Zoom-In Neural Network Deep-Learning Model for Alzheimer's Disease Assessments.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Deep neural networks (DNNs) show promise in analyzing complex medical data.
• Alzheimer's disease (AD) and mild cognitive impairment (MCI) require accurate diagnostic tools.

Purpose of the Study:

• To evaluate the efficacy of a novel deep-learning algorithm, the zoom-in neural network (ZNN), for classifying individuals with AD, MCI, and normal controls (NC).
• To identify brain regions implicated in AD progression using neuroimaging data.

Main Methods:

• Utilized a deep-learning algorithm, the zoom-in neural network (ZNN), which employs a hierarchical structure of zoom-in learning units (ZLUs) without backpropagation.
• Employed resting-state functional magnetic resonance imaging (rs-fMRI) data from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
• Applied the Automated Anatomical Labeling (AAL-90) atlas to analyze 90 neuroanatomical regions and extracted features from 140-time series rs-fMRI voxel values.

Main Results:

• Achieved high classification accuracies: 97.7% for AD vs. MCI and NC, 84.8% for NC vs. AD and MCI, and 72.7% for MCI vs. AD and NC.
• Identified seven discriminative regions of interest (ROIs) within the AAL-90 atlas crucial for classification.

Conclusions:

• The ZNN deep-learning algorithm demonstrates significant potential for accurate AD and MCI assessment using rs-fMRI data.
• The identified ROIs offer insights into the neuroanatomical underpinnings of AD and MCI.