A Lightweight ScaleDense-Transformer Framework with Auxiliary Quantum-Inspired Bottleneck Module for Whole-Lifespan...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Brain aging involves complex local and global changes.
• Accurate brain age prediction is crucial for monitoring neurological health.
• Existing models may lack biological plausibility or scalability.

Purpose of the Study:

• Introduce Lightweight Scale-Dense-Transformer (LST-Net) for whole-lifespan brain age prediction.
• Enhance prediction accuracy while ensuring structural and biological plausibility.
• Develop a scalable framework for large-scale brain health screenings.

Main Methods:

• Utilized a ScaleDense-Transformer architecture to capture local and global brain aging patterns.
• Incorporated a variational quantum circuit (VQC) for enhanced nonlinear latent representation.
• Evaluated the framework on a large cohort of 22,271 subjects across 17 data sources (0-96 years lifespan).

Main Results:

• Achieved a Mean Absolute Error (MAE) of 2.71 years in brain age prediction.
• Demonstrated stability across a wide lifespan and heterogeneous data sources.
• Confirmed high reproducibility with an Intraclass Correlation Coefficient (ICC) of 0.994 via longitudinal and test-retest analyses.

Conclusions:

• LST-Net provides a scalable and accurate method for brain age prediction.
• The framework supports structural and biological plausibility in modeling brain aging.
• LST-Net is a promising tool for widespread brain health monitoring and screening.