Transformer-Based Multi-Channel K-Complex Detection Algorithm Tailored for Elderly Patients With Amnestic Mild...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Sleep Medicine

Background:

• K-complexes (KCs) are crucial for sleep maintenance and memory consolidation.
• KC characteristics change with aging and are altered in amnestic mild cognitive impairment (aMCI).
• Current automated KC detection methods struggle with variability in elderly aMCI patients.

Purpose of the Study:

• To develop and validate AdaPatchFormer, an automated, multi-channel, Transformer-based algorithm for KC detection.
• To optimize KC detection for elderly individuals with aMCI, addressing limitations of existing tools.

Main Methods:

• AdaPatchFormer integrates period embedding and a multi-granularity encoder for feature fusion.
• The model was trained on polysomnography (PSG) data from 268 elderly aMCI patients.
• Validation was performed on four independent datasets, including private and public cohorts.

Main Results:

• AdaPatchFormer outperformed existing automated detectors in recall, precision, accuracy, and MCC.
• The algorithm demonstrated a well-balanced recall-precision profile across diverse datasets.
• KC density and amplitude detected by AdaPatchFormer closely matched expert annotations.

Conclusions:

• AdaPatchFormer is a robust and accurate algorithm for KC detection in elderly individuals, particularly those with aMCI.
• The tool offers objective and potentially cost-effective support for early aMCI identification.
• This advancement has implications for real-world clinical settings and sleep research.