Alzheimer's disease diagnosis framework from incomplete multimodal data using convolutional neural networks.

Area of Science:

• Neuroscience
• Medical Imaging
• Genetics

Background:

• Alzheimer's disease (AD) and mild cognitive impairment (MCI) require early detection for effective management.
• Current methods struggle with missing multimodal data (genetics, MRI, PET) and data heterogeneity.
• Existing studies often focus solely on classification, neglecting regression tasks for disease progression.

Purpose of the Study:

• To develop a unified framework for joint classification and clinical score regression of AD/MCI using multimodal data.
• To address challenges of missing data and data heterogeneity in neuroimaging and genetic datasets.
• To improve the accuracy and scope of predictive models for Alzheimer's disease and its prodromal stages.

Main Methods:

• A novel multimodal data fusion framework using convolutional neural networks (CNNs).
• Linear interpolation technique to impute missing genetic and imaging features.
• Joint learning of neuroimaging (MRI, PET) and genetic (SNP) data within a unified CNN architecture.
• Simultaneous classification of AD/MCI and regression of clinical scores.

Main Results:

• Achieved high classification accuracies: 98.22% (NC vs. AD), 93.11% (NC vs. sMCI), and 97.35% (NC vs. pMCI).
• Demonstrated superior performance in clinical score regression, achieving the lowest RMSE and highest correlation coefficients.
• Validated on 805 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

Conclusions:

• The proposed CNN-based multimodal fusion effectively handles missing data and heterogeneity for AD/MCI detection.
• The unified framework successfully performs both classification and regression tasks, outperforming existing methods.
• This approach offers a promising tool for early and accurate diagnosis and progression prediction in Alzheimer's disease.