Chronic Insomnia, Epilepsy, Sleep Disorders
Conditions
Keywords
Sleep staging, Polysomnography, EEG, Deep learning, Artificial intelligence, Sleep disorders, Chronic insomnia, Epilepsy
Brief summary
The objective of this study is to develop and validate deep learning algorithms for automated sleep stage and sub-stage classification using overnight polysomnography data. The models will be trained and evaluated on at least three independent datasets to ensure generalizability. \- Primary Outcome Measure : Accuracy of deep learning-based sleep stage classification compared to expert manual scoring (\>80% target agreement), evaluated across multiple polysomnography datasets including AP-HP (Assistance Publique - Hôpitaux de Paris) data. This is a retrospective, observational study.
Interventions
None listed
Sponsors
Assistance Publique - Hôpitaux de Paris
Idiap Research Institute, Switzerland
Study design
Observational model
COHORT
Time perspective
RETROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to 65 Years
Healthy volunteers
No
Inclusion criteria
* Patients with chronic insomnia and/or epilepsy who underwent polysomnography in a neurophysiology or neurology setting under the responsibility of Pr Navarro between 01 September 2011 and 31 December 2024. * Age ≥18 and ≤65 years at the time of the polysomnography recording.
Exclusion criteria
* Severe psychiatric disorder, including decompensated psychotic disorder, manic episode, or major depressive episode with melancholic features. * Use of continuous positive airway pressure (CPAP) therapy during the night of recording. * Patient refusal or documented opposition to data use.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Prediction accuracy of sleep stages and sub-stages | Single overnight polysomnography recording per participant (duration of approximately 8 to 12 hours) | Evaluation of the deep learning model's performance in accurately classifying different sleep stages and sub-stages compared to expert manual scoring. The metrics used to characterize this outcome are the macro F1-score and/or Cohen's Kappa (κ) score, with a target prediction accuracy of \>80%. The macro F1-score measures the model's ability to correctly recognize each sleep stage while compensating for the imbalance between frequent and rare classes. Cohen's Kappa quantifies the degree of agreement between automatic predictions and human annotations by correcting for the agreement expected by chance. The combination of these two metrics offers a robust and balanced evaluation. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Prediction accuracy of chronic insomnia profiles | Single overnight polysomnography recording per participant (duration of approximately 8 to 12 hours) | Evaluation of the deep learning algorithms' accuracy in identifying and predicting chronic insomnia profiles based on the electroencephalographic (EEG) analysis of polysomnographies. Performance will be assessed by comparing the automated predictions against established clinical diagnoses using standard machine learning classification metrics (such as macro F1-score and Cohen's Kappa). |
| Prediction accuracy of epilepsy profiles | Single overnight polysomnography recording per participant (duration of approximately 8 to 12 hours) | Evaluation of the deep learning algorithms' accuracy in identifying and predicting different epilepsy profiles based on the electroencephalographic (EEG) analysis of polysomnographies. Performance will be assessed by comparing the automated predictions against established clinical diagnoses using standard machine learning classification metrics (such as macro F1-score and Cohen's Kappa). |
Contacts
CONTACTVincent Navarro, MD, PhD
CONTACTJinmi BAEK
STUDY_DIRECTOROlivier Pallanca, MD, PhD
Idiap Research Institute, Switzerland
Outcome results
None listed