Aurora

Aurora is a Software as a Medical Device that leverages machine learning algorithms and advanced signal processing techniques to analyze photoplethysmogram (PPG) and oximetry data from compatible wearable sensors for the purpose of evaluating sleep quality in adult patients. It aims to assist healthcare professionals in diagnosing sleep disorders, such as sleep apnea, by automatically detecting respiratory events, determining sleep stages, and calculating comprehensive sleep metrics for clinical interpretation and management.

Aurora is a Software as a Medical Device (SaMD) designed to establish sleep quality by automatically analyzing photoplethysmogram (PPG) data, including SpO2 and pulse/heart rate, collected during sleep from compatible pulse oximeters. Intended for use by and under the order of healthcare professionals in adults, Aurora aids in the diagnosis of sleep disorders, particularly sleep apnea. The cloud-based system utilizes a sophisticated algorithm that includes verifying signal quality and study length, preprocessing data through normalization and artifact rejection, applying machine learning with multiple deep neural network models, and conducting statistical signal processing analyses. This process, guided by American Academy of Sleep Medicine (AASM) recommended scoring rules, generates outputs such as the Aurora Apnea-Hypopnea Index (eAHI), sleep stages (Wake, Light NREM, Deep NREM, REM), Total Sleep Time (TST), Sleep Efficiency (SE), Sleep Latency (SL), Wake After Sleep Onset (WASO), and Oxygen Desaturation Events Index (ODI), which clinicians can then review, edit, and use to inform clinical management. The performance of Aurora was validated using data from 158 adult patients, comparing its analysis of single-channel PPG and oximetry signals against manually scored polysomnography (PSG) and home sleep apnea test (HSAT) recordings. The primary metric, eAHI, when assessed at an AHI cut-off of ≥ 5 events/hour, demonstrated a sensitivity of 92.6% and specificity of 71.6% using a 3% desaturation criterion, and a sensitivity of 89.4% and specificity of 76.8% using a 4% desaturation criterion. Aurora also showed validated performance for sleep staging, with sensitivities and specificities for Wake (86.7%, 93.5%), Light Non-REM (80.9%, 85.5%), Deep Non-REM (63.4%, 95.9%), and REM sleep (83.6%, 97.5%) respectively. Further assessments through Deming regression and Bland-Altman analysis confirmed the performance of its derived sleep profile and oxygen saturation parameters against PSG-derived measures.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does Aurora's performance in detecting specific types of respiratory events (e.g., obstructive vs. central apneas, hypopneas) compare to polysomnography, beyond the overall eAHI?; What is the impact of varying photoplethysmogram (PPG) signal quality from different compatible oximeters on the accuracy of Aurora's sleep staging and eAHI calculation?; Could the machine learning models be further trained or adapted for specific patient subpopulations (e.g., elderly, patients with comorbid cardiovascular conditions) to improve diagnostic accuracy?; How does Aurora's algorithm differentiate between true sleep and periods of quiet wakefulness when analyzing single-channel PPG data for Total Sleep Time estimation?; What are the specific minimum data acquisition requirements (e.g., sampling rate, resolution) for PPG devices to ensure optimal performance with Aurora, and how sensitive is the algorithm to deviations from these requirements?