MPC-STANet: Alzheimer's Disease Recognition Method Based on Multiple Phantom Convolution and Spatial Transformation...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder requiring early detection for effective intervention.
• Accurate staging of AD is challenging due to subtle feature changes and data imbalance in MRI datasets.
• Existing methods struggle with recognizing scattered and small-scale pathological features in AD.

Purpose of the Study:

• To develop an advanced AI model for precise recognition of Alzheimer's disease stages using MRI data.
• To address data imbalance and enhance the detection of subtle, localized AD-related features.
• To improve the accuracy and efficiency of AD diagnosis through novel deep learning techniques.

Main Methods:

• Class-balancing operations including data expansion and SMOTE were applied to address dataset imbalance.
• A novel recognition network, MPC-STANet, utilizing Multi-Phantom Convolution (MPC) and Space Conversion Attention Mechanism (SCAM) with ResNet50 backbone was proposed.
• Multi-Phantom Residual Blocks (MPRB) were integrated into ResNet50's Conv and Identity blocks to enhance feature extraction for subtle AD indicators.

Main Results:

• The proposed MPC-STANet achieved an average recognition accuracy of 96.25%.
• The model demonstrated strong performance with an F1 score of 95% and mAP of 93%.
• The network's parameter count increased minimally (1.69 M) compared to the standard ResNet50.

Conclusions:

• The MPC-STANet model shows significant potential for accurate and reliable Alzheimer's disease stage recognition.
• The integration of MPC and SCAM effectively captures subtle structural changes crucial for AD diagnosis.
• This approach offers a promising tool for early AD detection, potentially aiding in timely treatment and disease management.