Differentiating Dementia with Lewy Bodies and Alzheimer's Disease by Deep Learning to Structural MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Dementia with Lewy bodies (DLB) is a common neurodegenerative dementia, second only to Alzheimer's disease (AD).
• Current DLB diagnostic criteria rely on biomarkers, but structural MRI shows limited sensitivity and specificity for medial temporal lobe preservation.
• Existing MRI-based biomarkers for DLB diagnosis have reliability issues.

Purpose of the Study:

• To investigate the efficacy of a deep learning approach in differentiating DLB from AD using structural MRI data.
• To compare the diagnostic performance of deep learning with conventional statistical analysis for DLB vs. AD classification.

Main Methods:

• A retrospective study involving 208 patients (101 DLB, 69 AD, 38 controls).
• Gray matter images were extracted using voxel-based morphometry (VBM).
• A classification model using a ResNet-type convolutional neural network was developed to differentiate DLB from AD based on gray matter images.

Main Results:

• Conventional statistical analysis revealed only minor differences in the middle temporal pole and hippocampal regions between DLB and AD.
• The deep learning model achieved 79.15% accuracy in differentiating DLB from AD.
• Conventional methods achieved 68.41% accuracy, significantly lower than the deep learning approach.

Conclusions:

• Deep learning methods applied to gray matter images can identify subtle differences between DLB and AD.
• These deep learning techniques offer superior diagnostic performance compared to conventional methods for distinguishing DLB from AD.
• The findings suggest deep learning can overcome limitations of traditional analysis in detecting early or subtle neurodegenerative changes.