Predicting Conversion from Mild Cognitive Impairment to Alzheimer's Disease Using a Vision Transformer and...

Area of Science:

• Artificial Intelligence
• Neuroimaging
• Alzheimer's Disease Research

Background:

• Convolutional Neural Networks (CNNs) dominate computer vision, including Alzheimer's disease (AD) applications.
• Vision Transformers (ViTs) present a novel alternative to CNNs in medical image analysis.
• Mild Cognitive Impairment (MCI) is a precursor to AD, but not all MCI patients convert.

Purpose of the Study:

• To evaluate the efficacy of Vision Transformers (ViTs) in classifying individuals with MCI who convert to AD versus those who do not.
• To explore the potential of ViTs as a reliable tool for early AD detection.

Main Methods:

• A transfer learning approach was employed, fine-tuning a pretrained ViT model on the ADNI dataset.
• Axial T1-weighted MRI slices of the hippocampal region from 575 individuals (299 stable MCI, 276 progressive MCI) were utilized.
• The model was trained to differentiate between MCI converters and non-converters.

Main Results:

• The ViT model achieved an average Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.74 ± 0.02.
• Classification accuracy was 0.69 ± 0.03, with a specificity of 0.72 ± 0.06 and sensitivity of 0.65 ± 0.07.
• The F1-score for the progressive MCI class was 0.67 ± 0.04.

Conclusions:

• The ViT approach demonstrates reasonable accuracy in classifying MCI converters versus non-converters.
• These findings suggest ViTs are a viable tool for predicting AD progression from MCI.
• Further research is needed to validate the generalizability and clinical utility of ViTs for AD diagnosis.