Developing a Treatment Clustering System for Obstructive Sleep Apnea Using Polysomnographic Physiological Signals
Developing a Treatment Clustering System for Obstructive Sleep Apnea Using Polysomnographic Physiological Signals
Obstructive sleep apnea syndrome (OSA) is marked by repeated upper airway obstructions during sleep, affecting approximately 14% of men and 5% of women aged 30-70 years.
However, precise clinical prediction tools for selecting optimal treatment strategies are lacking. This study aims to develop an automated treatment clustering system using artificial intelligence to classify patients based on etiology into (i) anatomical factors, (ii) reduced muscle responsiveness, and (iii) other non-anatomical factors. This system will analyze physiological sleep assessments, such as electromyography (EMG) and pneumotachographs, from a retrospective polysomnography (PSG) database. Cross-validation will be conducted on new OSA patients undergoing various management strategies, including surgical intervention, CPAP therapy, and oropharyngeal training (delivered face-to-face or via telerehabilitation). This system aims to enhance clinicians' ability to predict treatment success rates and improve patient outcomes.
Backgrounds:
Obstructive sleep apnea syndrome (OSA) is marked by repeated upper airway obstructions during sleep, affecting about 14% of men and 5% of women aged 30-70 years. The etiology of OSA is divided into anatomical and non-anatomical factors. Anatomical factors include upper airway narrowing or collapse, while non-anatomical factors encompass reduced muscle responsiveness, low arousal threshold, and high loop gain. Anatomical issues can be managed using surgical interventions or dental appliances. Non-anatomical issues like low arousal threshold and high loop gain may require pharmacological treatment or oxygen therapy. The genioglossus (GG) muscle's activity, crucial during sleep, is insufficient in about 30% of OSA patients. Regular oropharyngeal muscle exercises can reduce OSA severity and related symptoms.
However, precise clinical prediction tools for selecting optimal treatment strategies are lacking, and research on telerehabilitation for OSA patients is insufficient. This study aims to develop an automated treatment clustering system using artificial intelligence to classify patients based on etiology into: (i) anatomical factors, (ii) reduced muscle responsiveness, and (iii) other non-anatomical factors. This system will analyze physiological sleep assessments from a retrospective polysomnography (PSG) database. Cross-validation will be conducted on new OSA patients undergoing various management strategies, including surgical intervention, CPAP therapy, and oropharyngeal training (delivered face-to-face or via telerehabilitation).
Methods:
The automated treatment clustering system employs artificial intelligence to classify patients into etiological groups: (i) anatomical factors like upper airway narrowing or collapse; (ii) non-anatomical factors such as reduced muscle responsiveness; and (iii) other non-anatomical factors. The classification relies on analyzing multiple physiological sleep assessments, including electromyography (EMG) and pneumotachographs, from a retrospective PSG database. The system will undergo cross-validation with novel OSA patients, who will be screened based on inclusion and exclusion criteria and provide consent.
During the cross-validation phase, the OSA patients will undergo various assessments, including polysomnography, sleep-related questionnaire, drug-induced sleep endoscopy (DISE), computed tomography (CT) scans, functional magnetic resonance imaging (fMRI), tongue muscle strength and endurance tests, and mental state evaluations. Pre- and post-treatment measurements will be conducted. CT scans and DISE will assess anatomical structures before and after treatment, while fMRI will examine brain activation status. Muscle strength and endurance tests will evaluate the responsiveness level of tongue muscle before and after intervention.
The automated treatment clustering system, utilizing machine learning, will determine the phenotype of each case based on PSG, CT, sleep endoscopy, fMRI, and tongue strength and endurance results. These results will aid clinicians in categorizing patients and predicting treatment success rates. Treatment decisions will involve collaboration between physicians and patients, considering clinical expertise and patient preferences.
Participants classified as upper airway narrowing or collapse due to anatomical factors by the phenotyping system will be recommended for surgical management. For patients with reduced muscle responsiveness, a 12-week program of oropharyngeal muscle training is recommended. This training will be administered in two modes: face-to-face sessions and telerehabilitation. Each session will last 45-60 minutes, with participants attending face-to-face sessions in the lab or online classes (telerehabilitation) 1-3 days per week. Both groups will be instructed to perform additional oropharyngeal exercises at home. Patients not fitting these groups will use CPAP therapy, the gold standard for OSA management. During the treatment period, participants from all groups will have regular follow-ups to assess potential risks. Each group is expected to include 50 cases. After six months of treatment, the apnea-hypopnea index will be collected based on polysomnography to evaluate the success rates, comparing them to the predicted value analyzed using the phenotyping system.
Inclusion Criteria:
Exclusion Criteria:
junhui.ong611@gmail.com+886-9-37839992
chhung@mail.ncku.edu.tw+886-6-2353535 ext. 5939