Beyond the Ground Truth, XGBoost Model Applied to Sleep Spindle Event Detection

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Sleep spindles are crucial EEG microevents with unclear functions.
• Automatic detection of sleep spindles is vital for research reproducibility.
• Current deep learning methods lack interpretability.

Purpose of the Study:

• To develop an interpretable machine learning framework for automatic sleep spindle detection.
• To introduce a novel, symmetric evaluation metric for spindle detection.
• To propose a new performance assessment method for evaluating generalization and inter-expert agreement.

Main Methods:

• Developed SpinCo, a framework using sliding window feature extraction and XGBoost.
• Implemented a novel by-event evaluation metric for symmetric and probabilistic results.
• Designed a performance assessment test for generalization to unseen experts.

Main Results:

• SpinCo achieved performance close to state-of-the-art deep learning techniques.
• The novel metric enhanced evaluation interpretability and allowed direct assessment of inter-expert agreement.
• The proposed assessment test evaluated the method's generalization capabilities.

Conclusions:

• SpinCo is a robust and interpretable automatic sleep spindle detection technique.
• The new evaluation metric improves the understanding of spindle detection performance.
• This work provides a valuable tool for EEG signal labeling and analysis.