Dynamic Image for 3D MRI Image Alzheimer's Disease Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Alzheimer's disease (AD) diagnosis relies on neuroimaging, with MRI being crucial.
• 3D Convolutional Neural Networks (CNNs) show promise for AD classification from MRI but are computationally intensive.
• Efficient and accurate diagnostic tools are needed to combat the growing prevalence of Alzheimer's disease.

Purpose of the Study:

• To develop an efficient 2D CNN model for Alzheimer's disease classification from 3D MRI data.
• To reduce the computational cost and training time associated with 3D CNNs for neuroimaging analysis.
• To improve the accuracy of automated Alzheimer's disease detection using a novel image transformation technique.

Main Methods:

• Utilized approximate rank pooling to convert 3D MRI volumes into 2D images.
• Applied a 2D CNN architecture to the transformed 2D images for Alzheimer's disease classification.
• Compared the performance and training efficiency against baseline 3D CNN models.

Main Results:

• The proposed 2D CNN model achieved 9.5% higher accuracy in Alzheimer's disease classification compared to 3D models.
• The training time for the 2D CNN approach was reduced to only 20% of that required for 3D CNN models.
• Demonstrated the feasibility of using a simplified input representation for effective deep learning in neuroimaging.

Conclusions:

• The 2D CNN approach using approximate rank pooling offers a computationally efficient and accurate method for Alzheimer's disease classification from 3D MRI.
• This technique presents a viable alternative to resource-intensive 3D CNNs, facilitating wider adoption in clinical research.
• The developed model shows significant potential for improving early and accurate diagnosis of Alzheimer's disease.