A comparative analysis of automatic and manual scoring methods in polysomnography

The objective of this study was to compare twenty-six polysomnography (PSG) parameters between the groups utilizing automatic scoring (AS) software and manual scoring (MS) technique. Two MS groups, each comprising technicians with sleep-scoring experience ranging from 5 to 8 years, were established. One group consisted of three technicians while the other included four technicians. PSG record of 369 was initially scored by the first MS group and subsequently rescored by the second MS group. Finally, the PSG records were automatically scored once using the Embla RemLogic version 3.4 PSG Software (Embla Systems, Natus Medical Incorporated). All PSG records were evaluated three times independently and compared across the three groups: AS, MS group 1, and MS group 2. No significant differences were identified among the MS groups for any of the PSG parameters. AS differs from the two MS groups in many PSG parameters, including sleep stages and the apnea-hypopnea index (p < 0.05). A strong agreement was noted between MS group 1 and MS group 2 for almost all variables (p < 0.001, Kendall's tau-B (τb) = 0.747-0.922). The AS had a poor correlation with both MS groups regarding total apnea and hypopnea counts. As a result, AS methods yield imprecise results for numerous PSG parameters and should not be regarded as the exclusive scoring approach until more sophisticated artificial intelligence algorithms are developed. PSG scoring conducted by a technician with a minimum of five years of experience is both sufficient and dependable.