Validación de algoritmos de aprendizaje profundo Belun Ring Gen3 con sensor corporal subxifoideo

Potential participants will be identified from patients scheduled for in-lab PSG at the two medical centers of University Hospitals

The Belun Ring sensor should be placed on the palmar side of the proximal phalanx of the index finger and the sensor should be placed along the radial artery such that the accuracy of the device will be minimally affected by skin color. The Ring has 7 adjustable arms for different finger sizes. Each device is reusable after thorough cleaning with an alcohol swab.

The "Belun Cor" body sensor accessory is composed of an accelerometer, a temperature sensor, and a lithium battery. It will be placed immediately below the xiphoid process in the upper abdomen with a medical adhesive to detect the body temperature, body posture, respiratory rate, and respiratory efforts.

To establish mean, standard deviation, and target clinical agreement limit values for the differences in sleep-disordered breathing parameters (including AHI3%, AHI4%, REM AHI 3%, REM AHI4%, Supine AHI3%, Supine AHI4%, and CAI) from the BR against PSG. To calculate the sensitivity and specificity values, along with 95% confidence intervals (CIs), for sleep-disordered breathing parameters from the BR against PSG at PSG cutoffs of 5, 15, and 30 events/h. Pearson correlation and regression analysis will be employed to assess the association between sleep-disordered breathing parameters obtained from the BR and PSG.

Sensitivity and specificity of the Epoch-by-Epoch (EBE) 4-stage classification (wake, deep sleep, light sleep, REM) of the BR will be compared to EBE PSG data for each stage against the combination of the other three classifications. For instance, sensitivity for detecting wake will measure the proportion of correctly classified wake epochs against the combined classifications of deep sleep, light sleep, and REM sleep. Sensitivity and specificity values, accompanied by 95% CIs, of EBE body position (supine vs. non-supine) against PSG will also be computed.

To explore the minimum number of hours needed by the BR for an accurate OSA diagnosis, sensitivity, specificity, and Matthews Correlation Coefficient (MCC) values, along with 95% CIs, will be computed for sleep-disordered breathing parameters of Belun total record time (bTRT) and belun total sleep time (bTST) with different numbers of hours (e.g., 2, 3, 4, 5, 6, 7, and 8 hours) against PSG data. In cases where varying numbers of subjects meet different bTRT and bTST criteria at different durations, resampling methods like oversampling, undersampling, bootstrapping, etc., will be employed to address data imbalances.

To calculate the performance metrics, including sensitivity and specificity, along with 95% CIs, for the detection of arrhythmias by BR device in comparison to assessments made by a board-certified sleep physician. The physician will review the PSG ECG signals of the subjects suspected of having significant arrhythmias during the sleep study.

Pearson correlation and regression analysis will be employed: (1) to explore the relationships between autonomic nervous system (ANS) metrics and delta HR, hypoxic burden, and pulse wave amplitude drop index (PWADI); (2) to evaluate the association between HRV, Delta HR, hypoxic burden, PWADI, CPAP adherence, sleep-disordered breathing parameters, and demographic data (including age, gender, and BMI). If data distribution assumptions are violated, the Spearman correlation will be used instead of the Pearson correlation.

The investigators will compute sensitivity and specificity values, accompanied by their respective 95% CIs, for AHI3% and AHI4% from the NTUH/NEOMED models in comparison to PSG. Furthermore, the investigators will determine the ROC curve at PSG-defined cutoffs of 5, 15, and 30 events/h for both AHI3% and AHI4%.