Convolutional Neural Networks to Classify Alzheimer's Disease Severity Based on SPECT Images: A Comparative Study

Area of Science:

• Medical imaging analysis
• Artificial intelligence in neuroscience
• Neurodegenerative disease research

Background:

• Alzheimer's disease (AD) progression research relies on neuroimaging, but single-photon emission computed tomography (SPECT) data is scarce.
• Medical image analysis necessitates extensive labeled datasets, posing a challenge for AD studies.
• Developing robust classification models is crucial for epidemiological research in AD.

Purpose of the Study:

• To compare the detection performance of five distinct convolutional neural network (CNN) models for classifying AD.
• To evaluate the effectiveness of transfer learning in medical image analysis for AD detection.
• To establish a CNN-based classification model for epidemiological studies using SPECT images.

Main Methods:

• Collected brain SPECT scan data from 99 subjects, totaling 4711 images.
• Compared five CNN models: MobileNet V2, NASNetMobile, VGG16, Inception V3, and ResNet.
• Evaluated model performance using accuracy and loss functions, and analyzed demographic and cognitive data (CDR, CASI, MMSE).

Main Results:

• CNN models demonstrated varying classification performance on medical images.
• Transfer learning proved effective in distinguishing between healthy controls and different stages of cognitive impairment and AD.
• Cognitive scores significantly correlated with Clinical Dementia Rating (CDR) classifications.

Conclusions:

• CNNs, particularly with transfer learning, are effective tools for classifying Alzheimer's disease severity using SPECT imaging.
• This approach can aid in epidemiological research by overcoming limitations of small datasets.
• The study highlights the potential of AI in advancing neuroimaging analysis for AD.