Demonstrates an accurate multimodal classification of Alzheimer's disease, highlighting the integration of imaging and clinical data.
Key Points
The aim is to enhance the classification accuracy of Alzheimer's disease by integrating fMRI data with clinical tests using an explainable deep learning model.
Utilized a 3D convolutional neural network model for data analysis.
Supplemented fMRI data with five clinical tests to address small dataset limitations.
Applied leave-one-out cross-validation to prevent data leakage.
Employed perturbation ranking to identify feature importance in classifications.
Achieved 90% accuracy in classifying Alzheimer's disease versus controls.
Outperformed a model using only fMRI data, which achieved 58% accuracy.
Clinical tests showed varying importance depending on the diagnostic group, especially MoCA for controls.