Validation of a Fingertip Home Sleep Apnea Testing System Using Deep Learning AI and a Temporal Event Localization Analysis.

Study objectives: This paper validates TipTraQ, a compact home sleep apnea testing (HSAT) system. TipTraQ comprises a fingertip-worn device, a mobile application, and a cloud-based deep learning artificial intelligence (AI) system. The device utilizes PPG (red, infrared, and green channels) and accelerometer sensors to assess sleep apnea by the AI system.

Methods: We prospectively enrolled 240 participants suspected of obstructive sleep apnea (OSA) at a tertiary medical center for internal validation and 112 participants independently at another center for external validation. All participants underwent simultaneous polysomnography (PSG) and TripTraQ HSAT. We compared TipTraQ-derived total sleep time (TQ-TST) and TipTraQ-derived Respiratory Events Index (TQ-REI) with expert-determined total sleep time (TST) and apnea-hypopnea index (AHI), based on AASM standards with the 1B hypopnea rule. Temporal event localization analysis for respiratory event prediction was conducted at both event and hourly levels.

Results: In the external validation, the Spearman correlation coefficients for TQ-TST vs. TST and TQ-REI vs. AHI were 0.81 and 0.95. respectively. The root mean square error were 0.53 hours for TQ-TST vs. TST and 7.53 events/hour for TQ-REI vs. AHI. For apnea/hypopnea prediction with a 10s grace period, the true positive, false positive and false negative rates in temporal event localization analysis were 0.76, 0.24, and 0.23, respectively. The four-way OSA severity classification achieved a Cohen's kappa of 0.7.

Conclusions: TQ-TST and TQ-REI predict TST and AHI with comparable performance to existing devices of the same type, and respiratory event prediction is validated through temporal event localization analysis.