Improved Sleep Spindle Detection Using the BOSC Method: A Comparison with Traditional Approaches

Abstract

Sleep spindles are brief bursts of 6-20 Hz local field potential (LFP/EEG) activity that occur during non-rapid eye movement sleep. Traditional spindle detection methods rely on manually set amplitude and duration thresholds, but this approach can be vulnerable to false detections and could miss low-amplitude spindles due to the inflexible nature of the thresholding technique. The Better OSCillation (BOSC) detection method offers a more robust alternative by applying frequency-specific power thresholds calibrated to the signal itself and requiring a minimum number of oscillation cycles. In this study, we compared traditional and BOSC methods for spindle detection in a variety of ways. First, we created two synthetic datasets: one with synthetic spindles modelled on previous data and one with single wave pulses that had a period consistent with the spindle frequency band. These datasets were used to demonstrate each methods performance when given events that should be detected (synthetic spindles) and those that should not (single wave pulses). Second, we analyzed cortical local field potentials from recordings of rats during natural sleep using both methods to determine their relative effectiveness given biological LFP data. BOSC consistently outperformed the traditional approach in all situations, identifying more valid spindles while minimizing false (or likely false) detections. These findings validate BOSC as a superior method of spindle detection due to its better calibration to the actual signal.