Supervised Computer-Aided Diagnosis (CAD) Methods for Classifying Alzheimer's Disease-Based Neurodegenerative...

Area of Science:

• Medical Imaging
• Machine Learning
• Neurology

Background:

• Alzheimer's disease (AD) diagnosis is crucial for timely treatment, but manual methods are time-consuming and expensive.
• Machine learning offers a potential solution to expedite and improve diagnostic processes.

Purpose of the Study:

• To develop and evaluate a machine learning transfer learning method for Alzheimer's disease diagnosis using 3D MRI data.
• To reduce the time and computational cost associated with AD diagnosis.

Main Methods:

• A transfer learning approach utilizing a M-Net migration network to extract bottleneck features from 3D MRI data.
• Supervised training with an added top layer for dimensionality reduction and classification.
• Combining subject slice properties for final classification of AD symptoms and controls.

Main Results:

• The proposed transfer network demonstrated improved computational efficiency and reduced training time.
• Achieved 1.5 percentage points higher classification accuracy compared to using VGG16 alone for feature extraction.
• Showcased an 8% improvement in classification accuracy and a 60-fold reduction in training time compared to typical transfer learning networks, using OASIS dataset.

Conclusions:

• The developed transfer learning method offers significant advantages in terms of speed and accuracy for Alzheimer's disease diagnosis.
• This approach can effectively aid in the early and efficient diagnosis of Alzheimer's disease.
• The method shows promise for clinical application in reducing diagnostic burden.