Here is How AI Detects Sleep Apnea From Sound Data

Here is How AI Detects Sleep Apnea From Sound Data

Healthcare Tech Outlook | Monday, September 16, 2019

Tracheal sounds are capable of detecting sleep disorders precisely by differentiating between sleep and waking states.

FREMONT, CA: Sleep apnea includes daytime sleepiness, loud snoring, and restless sleep. The individuals’ breathing repeatedly stops and starts during sleep in a common disorder. Tracheal sounds are one of the best ways to estimate respiratory flow, which is why it is a reliable way to detect obstructive and central apnea.

There are several ways to detect sleep disorders like overnight monitoring at a sleep center of breath and other body functions, nocturnal polysomnography, Home sleep tests, and AI analysis of spectrograms.

As a substitute for polysomnography (PSG), various simplified methods exist. Such uncomplicated methods are widely used as home sleep apnea test (HSAT) and are officially accepted diagnostic tools. However, HSAT devices often have limited diagnostic abilities; these devices cannot discriminate the wake/sleep status easily. Therefore, they use the total recording time, instead of the whole sleep time, as the denominator of the respiratory event index, which results in an underestimation of OSA severity.

The latest AI-enabled automated methods detect sleep apnea using artificial intelligence algorithms, which are more convenient and comfortable for patients. In AI analysis of spectrograms, the sound is converted into images after recording patients' breathing patterns as they sleep. Japanese researchers have come up with this working method as reported by the American Academy of Sleep Medicine's Journal of Clinical Sleep Medicine.

For the development of neural network structure, TS spectrogram images were constructed related to several thousand typical breathing events that were sampled as pilot data. The breathing process is categorized into eight patterns: normal breathing, snoring, snoring with hypopnea, obstructive apnea, central apnea, body movement, vocalization, and irregular breathing. These images were discriminated using convolutional and recurrent DNNs. After that, a DNN structure is selected to facilitate the subsequent analysis.

Many portable devices for testing sleep apnea can fail to distinguish between sleep and waking states. Tracheal sounds, which are pictured with a spectrogram, carry information about sleep apnea and sleep/wake status. A deep neural network with convolutional layers can discriminate breathing status and sleep/wake status with accuracy.

The researchers used tracheal spectrograms obtained every 60 seconds from more than 1,500 patients to test their AI-based experimental technique.

Many studies were conducted to analyze the tracheal sound signals in the time domain and to investigate how the tracheal intensity signal changes with time. Later studies explored frequency analysis of tracheal sound signals to examine how much of the signal lies within each given frequency band over a range of frequencies. These studies have proved that tracheal sounds that are recorded with appropriate sensors can pick up breathing, snoring, and intra-thoracic pressure variations.

Check out: Top Sleep Disorder Care Solution Companies

Weekly Brief