Lightweight neural network for Alzheimer's disease classification using multi-slice sMRI.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder.
• Early detection and intervention are critical for managing AD progression.
• Structural Magnetic Resonance Imaging (sMRI) is a key tool for AD diagnosis.

Purpose of the Study:

• To develop an efficient and scalable automated system for AD detection using sMRI.
• To leverage a lightweight neural network for resource-constrained environments.
• To improve the accuracy and speed of AD diagnosis.

Main Methods:

• A lightweight neural network based on ShuffleNet V1 architecture was employed for feature extraction from multi-slice sMRI data.
• Efficient Channel Attention (ECA) was integrated to enhance disease-related brain region features.
• Cross-entropy and triplet loss functions were utilized for model optimization.

Main Results:

• The model achieved high classification accuracies: 95.00% for AD vs. CN, 87.50% for AD vs. MCI, and 85.62% for MCI vs. CN.
• The proposed method uses only 3.42 million parameters and 6.08G FLOPs.
• Performance was comparable to other state-of-the-art lightweight methods.

Conclusions:

• The developed lightweight neural network offers a computationally efficient solution for AD detection using sMRI.
• The model demonstrates potential for intelligent AD detection on devices with limited computing capabilities.
• This approach facilitates rapid and accurate analysis of large sMRI datasets for timely diagnosis.