Alzheimer's disease detection using depthwise separable convolutional neural networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) diagnosis relies on neuroimaging, with deep learning (DL) showing promise.
• Current DL models for AD detection require extensive data and computational resources, limiting mobile integration.
• Existing research prioritizes classification accuracy over model efficiency and compactness.

Purpose of the Study:

• To develop a more efficient and compact deep learning model for Alzheimer's disease classification.
• To address the limitations of current DL algorithms in terms of data requirements and computational cost.
• To enable the integration of AD detection models into mobile embedded devices.

Main Methods:

• A deep separable convolutional neural network (DSC) model was proposed, utilizing depthwise separable convolution to replace conventional convolution.
• The proposed DSC model was evaluated on the OASIS magnetic resonance imaging dataset for AD detection.
• Transfer learning was implemented using pre-trained AlexNet and GoogLeNet models to enhance performance.

Main Results:

• The DSC model significantly reduced parameters and computational costs compared to traditional neural networks.
• The proposed model demonstrated high success rates in AD detection on the OASIS dataset.
• Transfer learning with AlexNet and GoogLeNet achieved average classification rates of 91.40% and 93.02%, respectively, with low power consumption.

Conclusions:

• The deep separable convolutional neural network is an efficient and effective model for Alzheimer's disease detection.
• The model's reduced complexity and low power consumption make it suitable for deployment on mobile devices.
• This approach advances the potential for accessible and widespread AD screening using AI and neuroimaging.