Novel hippocampus-centered methodology for informative instance selection in Alzheimer's disease data.

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Data Analysis

Background:

• Dataset size and quality are critical for prediction model performance.
• Instance selection techniques are vital for optimizing models by reducing data size and computational costs.
• Magnetic resonance imaging (MRI) data holds potential for Alzheimer's disease (AD) diagnosis.

Purpose of the Study:

• To propose a novel methodology for identifying informative 2D MRI slices for Alzheimer's disease study.
• To leverage hippocampus-centered analysis across multiple atlases for enhanced data selection.
• To improve the efficiency and accuracy of Alzheimer's disease prediction models.

Main Methods:

• Developed a novel methodology for selecting informative 2D MRI slices.
• Employed a hippocampus-centered analysis utilizing data from multiple atlases.
• Constructed convolutional neural networks (CNNs) for Alzheimer's disease classification.
• Consolidated data from three standard datasets: ADNI, AIBL, and OASI.

Main Results:

• The proposed methodology effectively identified informative MRI slices.
• CNNs trained on the selected data achieved high subject-level classification accuracy.
• Demonstrated approximately $$ accuracy in distinguishing between normal cognition and Alzheimer's disease.

Conclusions:

• The novel instance selection methodology significantly enhances Alzheimer's disease detection using MRI data.
• Hippocampus-centered analysis combined with multi-atlas data improves the identification of informative slices.
• This approach offers a promising strategy for developing more accurate and computationally efficient Alzheimer's disease prediction models.