Fully Connected Multi-Kernel Convolutional Neural Network Based on Alzheimer's Disease Diagnosis.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) and mild cognitive impairment (MCI) diagnosis lacks sufficient clinical expertise.
• Current diagnostic methods like cerebrospinal fluid analysis and PET scans are invasive.
• Diffusion tensor imaging (DTI) offers a potential non-invasive alternative for diagnosis.

Purpose of the Study:

• To develop a high-precision deep learning model for automatic AD and MCI diagnosis using DTI.
• To create feature probability maps for auxiliary clinical diagnosis.
• To evaluate the performance of a novel Factorization Machine combined Neural Network (FMCNN) model.

Main Methods:

• Preprocessing DTI data to remove external influences.
• Tracking fiber bundles (corpus callosum, cingulum, uncinate fasciculus, white matter) using deterministic fiber tracking.
• Classifying streamlines with Convolutional Neural Network (CNN), Multi-function CNN (MCNN), and FMCNN models.
• Generating fiber risk probability maps using the FMCNN model.

Main Results:

• The FMCNN model demonstrated superior performance in accuracy, specificity, sensitivity, and area under the curve compared to CNN and MCNN.
• White matter (WM) analysis using FMCNN achieved the highest accuracy of 96.95%.
• The model successfully generated fiber probability maps indicating disease risk.

Conclusions:

• The proposed FMCNN model accurately diagnoses Alzheimer's disease and mild cognitive impairment.
• The generated fiber probability maps effectively represent the risk status for AD and MCI.
• This DTI-based deep learning approach offers a promising non-invasive diagnostic tool.