Structural and Statistical Knowledge-Enhanced Attention Network for early Parkinson's disease diagnosis

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurological Disorders

Background:

• Early diagnosis of Parkinson's disease (PD) is critical but challenging.
• Current deep learning (DL) models for PD diagnosis lack domain-specific knowledge integration.
• This limits diagnostic accuracy and clinical interpretability.

Purpose of the Study:

• To develop a specialized DL framework for early PD detection.
• To integrate neuroimaging domain knowledge for enhanced diagnostic performance.
• To improve the clinical interpretability of PD diagnostic models.

Main Methods:

• Proposed the Structural and Statistical Knowledge-Enhanced Attention Network (SSKEA-Net).
• Incorporated Gray-White Interactive Modulation (GWIM) and Statistical Prior-Guided Attention (SPGA) modules.
• Evaluated using diffusion tensor imaging on a matched early-stage PD dataset.

Main Results:

• SSKEA-Net achieved high diagnostic accuracy (ACC 0.8798±0.0158).
• Outperformed existing DL and neuroimaging models.
• Activation heatmaps showed precise localization in clinically relevant brain regions.

Conclusions:

• SSKEA-Net effectively combines domain knowledge with DL for accurate and interpretable early PD detection.
• The framework offers a valuable approach for clinical neuroimaging AI.
• This enhances the reliability of AI systems in diagnosing neurological diseases.