next up previous contents
Next: 3. Database Access Up: 2.3 High resolution waveforms Previous: 2.3.4.1 Annotated alarms   Contents


2.3.5 Signal Quality

Although our search engines allow a researcher to determine what signals exist for which patients, this is no guarantee of quality, and sometimes the data can be so noisy that no useful clinical information can be extracted from the data. To avoid requesting noisy data, and using this data for further processing, we have developed a set of signal quality indices (SQI's) for both electrocardiogram and blood pressure data. We have also used these indices and a multi-channel weighting algorithm, to generate our best estimates of heart rate and blood pressure for every 10 second window of waveform data.

The ECG signal quality metrics are a combination of statistical measures, in both the time and frequency domains, multi-channel QRS detector performance, and correlation to past data. A more in-depth description of the formation of these annotations, together with an evaluation of a robust HR and ABP tracking algorithm that utilizes this information can be found in (9) and (10). The blood pressure signal quality metric is based upon two earlier developed metrics by Zong (7) and Sun (8).

There are several annotations associated with each beat in the ECG and ABP signals. Table 2.8 describes the available SQI annotations for waveform records. Full descriptions of how to interpret the SQI output for ECG and ABP can be found in in (9) and (10) respectively. Note that although many of the artifact types for each of these signals have been incorporated, and known errors in heart rate and blood pressure calibrated to the SQI output, some artifacts are not well represented. In particular, the tricky problem of blood pressure damping is not yet fully solved in our ABP SQI metric. Any analysis of blood pressure should therefore be tempered by the fact that damping may lead to an error, and in particular, an under-estimation of the SBP and pulse pressure.

Annotation Description
file.ecgsqid A combined beat-by-beat ECG SQI created by selecting the best ECG SQI between different ECG leads.
file.ecgsqidn The ECGSQI annotation of the $n^{th}$ ($n=0,1,2, ...$) ECG lead, used to create a combined annotation file'.multid'
file.epltdn The EPLTD (DF) annotation of the $n^{th}$ ($n=0,1,2, ...$) ECG lead, used to create EPSQI and ICHSQI
file.epsqin The EPSQI (bSQI) annotation of the $n^{th}$ ($n=0,1,2, ...$) ECG lead, used to create Kurtosis (kSQI) and SDR (sSQI)
file.ichsqin The ICHSQI (iSQI) annotation of the $n^{th}$ ($n=0,1,2, ...$) ECG lead, used to create ECGSQI
file.kurtdn The Kurtosis (kSQI) and SDR (sSQI) annotation of the $n^{th}$ ($n=0,1,2, ...$) ECG lead, used to create ECGSQI and the sample-and-hold HR (HRsh1) of the first ECG lead
file.multid An all-in-one annotation include all leads of ECG beats with ECGSQI and ABP beat with ABPSQI, used to create the HR and ABP
fileTd A WFDB trend data file that includes different calculations of HR, HRSQI and ABP sampled at 0.1Hz
fileTd.hea A header file of fileTd
file.wqrsn The WQRS (LT) annotation of the $n^{th}$ ($n=0,1,2, ...$) ECG lead, used to create EPSQI (bSQI)
fileT A WFDB trend data file similar to fileTd, but without baseline wander filter to calculate kurtosis.
fileT.hea Header file for fileT
file.wsqi The WSQI annotation of the ABP lead, used to create ABPSQI
file.jsqi The JSQI annotation of the ABP lead, used to create ABPSQI
file.abpsqi The ABPSQI annotation of the ABP lead, used to create '.multid'
   

next up previous contents
Next: 3. Database Access Up: 2.3 High resolution waveforms Previous: 2.3.4.1 Annotated alarms   Contents
djscott 2011-09-07