A Deep Spatiotemporal Attention Network for Mild Cognitive Impairment Identification

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Mild cognitive impairment (MCI) is a precursor to Alzheimer's disease (AD), but subtle structural brain changes make early diagnosis challenging.
• Current diagnostic methods for MCI lack sensitivity, necessitating advanced techniques for early detection and intervention.
• Functional brain network (FBN) analysis shows potential as a sensitive biomarker for neurological disease diagnosis.

Purpose of the Study:

• To develop a novel deep learning framework for accurate recognition of MCI based on brain functional networks.
• To improve early detection of MCI, thereby facilitating timely intervention for potential Alzheimer's disease (AD) progression.

Main Methods:

• Proposed a Deep Spatiotemporal Attention Network (DSTAN) framework for MCI recognition.
• Employed a spatiotemporal convolution strategy (ST-CONV) to extract features from brain functional signals and FBNs.
• Integrated an attention mechanism to identify key brain regions correlated with MCI and fused features for classification.

Main Results:

• The DSTAN framework achieved a classification accuracy of 84.21% in identifying MCI.
• The proposed method significantly outperformed existing baseline and state-of-the-art approaches.
• End-to-end training of the DSTAN network demonstrated its effectiveness in MCI recognition.

Conclusions:

• The DSTAN framework offers a powerful and accurate method for MCI recognition using FBN analysis.
• This approach holds significant promise for the early diagnosis of neurological diseases like Alzheimer's disease (AD).
• The study highlights the potential of deep learning and attention mechanisms in advancing neuroimaging-based diagnostics.