Convolutional Recurrent Neural Network for Dynamic Functional MRI Analysis and Brain Disease Identification.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) provides insights into brain dynamics through dynamic functional connectivity (dFC) networks.
• dFC network analysis aids in the automated identification of neurological conditions, including Alzheimer's disease (AD) and its early stages.
• While deep learning methods show promise in dFC analysis, they often overlook the sequential nature of these networks, limiting diagnostic performance.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach, a convolutional recurrent neural network (CRNN), for automated brain disease classification using rs-fMRI data.
• To enhance diagnostic accuracy by incorporating sequential information from dFC networks, which is often neglected in existing methods.

Main Methods:

• Dynamic functional connectivity (dFC) networks were constructed from rs-fMRI data using a sliding window approach.
• A CRNN architecture, comprising convolutional layers for feature extraction and a long short-term memory (LSTM) layer for preserving sequential information, was employed.
• The model utilized fully connected layers for the final classification of brain diseases.

Main Results:

• The proposed CRNN method demonstrated effectiveness in both binary and multi-category classification tasks for brain diseases.
• Experimental validation was conducted on a dataset of 174 subjects and 563 rs-fMRI scans from the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Conclusions:

• The CRNN model successfully integrates dFC network analysis with sequential feature learning for improved brain disease classification.
• The findings highlight the potential of leveraging sequential information in dFC networks with advanced deep learning techniques for more accurate neurological disorder diagnosis.