|
|
October 17-21, 2009 New Orleans, LA Home Abstract Archive Search Abstracts 2009 Session Grid 2009 Meeting Website Policy Statements ASA Website Feedback 
Visit Anesthesiology.org |
| Previous Abstract | Next Abstract |
| Printable Version |
|
A188
October 17, 2009 10:00 AM - 11:30 AM Room Room 356 |
| Artificial Neural Networks Improve Accuracy of Respiratory Rate from Capnometry during Sedation |
|
* Joseph A. Orr, Ph.D., Ken B. Johnson, M.D., Lara M. Brewer, M.S. Anesthesiology, University of Utah, Salt Lake City, Utah |
|
Introduction:
Capnometers are used during sedation in non-intubated patients to monitor respiratory rate and end-tidal CO
2
concentration. Because capnometers detect changes in CO
2
concentration and not gas movement, they often confuse spurious CO
2
concentration changes with actual breaths leading to false alarms and missed detection of periods of apnea. Cardiogenic oscillations, patient speaking and unsuccessful breath attempts during obstruction are causes of changes in CO
2
concentration that can be falsely detected as breaths.
An artificial neural network (ANN) was trained to estimate the tidal volume of each breath based on the shape of the capnogram. If the trained ANN identified the capnogram as having a tidal volume of at least 200 ml, then it was considered a valid breath, otherwise it was rejected as artifact. We evaluated the accuracy of this combination ANN algorithm, and a simple threshold algorithm as compared to a reference breath rate as measured using a pneumotach. Methods: 24 volunteers were fitted with a tight fitting sealed mask connected to a combination flow and CO 2 sensor. Combinations of propofol and remifentanil concentrations delivered as target controlled infusions were administered to each volunteer to simulate various sedation conditions. A simple threshold algorithm was applied in which any deviation of the CO 2 signal above 10 mm Hg and below 5 mm Hg was used to identify a breath. For the ANN, the CO 2 signal of each detected breath was down-sampled (from 100 samples/sec) and applied as the inputs to an ANN. Using data from 12 volunteers, the ANN was trained to estimate the tidal volume associated with the detected breath. The threshold breath detection was considered valid only for breaths for which the ANN identified a tidal volume sufficient to clear the airway dead space (200 ml). Both the simple threshold algorithm and the ANN detection algorithm were compared against the reference pneumotach breath rate using data from the remaining 12 volunteers. Results: Using the threshold algorithm, the average detected breath rate was 3 breaths per minute (bpm) higher than the reference rate. The standard deviation of the difference was 4.7 bpm. When the ANN modification was applied, the average breath rate was 0.07 bpm less than the reference with the standard deviation of the difference of 2.5 bpm. The plot below shows the breath rate data for a typical volunteer using both methods.[figure1] Discussion: The capnometry signal observed during sedation is susceptible to signal corruption and misinterpretation. Inaccuracy of the breath rate measurement is a cause of distracting false positive high respiratory rate alarms and failure to detect periods of actual apnea during sedation. Analysis of the CO 2 signal for each breath to assess validity is beyond the ability of the typical user. An artificial neural network can be applied as a intelligent filter providing breath-by-breath analysis of the capnogram signal to improve reliability during sedation. From Proceedings of the 2009 Annual Meeting of the American Society Anesthesiologists. |
Figure 1
 |