Fine-Grained and Multiple Classification for Alzheimer's Disease With Wavelet Convolution Unit Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Alzheimer's disease (AD) classification requires accurate feature extraction.
• Existing methods struggle with fine-grained, multi-class AD classification.
• Integrating wavelet analysis with neural networks can improve feature representation.

Purpose of the Study:

• To propose a novel wavelet convolution unit for image-oriented neural networks.
• To develop a network for fine-grained Alzheimer's disease classification.
• To achieve state-of-the-art accuracy in multi-class AD classification.

Main Methods:

• A novel wavelet convolution unit combining traditional convolution with wavelet decomposition (single and multi-scale).
• Fusion of cross-scale features to enhance localization of singular points.
• Development of a neural network utilizing the wavelet convolution unit for AD classification on diffuse tensor images.

Main Results:

• The proposed network achieved state-of-the-art accuracy for all eight fine-grained classifications, reaching up to 97.30%.
• The method demonstrated high accuracy across twelve coarse-grained and fine-grained AD classifications.
• The classification performance significantly surpassed existing Alzheimer's disease classification methods.

Conclusions:

• The novel wavelet convolution unit effectively extracts deep abstract features for improved image analysis.
• The developed network provides a robust and highly accurate solution for fine-grained Alzheimer's disease classification.
• This approach sets a new benchmark for Alzheimer's disease classification accuracy and capability.