Early-Stage Alzheimer's Disease Categorization Using PET Neuroimaging Modality and Convolutional Neural Networks in...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's Disease (AD) is a progressive neurodegenerative disorder impacting global aging populations.
• AD is characterized by neuronal loss and the hallmark pathologies of neurofibrillary tangles and amyloid plaques.
• Neuroimaging and deep learning are crucial for identifying AD-related brain changes and patterns.

Purpose of the Study:

• To investigate the efficacy of 2D and 3D Convolutional Neural Network (CNN) architectures for classifying early-stage Alzheimer's Disease (AD).
• To differentiate between AD, Mild Cognitive Impairment (MCI), and Normal Control (NC) using Positron Emission Tomography (PET) neuroimaging.
• To evaluate the impact of data augmentation and novel deep learning techniques on classification performance.

Main Methods:

• Utilized 2D and 3D CNN architectures for classification tasks, including binary (AD/NC, AD/MCI, MCI/NC) and multiclass (AD/NC/MCI).
• Employed Positron Emission Tomography (PET) neuroimaging data.
• Implemented data augmentation via random zooming, blurring before subsampling, and distant domain transfer learning for 2D CNNs.
• Performed classification using a five-fold cross-validation approach for hyperparameter selection.

Main Results:

• The 3D-CNN architecture demonstrated superior performance across all classification tasks.
• Achieved highest accuracy of 89.21% for AD/NC, 71.70% for AD/MCI, 62.25% for MCI/NC, and 59.73% for the multiclass AD/NC/MCI task.
• Data augmentation significantly improved performance on the multiclass classification task.

Conclusions:

• Deep learning models, particularly 3D-CNNs, are effective tools for the early recognition of Alzheimer's Disease.
• The findings support the clinical application of advanced AI techniques in neuroimaging for AD diagnosis.
• Optimized CNN architectures and data augmentation strategies enhance the accuracy of early AD detection.