Classification of EEG and Sleep Stages Criteria

AI LAB- STUDY_ Sleep Scoring Modeling

• Infrared sleep video and voice database construction project for disease diagnosis using artificial intelligence technology
• [Seoul National University Hospital participates] in research activities to enhance artificial intelligence technology for sleep and speech disorders through the construction of infrared sleep video data and voice data, under the supervision of the institution.

Dataset : infrared sleep video data with over 5000 hours, anonymized and standardized after completion of the necessary tasks.

Infrared Sleep Video Data:

• Constructing sleep polygraph data correlated with infrared sleep video data.
• Developing disease prediction models and algorithms based on this data.
• Establishing a service model based on artificial intelligence models.

Speech Disorder Discrimination Using Voice Data:

• Constructing learning datasets for infrared sleep videos and voice data to lead the development of private and public artificial intelligence technologies.
• Establishing pilot services for infrared sleep video and voice data.

Introduction:

The interpretation of polysomnography (PSG) results follows the guidelines outlined in The AASM Manual for the Scoring of Sleep and Associated Events. Sleep stages, arousals, and associated events are labeled according to the rules defined in the manual. The PSG data, including EEG, EOG, EMG, and respiratory signals, are analyzed in 30-second epochs, and specific criteria are applied for labeling sleep stages and events.

Sleep Stages:

a. Stage W (Wakefulness):

• Characterized by more than 50% alpha rhythm before sleep onset.
• Occasional eye movements with irregular patterns and increased tension in jaw muscles.

b. Stage N1 (NREM 1 - Shallow Sleep):

• No alpha waves; lower amplitude and vertex sharp waves in more than 50% of epochs.
• Slow waves (4-7Hz) higher than in Stage W, dominant in posterior head regions, especially in children.

c. Stage N2 (NREM 2 - Shallow Sleep):

• Presence of sleep spindles and K complexes defines N2 sleep.
• If sleep spindles or K complexes are absent, EEG activity continues.
• Transitions out of N2 occur when switching to Stage W, transitioning to N1 without sleep spindles or K complexes, or transitioning to slow-wave sleep (N3) or REM sleep.

d. Stage N3 (NREM 3 - Deep Sleep):

• Absence of sleep spindles and eye movements.
• Lower chin muscle tone compared to N2; occasionally slower than REM.

e. Stage R (REM - Dream Sleep):

• Low amplitude and mixed-frequency EEG, low chin muscle tone, and rapid eye movements characterize REM sleep.
• Scoring Periodic Limb Movements (PLM) and PLM-Associated Arousals:
• LM with a duration of 0.5-10 seconds and starting from resting EMG above 8uV, ending when below 2uV.
• PLM defined as at least 4 LMs with intervals of 5-90 seconds; if multiple LMs occur within 5 seconds, they are considered a single LM.
• PLM-associated arousals occur within 0.5 seconds of LM, with specific EEG criteria and increased chin EMG during REM sleep.

Sleep Scoring Principles:

• If arousals are observed in N2, N3, or R, classify the epoch as N1.
• Increased chin muscle activity during R indicates N1.
• If N3 criteria are not met in the following epoch after N3, reclassify as N2.
• If even partial alpha rhythm is observed during major body movements, classify as W.

Conclusion:

• The PSG interpretation adheres to the principles outlined in The AASM Manual for the Scoring of Sleep and Associated Events. Proper scoring of sleep stages, arousals, and associated events is crucial for accurate diagnosis and treatment planning.
• Establishing a system that enables easy access to ethical data necessary for AI research and development related to sleep and speech for private companies, public institutions, and research centers having difficulty accessing medical data:
• Utilizing trained data to conduct trial learning according to data guidelines and opening a system to save models, thereby enhancing the usability of the data.

Reference: https://github.com/rootofdata/AILAB_Sleep-Scoring-Modeling?tab=readme-ov-file

GitHub - rootofdata/AILAB_Sleep-Scoring-Modeling: Infrared sleep video and voice database construction project for disease diagn