1
|
Automatic sleep scoring with LSTM networks: impact of time granularity and input signals. BIOMED ENG-BIOMED TE 2022; 67:267-281. [DOI: 10.1515/bmt-2021-0408] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/08/2021] [Accepted: 05/17/2022] [Indexed: 11/15/2022]
Abstract
Abstract
Supervised automatic sleep scoring algorithms are usually trained using sleep stage labels manually annotated on 30 s epochs of PSG data. In this study, we investigate the impact of using shorter epochs with various PSG input signals for training and testing a Long Short Term Memory (LSTM) neural network. An LSTM model is evaluated on the provided 30 s epoch sleep stage labels from a publicly available dataset, as well as on 10 s subdivisions. Additionally, three independent scorers re-labeled a subset of the dataset on shorter time windows. The automatic sleep scoring experiments were repeated on the re-annotated subset.The highest performance is achieved on features extracted from 30 s epochs of a single channel frontal EEG. The resulting accuracy, precision and recall were of 92.22%, 67.58% and 66.00% respectively. When using a shorter epoch as input, the performance decreased by approximately 20%. Re-annotating a subset of the dataset on shorter time epochs did not improve the results and further altered the sleep stage detection performance. Our results show that our feature-based LSTM classification algorithm performs better on 30 s PSG epochs when compared to 10 s epochs used as input. Future work could be oriented to determining whether varying the epoch size improves classification outcomes for different types of classification algorithms.
Collapse
|
2
|
Bates JHT, Irvin CG, Farré R, Hantos Z. Oscillation mechanics of the respiratory system. Compr Physiol 2013; 1:1233-72. [PMID: 23733641 DOI: 10.1002/cphy.c100058] [Citation(s) in RCA: 118] [Impact Index Per Article: 10.7] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 12/21/2022]
Abstract
The mechanical impedance of the respiratory system defines the pressure profile required to drive a unit of oscillatory flow into the lungs. Impedance is a function of oscillation frequency, and is measured using the forced oscillation technique. Digital signal processing methods, most notably the Fourier transform, are used to calculate impedance from measured oscillatory pressures and flows. Impedance is a complex function of frequency, having both real and imaginary parts that vary with frequency in ways that can be used empirically to distinguish normal lung function from a variety of different pathologies. The most useful diagnostic information is gained when anatomically based mathematical models are fit to measurements of impedance. The simplest such model consists of a single flow-resistive conduit connecting to a single elastic compartment. Models of greater complexity may have two or more compartments, and provide more accurate fits to impedance measurements over a variety of different frequency ranges. The model that currently enjoys the widest application in studies of animal models of lung disease consists of a single airway serving an alveolar compartment comprising tissue with a constant-phase impedance. This model has been shown to fit very accurately to a wide range of impedance data, yet contains only four free parameters, and as such is highly parsimonious. The measurement of impedance in human patients is also now rapidly gaining acceptance, and promises to provide a more comprehensible assessment of lung function than parameters derived from conventional spirometry.
Collapse
Affiliation(s)
- Jason H T Bates
- Vermont Lung Center, University of Vermont College of Medicine, Burlington, Vermont, USA.
| | | | | | | |
Collapse
|
3
|
Constant-phase descriptions of canine lung, chest wall, and total respiratory system viscoelasticity: effects of distending pressure. Respir Physiol Neurobiol 2012; 183:75-84. [PMID: 22691447 DOI: 10.1016/j.resp.2012.06.008] [Citation(s) in RCA: 16] [Impact Index Per Article: 1.3] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 02/08/2012] [Revised: 06/03/2012] [Accepted: 06/05/2012] [Indexed: 11/21/2022]
Abstract
The dynamic mechanical properties of the respiratory system reflect the ensemble behavior of its constituent structural elements. This study assessed the appropriateness of constant-phase descriptions of respiratory tissue viscoelasticity at various distending pressures. We measured the mechanical input impedance (Z) of the lungs, chest wall and total respiratory system in 12 dogs at mean airway pressures from 5 to 30 cm H(2)O. Each Z was fitted with a constant-phase model which provided estimates tissue damping (G), elastance (H), and hysteresivity (η=G/H). Both G and H sharply increased with increasing distending pressure for the lungs and chest wall, while η attained a minimum near 15-20 cm H(2)O. Model fitting errors for the lungs and total respiratory system increased for distending pressures greater than 20 cm H(2)O, indicating that constant-phase descriptions of parenchymal and respiratory system viscoelasticty may be inappropriate at volumes closer to total lung capacity. Such behavior may reflect alterations in load distribution across various parenchymal stress-bearing elements.
Collapse
|
4
|
|
5
|
Fahy BG, Barnas GM, Nagle SE, Flowers JL, Njoku MJ, Agarwal M. Effects of Trendelenburg and reverse Trendelenburg postures on lung and chest wall mechanics. J Clin Anesth 1996; 8:236-44. [PMID: 8703461 DOI: 10.1016/0952-8180(96)00017-7] [Citation(s) in RCA: 55] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 02/01/2023]
Abstract
STUDY OBJECTIVE To test whether the Trendelenburg ("head-down") or reverse Trendelenburg ("head-up") postures change lung and chest wall mechanical properties in a clinical condition. DESIGN Unblinded study, each patient serving as own control. SETTING University of Maryland at Baltimore Hospital, Baltimore, Maryland. PATIENTS 15 patients scheduled for laparoscopic surgery. INTERVENTIONS Patients were anesthetized and paralyzed, tracheally intubated and mechanically ventilated at 10 to 30 per minute and at a tidal volume of 250 to 800 ml. Measurements were made before surgery in supine, head-up (10 degrees from horizontal) and head-down (15 degrees from horizontal) postures. MEASUREMENTS AND MAIN RESULTS Airway flow and airway and esophageal pressures were measured. From these measurements, discrete Fourier transformation was used to calculate elastances and resistances of the total respiratory system, lungs, and chest wall. Total respiratory elastance and resistance increased in the head-down posture compared with supine due to increases in lung elastance and resistance (p < 0.05); but chest wall elastance and resistance did not change (p > 0.05). Lung elastance also exhibited a negative dependence on tidal volume while head-down that was not observed in the supine posture. The change in lung elastance compared with supine was positively correlated to body mass index (weight/height2) and negatively correlated to tidal volume. Lung and chest wall elastance and resistance were not affected by shifting from supine to head-up (p > 0.05). CONCLUSIONS The Trendelenburg posture increases the mechanical impedance of the lung to inflation, probably due to decreases in lung volume. This effect may become clinically relevant in patients predisposed with lung disease and in obese patients.
Collapse
Affiliation(s)
- B G Fahy
- Department of Anesthesiology, University of Maryland Hospital, Baltimore, USA
| | | | | | | | | | | |
Collapse
|
6
|
Barnas GM, Gilbert TB, Watson RJ, Sequeira AJ, Roitman K, Nooroni RJ. Respiratory mechanics in the open chest: effects of parietal pleurae. RESPIRATION PHYSIOLOGY 1996; 104:63-70. [PMID: 8865383 DOI: 10.1016/0034-5687(96)00010-2] [Citation(s) in RCA: 6] [Impact Index Per Article: 0.2] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 02/02/2023]
Abstract
To understand how the parietal pleurae affect the mechanical behavior of the human respiratory system after the chest wall is opened by median sternotomy, we studied 18 anesthetized/paralyzed patients immediately before coronary artery bypass grafting surgery. Elastances and resistances of the total respiratory system (ETr, Rrs) were calculated from measurements of airway pressure and flow during mechanical ventilation in the frequency and tidal volume ranges of normal breathing. Elastances and resistances of the lungs (EL, RL), chest wall (Ecw, Rcw) were also estimated from measurements of esophageal pressure. Data were collected in the closed chest, after median sternotomy with the parietal pleurae intact and after the left parietal pleura was opened for internal mammary artery harvest. After sternotomy with pleurae intact (n = 14), Ers did not change but Rrs decreased (p < 0.05). Ecw (including the contribution of the pleurae) was higher than in the closed chest (p < 0.05) while EL and RL were lower (p < 0.05); Rcw did not change. Opening the left pleura (n = 10) decreased Ers (p < 0.05), but Rrs did not change. We conclude that the chest wall/pleurae compartment offers significant impedance to lung expansion after sternotomy and rib retraction, unless one pleura is opened.
Collapse
Affiliation(s)
- G M Barnas
- Department of Anesthesiology Research Labs, University of Maryland, Baltimore 21201, USA
| | | | | | | | | | | |
Collapse
|
7
|
Fahy BG, Barnas GM, Flowers JL, Nagle SE, Njoku MJ. The effects of increased abdominal pressure on lung and chest wall mechanics during laparoscopic surgery. Anesth Analg 1995; 81:744-50. [PMID: 7574004 DOI: 10.1097/00000539-199510000-00015] [Citation(s) in RCA: 19] [Impact Index Per Article: 0.7] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 01/26/2023]
Abstract
We tested the hypothesis that increases in pressure in the abdomen (Pab) exerted by CO2 insufflation during laparoscopy would increase elastance (E) and resistance (R) of both the lungs and chest wall. We measured airway flow and airway and esophageal pressures of 12 anesthetized/paralyzed tracheally intubated patients during mechanical ventilation at 10-30/min and tidal volume of 250-800 mL. From these measurements, we used discrete Fourier transformation to calculate E and R of the lungs and chest wall. Measurements were made at 0, 15, and 25 mm Hg Pab in the 15 degrees head-down (Trendelenburg) posture and at 0 and 15 mm Hg Pab in the 10 degrees head-up (reverse Trendelenburg) posture. Lung and chest wall Es and Rs while head-down increased at Pab = 15 mm Hg, and both Es increased further at Pab = 25 mm Hg (P < 0.05). Both Es and Rs also increased while head-up at Pab = 15 mm Hg (P < 0.05), but increases in lung E and R were less than while head-down (P < 0.05). The increase in lung E and R at Pab = 15 mm Hg in either posture were positively correlated to body weight or body mass index, whereas the increases in chest wall E and R were negatively correlated to the same factors (P < 0.05). Lung and chest wall mechanical impedances increase with increasing Pab; the increases depend on body configuration and are greater while head-down.(ABSTRACT TRUNCATED AT 250 WORDS)
Collapse
Affiliation(s)
- B G Fahy
- Department of Anesthesiology, University of Maryland, Baltimore, USA
| | | | | | | | | |
Collapse
|
8
|
Fahy BG, Barnas GM, Flowers JL, Nagle SE, Njoku MJ. The Effects of Increased Abdominal Pressure on Lung and Chest Wall Mechanics During Laparoscopic Surgery. Anesth Analg 1995. [DOI: 10.1213/00000539-199510000-00015] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/05/2022]
|
9
|
Suki B, Davey BL, Sato J, Bates JH. A model of transient oscillatory pressure-flow relationships of canine airways. Ann Biomed Eng 1995; 23:682-90. [PMID: 7503468 DOI: 10.1007/bf02584465] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.1] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 01/25/2023]
Abstract
In a previous paper (27) we developed a lumped parameter model of canine pulmonary airway mechanics featuring airway wall elasticity, gas inertance, and laminar and turbulent gas flow. The model accurately accounted for the steady-state pressure-flow data we obtained during sinusoidal cycling of the lung following a period of apnea. In the present paper, we extend the model to account for the transient decrease in the amplitude of the trans-airway pressure swings that we observed immediately following the apnea, which we have shown to be due to a vagally mediated bronchodilatation reflex. The extended model accounts for this transient in terms of a sudden change in airway smooth muscle tone acting on the viscoelastic properties of the airway wall and tissues mechanically coupled to it. Consequently, this model is able to temporarily store a volume of gas in the conducting airway tree as its volume changes cyclically with that of the whole lung. This means that the flow entering the airway tree from the trachea at any instant (V) is not precisely equal to that entering the alveoli (Valv) even when the gas is considered incompressible. We found that assuming V to be equal to Valv can lead to errors in estimating respiratory tissue impedance of as much as 10%. However, tissue hysteresivity remained almost unaffected, suggesting that the hysteretic properties of respiratory system tissues and airway wall are well matched.
Collapse
Affiliation(s)
- B Suki
- Meakins-Christie Laboratories, McGill University, Montreal, Quebec, Canada
| | | | | | | |
Collapse
|
10
|
Affiliation(s)
- V Brusasco
- Dipartimento di Scienze Motorie, Università di Genova, Italy
| | | |
Collapse
|
11
|
Barnas GM, Harinath P, Green MD, Suki B, Kaczka DW, Lutchen KR. Influence of waveform and analysis technique on lung and chest wall properties. RESPIRATION PHYSIOLOGY 1994; 96:331-44. [PMID: 8059094 DOI: 10.1016/0034-5687(94)90137-6] [Citation(s) in RCA: 15] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 01/28/2023]
Abstract
To test an approach for measuring respiratory system resistance (R) and elastance (E) during non-sinusoidal forcing, we measured airway and esophageal pressures and flow at the trachea of 9 anesthetized-paralyzed dogs during sinusoidal forcing (SF) and 4 types of non-sinusoidal forcings at 0.15 and 0.6 Hz and 300 ml tidal volume. During SF, calculations of E and R of the lungs, chest wall or total system from discrete Fourier transform (DFT) and two other widely used methods (multiple regression and volume-pressure loop analysis) did not differ from each other (P > 0.05). During forcing with sinusoidal or step inspiration with passive expiration (inspiratory to expiratory ratio, I/E, = 1:1), Es from any analysis method were within 10% of values during SF. Although Rs of the lungs, chest wall or total system were not affected by waveform shape with DFT (P > 0.05), the other analysis methods gave values for R during non-SF that differed (P < 0.05) from those during SF by up to 77%. If I/E was changed to 1:2, with or without an added 10% inspiratory pause, values for E and R differed least from values during SF if DFT was used. During severe pulmonary edema induced by infusion of oleic acid in the right atrium, results for lung properties were similar to controls, despite large increases in E and R of the lungs. We conclude that E and R of the lungs and chest wall can be measured by DFT using nonsinusoidal forcing waveforms available on most clinical ventilators, incurring only modest error.
Collapse
Affiliation(s)
- G M Barnas
- Department of Anesthesiology, Anesthesiology Research Labs, Baltimore, MD 21201
| | | | | | | | | | | |
Collapse
|
12
|
Barnas GM, Ho G, Green MD, Harinath P, Smalley AJ, Campbell DN, Mendham JE. Effects of analysis method and forcing waveform on measurement of respiratory mechanics. RESPIRATION PHYSIOLOGY 1992; 89:273-85. [PMID: 1410841 DOI: 10.1016/0034-5687(92)90086-c] [Citation(s) in RCA: 14] [Impact Index Per Article: 0.4] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 12/26/2022]
Abstract
The respiratory system has been shown to exhibit nonlinear mechanical properties in the frequency (f) range of normal breathing, manifested by tidal volume (Vt) dependence. Calculations of respiratory system resistance (R) and elastance (E) from pressure-flow measurements during external forcing at a given f may be ambiguous, especially if non-sinusoidal forcing waveforms are used. We evaluated the degree to which R and E depended upon: (1) analysis method (Fourier transform, multiple regression and pressure-volume loop analysis) and; (2) shape of the forcing waveform (sinusoidal, quasi-sinusoidal and step). We measured pressure and flow at the mouth of 5 healthy, awake subjects, relaxed at functional residual capacity, during forcing with the three different waveforms in the normal range of f (0.2-0.6 Hz) and Vt (250-750 ml). During sinusoidal forcing, E and R were not affected by analysis method (P greater than 0.2). With Fourier transform and multiple regression, E was not affected by waveform shape (P greater than 0.05); with loop analysis, E was slightly (less than 10%) higher during quasi-sinusoidal and step forcing than during the sine (P less than 0.05). R was least affected by waveform shape with Fourier transform. We conclude that, in the f and Vt range of normal breathing: (1) respiratory system impedance is 'quasi-linear,' i.e. despite dependencies of R and E on Vt, non-linearities are not large enough to restrict interpretation of R and E at a given f and Vt; (2) it may be possible to measure R and E using non-sinusoidal forcing waveforms available on most clinical ventilators, incurring only modest error.
Collapse
Affiliation(s)
- G M Barnas
- Department of Anesthesiology, University of Maryland, Baltimore 21201
| | | | | | | | | | | | | |
Collapse
|
13
|
Barnas GM, Hempleman SC, Harinath P, Baptiste JW. Respiratory system mechanical behavior in the chicken. RESPIRATION PHYSIOLOGY 1991; 84:145-57. [PMID: 1876756 DOI: 10.1016/0034-5687(91)90113-w] [Citation(s) in RCA: 5] [Impact Index Per Article: 0.2] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 12/29/2022]
Abstract
We evaluated whether the avian respiratory system displays the same fundamental mechanical behavior during external forcing as found in mammals. We measured airway flow and pressures in the trachea, air sacs and thoracoabdominal cavity in 4 anesthetized-paralyzed roosters during sinusoidal volume oscillations at the trachea in the normal range of euthermic breathing frequency, f(0.2 to 1.0 Hz), and tidal volume, VT (10-50 ml). From the pressure and flow waveforms, we calculated resistance (R) and elastance (E) of the total respiratory system and its major compartments (lungs, air sacs and chest wall). E of the chest wall was minimum (147 cmH2O.L-1 +/- 7 SE) at 0.2 Hz-50 ml and was consistently, slightly lower than E of the total respiratory system over the entire range studied. Both elastances showed the same dependence on f and VT, increasing slightly with increasing f and decreasing with increasing VT. R of the chest wall was maximum (35.6 cmH2O.L- 1.sec-1 +/- 2.2 SE) at 0.2 Hz-10 ml and decreased with increasing f and VT, although the VT effect diminished at the higher f. E and R of the air sacs were much smaller than those of the chest wall, but showed similar f and VT dependencies. R of the lungs, due to resistance of the airways, was minimum (6.8 cmH2O.L-1.sec-1 +/- 1.5 SE) at 0.2 Hz-10 ml and increased with both f and VT. Total respiratory R reflected R of the air sacs and chest wall at low f and R of the lungs at high f. The f and VT dependencies of E and R in the chicken were strikingly similar to those measured in various types of mammalian respiratory tissues (Stamenović et al. (1990) J. Appl. Physiol. 69: 973-988. We conclude that, despite important anatomical differences between species, avian and mammalian respiratory tissues exhibit fundamentally similar mechanical behavior.
Collapse
Affiliation(s)
- G M Barnas
- Department of Anesthesiology, University of Maryland, Baltimore 21201
| | | | | | | |
Collapse
|