzheimer's disease (AD) biomarkers.

Purpose of the Study:

• To classify individuals with MCI versus cognitively normal (CN) adults using FDG-PET.
• To predict MCI individuals who will convert to AD dementia.
• To compare the diagnostic performance of MRI against FDG-PET for MCI classification.

Main Methods:

• A 3D DenseNet deep learning model was trained on FDG-PET images from 805 participants (MCI/CN).
• MRI and functional MRI data from 348 participants were used to train alternative classification models.
• Performance was evaluated using 10-fold cross-validation across four ADNI cohorts.

Main Results:

• The FDG-PET model achieved 88.02% accuracy in classifying MCI vs. CN.
• MRI-based models yielded lower accuracy, with the highest reaching 76.86%.
• The FDG-PET model showed 63.23% accuracy in predicting MCI converters versus stable MCI.

Conclusions:

• FDG-PET imaging demonstrates superior accuracy for classifying MCI versus CN.
• Although less accurate than FDG-PET, multi-contrast MRI provides valuable insights for neurodegeneration classification.
• Deep learning models show promise for analyzing neuroimaging data in MCI and AD research.