Knowledge Vault 3/54 - G.TEC BCI & Neurotechnology Spring School 2024 - Day 5
Sleep analyses when we go beyond EEG?
Samu Kainulainen, University of Eastern Finland (FI)
<Resume Image >

Concept Graph & Resume using Claude 3 Opus | Chat GPT4 | Llama 3:

graph LR classDef sleep fill:#f9d4d4, font-weight:bold, font-size:14px; classDef staging fill:#d4f9d4, font-weight:bold, font-size:14px; classDef ai fill:#d4d4f9, font-weight:bold, font-size:14px; classDef ppg fill:#f9f9d4, font-weight:bold, font-size:14px; classDef osa fill:#f9d4f9, font-weight:bold, font-size:14px; classDef insomnia fill:#d4f9f9, font-weight:bold, font-size:14px; classDef future fill:#f9d4d4, font-weight:bold, font-size:14px; A[Samu Kainulainen] --> B[Sleep divided into stages, cycles. 1] A --> C[Non-REM: N1-N4. REM: sympathetic overdrive. 2] B --> D[2024: N3, N4 combined. Agreement 70-80%. 3] A --> E[Samu Kainulainen: sleep researcher. 4] A --> F[Limitations: intervals, majority rule, interactions. 5] A --> G[Sleep disrupts scoring. OSA complicates. 6] A --> H[AI matches human accuracy. N1 problematic. 7] H --> I[Explainable AI highlights uncertainty. Trialed in lab. 8] A --> J[PPG classifies sleep, not as accurate. 9] J --> K[ANS changes not synced with EEG. 10] A --> L[OSA decreases agreement on stages, events. 11] L --> M[OSA decreases N3, fragments REM. Research lacking. 12] A --> N[Insomnia types: onset, maintenance, paradoxical. 13] N --> O[Objective, subjective quality differ. Combining may explain. 14] A --> P[Beyond EEG: wearables, biosignals, subjective experience. 15] A --> Q[Automatic staging needed to reduce workload. 16] A --> R[Phasic vs tonic REM, dementia links. 17] A --> S[Coma differs from sleep. Sleep dynamic. 18] A --> T[Body, brain likely interact during sleep. 19] A --> U[Adaptation to PSG setup takes nights. 20] A --> V[Sleep needed for growth, development. 21] A --> W[6+ hours recommended. Chronic deprivation detrimental. 22] A --> X[Classification system to improve with research. 23] class A,B,C,F,G,S,T,U,V,W sleep; class D,E,H,I,L,M,Q,X staging; class J,K ppg; class N,O insomnia; class P,R future;


1.- Sleep is divided into stages and cycles based on Registrar van Kael's 1968 paper using healthy volunteers. Why we sleep remains unanswered.

2.- Non-REM sleep is divided into stages 1-4 (from light to deep sleep). REM sleep has sympathetic overdrive. 4-5 cycles occur per night.

3.- In 2024, the only major sleep scoring change is combining N3 and N4 into one deep sleep stage (N3). Scoring agreement is ~70-80%.

4.- Samu Kainulainen is a sleep researcher at University of Eastern Finland and Kuopio University Hospital. His group was established in 2017.

5.- Limitations in sleep staging include using fixed 30-second intervals, the majority rule, losing interactions, and neglecting micro-features. More tools are needed.

6.- Sleep disrupts the scoring process. Severe OSA makes distinguishing sleep stages hard. Manual scoring takes 1-5 hours depending on the case.

7.- AI sleep staging tools match human accuracy. N1 is problematic to score. Combining N1 and N2 into light sleep is being discussed.

8.- Explainable AI highlights uncertain areas for human review, saving time. The system is being trialed in Samu's lab to set scoring thresholds.

9.- PPG signals can be used to classify sleep stages, though not as accurately as EEG. 3-stage (wake, NREM, REM) classification works well.

10.- Changes in autonomous nervous system activity are not synchronized with EEG sleep stages. PPG and EEG may quantify different phenomena.

11.- Severe sleep apnea decreases human scorer agreement on sleep stages, arousals, and respiratory events. AI accuracy also decreases due to uncertain training data.

12.- OSA decreases N3 and fragments REM. Literature is lacking on spatial EEG information and physiological interactions in OSA. More research is needed.

13.- Insomnia types include onset, maintenance, and paradoxical. PSG may show good sleep while the patient feels they barely slept. Brain region differences may explain this.

14.- Objective and subjective sleep quality often don't match. Combining them may help explain links between sleep disorders, psychiatric disorders, and neurological diseases.

15.- Going beyond EEG sleep analysis means transforming it to wearables, diving into biosignal properties and interactions, and incorporating subjective sleep experience.

16.- Automatic sleep staging is needed to reduce sleep clinic workload. PPG-based methods enable staging for home sleep tests lacking EEG.

17.- Phasic vs tonic REM distinction and links to outcomes like dementia risk is an area for further research involving human physiology experts.

18.- Coma differs from sleep - coma EEGs are flat without oscillations, making sleep staging impossible. Sleep is a more dynamic process.

19.- The body and brain likely interact during sleep in ways we can't yet measure well. Deeper brain regions may drive cortical EEG changes.

20.- It takes most people at least two nights to adapt to sleeping with a full PSG setup. Single-night PSGs have limitations.

21.- One reason we need sleep is for growth and development, but human sleep patterns differ from other mammals in still unexplained ways.

22.- 6+ hours of sleep per night is needed for most people. Chronically sleeping less is likely detrimental even if someone feels okay.

23.- The current sleep stage classification system is expected to improve substantially in the future as research advances.

Knowledge Vault built byDavid Vivancos 2024