IntelliSleepScorer

IntelliSleepScorer performs automated sleep stage scoring in mice using LightGBM-based machine learning on EEG and EMG recordings to classify NREM, REM, and wake states.

Key Features:

• Machine Learning Algorithm: Uses LightGBM, a gradient boosting framework, for sleep stage classification.
• High Accuracy and Reliability: Achieves overall accuracy of 95.2% and Cohen's kappa of 0.91, exceeding logistic regression (accuracy 93.3%, kappa 0.88) and random forest (accuracy 94.3%, kappa 0.89).
• Extensive Training Data: Models were developed on 5776 hours of EEG and EMG signals from 519 recordings across 124 mice.
• Generalizability (sampling frequency and epoch length): Validated on independent datasets with different sampling frequencies and epoch lengths, maintaining Cohen's kappa ≥ 0.80.
• Performance at low sampling frequencies: Retains high performance at lower sampling frequencies with kappa = 0.90.
• Robustness to light/dark cycle: Reported accuracy is unaffected by variations in the light/dark cycle.
• Single-electrode scoring: A modified LightGBM model scores sleep stages from one EEG and one EMG electrode with kappa ≥ 0.89.

Scientific Applications:

• Rodent sleep research: Automated scoring of NREM, REM, and wake states from mouse EEG and EMG supports studies of sleep physiology and neurological processes.
• Cross-condition and multi-dataset studies: High accuracy and validated generalizability enable application across diverse datasets, sampling settings, and experimental conditions.

Methodology:

LightGBM models were trained on 5776 hours of EEG and EMG signals from 519 recordings across 124 mice and validated across independent datasets, sampling frequencies, epoch lengths, and light/dark conditions.