Dual Attention Multi-Instance Deep Learning for Alzheimer's Disease Diagnosis With Structural MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Structural magnetic resonance imaging (sMRI) is crucial for diagnosing neurological diseases by detecting brain variations.
• Identifying subtle, localized structural changes in sMRI is challenging but key for accurate disease diagnosis.
• Early detection of Alzheimer's disease (AD) and mild cognitive impairment (MCI) is vital for effective management.

Purpose of the Study:

• To develop an advanced deep learning model for enhanced feature identification in sMRI for early AD and MCI diagnosis.
• To improve the accuracy and generalizability of brain disease diagnosis using structural neuroimaging data.

Main Methods:

• Proposed a dual attention multi-instance deep learning network (DA-MIDL) incorporating Patch-Nets with spatial attention.
• Utilized an attention multi-instance learning (MIL) pooling for weighted global representation of brain structures.
• Implemented an attention-aware global classifier for integrated feature learning and classification.

Main Results:

• DA-MIDL effectively identified discriminative pathological locations in sMRI scans.
• The model demonstrated superior classification performance in accuracy and generalizability compared to existing methods.
• Evaluated on 1689 subjects across ADNI and AIBL datasets, confirming robustness.

Conclusions:

• DA-MIDL offers a promising approach for early and accurate diagnosis of Alzheimer's disease and mild cognitive impairment.
• The dual attention mechanism enhances the detection of subtle structural abnormalities in sMRI.
• This method advances the application of AI in neurodegenerative disease diagnostics.