Accurate scoring of the apnea-hypopnea index using a simple non-contact breathing sensor

Abstract

Sleep apnea is a serious condition that afflicts many individuals and is associated with serious health complications. Polysomnography, the gold standard for assessing and diagnosing sleep apnea, uses breathing sensors that are intrusive and can disrupt the patient's sleep during the overnight testing. We investigated the use of breathing signals derived from non-contact force sensors (i.e. load cells) placed under the supports of the bed as an alternative to traditional polysomnography breathing sensors (e.g. nasal pressure, oral-nasal thermistor, chest belt and abdominal belt). The apnea-hypopnea index estimated using the load cells was not different than that estimated using standard polysomnography leads (t44 = 0.37, P = 0.71). Overnight polysomnography sleep studies scored using load cell breathing signals had an intra-class correlation coefficient of 0.97 for the apnea-hypopnea index and an intra-class correlation coefficient of 0.85 for the respiratory disturbance index when compared with scoring using traditional polysomnography breathing sensors following American Academy of Sleep Medicine guidelines. These results demonstrate the feasibility of using unobtrusive load cells installed under the bed to measure the apnea-hypopnea index.

Figure 1

(A–B) Illustration of how the load cells detect breathing via small mass (M) displacements. As an individual lies on the bed, the load cells detect the forces supported by each bed leg. (A) During inspiration mass is displaced towards the foot of the bed. (B) During expiration mass is displaced towards the head of the bed. (C) An example of a load cell breathing signal from a patient is shown. Periods of inspiration are marked in red and periods of expiration are marked in black.

Figure 2

Screen shots comparing the scoring montages used for scoring with typical PSG signals (upper) and with load cell breathing signals (lower). The screen shots were taken from the same 120 seconds for both scoring results from one patient. The nasal pressure, oral-nasal thermistor, chest belt, and abdominal belt are colored purple in the PSG scoring montage (upper), and the load cell breathing signals are similarly colored purple in the load cell scoring montage (lower). The load cell tracing “All_Sum_HP” is the summation of all the load cells, and the “COP_Y_HP” is the center of pressure load cell signal. The purple, horizontal boxes in both cases indicate the locations of scored respiratory events.

Figure 3

Segments of the load cell breathing signal from a single patient illustrating the scoring of respiratory events using the load cell trace. (Upper) A scored hypopnea showing a slight reduction in the excursion of the load cell signal. (Middle) A scored obstructive apnea showing a major reduction in the excursion of the load cells signal. However, some LC excursion appears to still be present suggesting breathing effort may still exist. (Lower) A scored central apnea showing a complete absence of excursion in the load cell signal.

Figure 4

Linear least squares regression plots for AHI-LC vs. AHI-PSG and RDI-LC vs. RDI-PSG.

Figure 5

Bland-Altman plots for visualization of the agreement between the PSG and load cell scoring of AHI and RDI.