Table 1.
Unique, mutually-exclusive patterns of respiratory inductive plethysmography and their scoring rules.
Fig 1.
Elements of the RIPScore interface.
(A) Respiratory Inductive Plethysmography (RIP) Pattern; (B) Signals from ribcage (RCG), abdomen (ABD), photoplethysmograph (PPG), and blood oxygen saturation (SAT); (C) Notes; (D) Segment and Epoch Control; (E) Lissajous Figure; (F) RIP Pattern Scoring; and (G) Mode Control. The epoch shows a representative example of Pause (PAU). The quasi-sinusoidal pattern in RCG and ABD stops during the PAU highlighted in red. The horizontal dotted cursors in RCG show an estimated variation of ± 90% of the amplitude of the breath preceding the PAU. Note that these cursors do not take into account low frequency trends, and so are only an approximate reference. a.u. = arbitrary units. (A) RIP Pattern: a color-coded bar showing the RIP pattern assigned by the scorer at each time; (B) Signals: plots of the cardiorespiratory signals including ribcage (RCG), abdomen (ABD), photoplethysmograph (PPG), and blood oxygen saturation (SAT). Clicking on a breath from RCG or ABD plots three horizontal cursors, one at the estimated breath’s amplitude, and two at ± 90% of that amplitude. Note that these cursors are not an exact amplitude reference for the epoch because they do not take into account low frequency trends frequently observed in RIP signals [35]; (C)Notes: text boxes showing time stamped notes made during data acquisition, and comments entered by the scorer during analysis; (D) Segment and Epoch Control: text boxes showing the start and end times for the current segment (highlighted in red in Signals); command buttons to add a “Comment” or “Delete” the RIP pattern assigned to the current segment; command buttons to scroll through epochs (“Previous”, “Next”), and a text box with the start time of the current epoch; (E) Lissajous Figure: a plot of RCG versus ABD for the current segment to aid the user in evaluating thoraco-abdominal synchrony. During breathing, the plot will be an ellipse tilted to the right for a phase less than 90 degrees, a circle for a phase of 90 degrees, and an ellipse tilted to the left for a phase greater than 90 degrees; (F) RIP Pattern Scoring: color-coded command buttons that assign a RIP pattern to the current segment; each button may also be activated by hitting the corresponding keyboard “hot-key” defined by the character in parenthesis for each button (e.g., the hot-key for Pause is ‘1’); (G) Mode Control: command button to switch between scoring and visualization mode.
Fig 2.
Concatenation of signal segments.
(A) Sample input segments. (B) Input segments are aligned and overlapped over a transition window T. (C) The output during this window is computed by gradually attenuating the end of the first segment, gradually incrementing the start of the second segment, and adding the two parts to yield a smooth transition. (D) The output signal consists on the first segment up to the start of T, followed by the transition, followed by the second segment starting after T.
Fig 3.
Representative example of Synchronous-Breathing (SYB).
The ellipse in the Lissajous plot of ribcage (RCG) against abdomen (ABD) is tilted to the right. PPG = photoplethysmograph, SAT = blood oxygen saturation, a.u. = arbitrary units.
Fig 4.
Representative example of Asynchronous-Breathing (ASB).
The Lissajous plot of ribcage (RCG) against abdomen (ABD) for the segment highlighted in red shows ellipses tilted to the left. PPG = photoplethysmograph, SAT = blood oxygen saturation, a.u. = arbitrary units.
Fig 5.
Representative example of Sigh (SIH).
The SIH highlighted in red has larger amplitude and longer duration than the other breaths. The horizontal dotted cursors in the ribcage (RCG) signal show an estimated variation of ± 90% of the amplitude of the breath preceding the SIH. Note that these cursors are not an exact amplitude reference. Also, the Lissajous plot shows an ellipse tilted to the right. ABD = abdomen, PPG = photoplethysmograph, SAT = blood oxygen saturation, a.u. = arbitrary units.
Fig 6.
Representative example of Movement Artifact (MVT).
The MVT in the ribcage (RCG) and abdomen (ABD) signals is highlighted in red. PPG = photoplethysmograph, SAT = blood oxygen saturation, a.u. = arbitrary units.
Fig 7.
Representative example of a Pause (PAU) which follows a Sigh (SIH).
The horizontal dotted cursors in the abdomen (ABD) signal show an estimated variation of ± 90% of the amplitude of the breath that precedes the SIH. Note that these cursors are not an exact amplitude reference. RCG = ribcage, PPG = photoplethysmograph, SAT = blood oxygen saturation, a.u. = arbitrary units.
Fig 8.
Representative example of a Pause (PAU) which follows a Movement Artifact (MVT).
The horizontal dotted cursors in the ribcage (RCG) signal show an estimated variation of ± 90% of the amplitude of the breath that follows the PAU. Note that these cursors are not an exact amplitude reference. ABD = abdomen, PPG = photoplethysmograph, SAT = blood oxygen saturation, a.u. = arbitrary units.
Fig 9.
It is not possible to determine the pattern in the selected segment (red) because the ribcage (RCG) signal shows a low-frequency, chaotic pattern, while the abdomen (ABD) signal has a quasi-sinusoidal breathing pattern with an additional low-frequency movement component. PPG = photoplethysmograph, SAT = blood oxygen saturation, a.u. = arbitrary units.
Fig 10.
Study Data Flowchart.
Table 2.
Proportion of “true-pattern” samples in the records used to create the validation data subset.
Fig 11.
Criteria to successfully complete levels: (A) Level 1, the trainee obtained accuracy and consistency values of κ ≥ 0.8; and (B) Level 2, the trainee obtained accuracy and consistency values of κ ≥ 0.8 on two consecutive sessions.
Fig 12.
(A) Type I, and (B) Type II data segments were concatenated to generate the training records. (C) Validation records were pre-processed such that Type II segments were inserted into the validation subset. Red vertical lines indicate the concatenation point.
Table 3.
Training accuracy.
Table 4.
Training consistency.
Fig 13.
(A) Accuracy (Fleiss’ κ); (B) consistency (Fleiss’ κ); and (C) rate (hours of data per hour of scoring) as a function of number of data records analyzed. SC1 was a pediatric anesthesiologist; SC2 was an experienced sleep laboratory scorer; and SC3 was a data networks analyst with no clinical experience. Standard deviation of each accuracy and consistency point was < 0.01.
Fig 14.
Evaluation of manual scoring of Pause.
(A) Accuracy (Fleiss’ κ); (B) consistency (Fleiss’ κ); and (C) rate (hours of data per hour of scoring) as a function of number of data records analyzed. Results are shown for the 42 data records analyzed (21 files scored twice).
Table 5.
Intra-scorer repeatability.
Table 6.
Inter-scorer repeatability of scorers SC1, SC2, and SC3.
Table 7.
Proportion of consensus patterns for the confusion analysis.
Fig 15.
Conditional probability of each respiratory inductive plethysmography (RIP) pattern for samples with the consensus pattern of: (A) synchronous-breathing (SYB), (B) asynchronous-breathing (ASB), (C) pause (PAU), (D) sigh (SIH), (E) movement artifact (MVT), and (F) unknown (UNK). When there is no confusion, the consensus pattern has a probability of 1 and the others have probabilities of 0. During total confusion all patterns have equal probabilities. Standard deviations of all probabilities were < 0.01.
Fig 16.
Confusion of SC1 on samples with consensus pattern of pause as a function of segment length.
SYB = synchronous-breathing, ASB = asynchronous-breathing, SIH = sigh, PAU = pause, MVT = movement artifact, UNK = unknown. A probability of 1 for PAU indicates no confusion. Lower PAU probabilities indicate increased confusion. Standard deviations of all probabilities were < 0.01.