Combined Quantitative amyloid-β PET and Structural MRI Features Improve Alzheimer's Disease Classification in Random...

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Machine Learning in Medicine

Background:

• Neuropathological changes like amyloid-β (Aβ) plaques and brain atrophy precede clinical Alzheimer's Disease (AD) symptoms.
• Multimodal biomarker assessment can improve the accuracy of AD diagnosis.

Purpose of the Study:

• To evaluate the discriminative power of Aβ PET imaging features in classifying AD.
• To determine if combining Aβ PET and structural MRI (sMRI) features improves machine learning model performance for differentiating AD from older healthy controls (OHC) and mild cognitive impairment (MCI).

Main Methods:

• Utilized data from 94 AD, 82 MCI, and 85 OHC patients across three cohorts.
• Extracted 17 Aβ PET features (Centiloid), 122 regional volumes, and 68 cortical thickness measures.
• Trained a random forest model using single or combined imaging modalities.

Main Results:

• Aβ PET features achieved AUC scores of 0.86 in differentiating AD from OHC and 0.68 in differentiating AD from MCI.
• sMRI features yielded AUC scores of 0.81 for AD vs. OHC and 0.69 for AD vs. MCI.
• Combining Aβ PET and sMRI features improved classification performance to AUC 0.89 (AD vs. OHC) and 0.71 (AD vs. MCI).
• Specific Aβ PET features demonstrated higher discriminative value than sMRI features.

Conclusions:

• Aβ PET imaging possesses significant discriminative ability for identifying AD.
• Integrating Aβ PET and sMRI data enhances the performance of random forest models for AD classification.