Establishment of a reference library for evaluating computer ECG measurement programs

Willems, Jos L.; Arnaud, Pierre; Bemmel, Jan H. van; Bourdillon, Peter J.; Degani, R.; Denis, B; Harms, Frits M.A.; Macfarlane, Peter W.; Mazzocca, G.; Meyer, Jürgen; Eck, Henk J. Ritsema van; Medina, Etienne O. Robles de; Zywietz, C.

doi:10.1016/0010-4809(85)90021-7

Cited by 57 publications

(38 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bortolan trained on the CSE and tested on the PTB set and also reported a bias in reference QT measurements with a much larger value of 25 ms [13]. This bias could easily be explained by the fact that the PTB annotators concentrated on lead II while the CSE annotators used 8 available leads I, II and V1 -V6 [14,15].…”

Section: Discussionmentioning

confidence: 99%

“…Through measurements made on the representative beat, results from longer ECG records and short 10 second ECG records can be compared. A second data set from the CSE project was used here [14]. The 100 ten-second records were selected from the total 125 according to the selection for computerized algorithm testing in the essential performance standard IEC 60601-2-51.…”

Section: Study Populationmentioning

confidence: 99%

See 1 more Smart Citation

Comparison of two automated methods for QT interval measurement

Gregg

Babaeizadeh

Feild

et al. 2007

2007 Computers in Cardiology

View full text Add to dashboard Cite

In this paper we compared two methods of automated QT interval measurement on standard ECG databases: the Root-Mean-Square (RMS) lead combining method aimed at QT monitoring and the method of median of lead-by-lead QT interval measurements.We IntroductionGlobal QT interval is one of the fundamental ECG measurements reported on virtually every 12 lead ECG. A longer than normal QT interval may indicate a congenital or acquired long QT condition [1][2][3]. The AHA/ACC practice guideline for ECG monitoring now includes a recommendation to monitor QT interval for the purpose of drug titration of drugs known to have a pro-arrhythmic effect [4]. If the QT interval lengthens by more than 60ms after starting the drug or the QT interval extends beyond 500 ms the administration of the drug should be stopped or the dosage can be reduced.We have previously presented Philips automated QT interval measurement algorithms for 12-lead ECG, Holter and ECG monitoring applications [5][6][7][8][9][10]. In the ambulatory Holter and patient monitoring ECG applications, the QT interval algorithm uses an RMS waveform from combined available high quality leads and measures QT interval on the RMS ECG. In the resting 12-lead ECG application, the global QT interval measurement is based on a lead-by-lead method. The open question is, of the two, which method is better ignoring the constraints of the application. In this paper, we compared the two automated methods and reported the results using the same ECG datasets. Study PopulationA comparison between automated methods aimed at either monitoring or ambulatory ECG versus 12-lead ECG is hampered by the gross difference in available ECG in each case. Ambulatory ECGs cannot be used to test the 12-lead QT algorithm because of low sample rate, narrow bandwidth and limited leads. For 12-lead ECG analysis, a sample rate of 500 sps and a bandwidth of 0.05 to 150Hz are required. Ambulatory ECG recordings often have sample rates around 200 sps and a cut-off frequency of 40Hz. Only selected parts of the 12-lead analysis are possible with the small number of leads used in an ambulatory or monitoring recording. On the other hand, the sample rate, bandwidth and number of leads may be adequate for a 12-lead ECG to be used for an ambulatory analysis, but the 10 second recording is not long enough for even the learning period of the ambulatory and monitoring algorithmsThe addition of the PTB set to the data publicly available at PhysioNet made this study possible because it has the features that allow direct comparison between ambulatory and 12 lead algorithms [11]. The PTB dataset consists of 549 records from 294 subjects with a sample

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Study Populationmentioning

confidence: 99%

Comparison of two automated methods for QT interval measurement

Gregg

Babaeizadeh

Feild

et al. 2007

2007 Computers in Cardiology

View full text Add to dashboard Cite

show abstract

“…The computation of the probabilities P i FP and P i NP for each channel is done in the evaluation procedure according to the ratio P i ϭ false detections overall detections . [11] Each time a decision is made, in those leads for which the prediction was the same as the decision, we will increase the denominator of the above ratio by 1 and will calculate the new lower values of P. Correspondingly, in the leads for which a false prediction was made, we will also increase the denominator by 1, but at the same time we will also increase by 1 the nominator of the P value of the false prediction, which will lead to an increase of this P value. If, for example, a false negative prediction was made, the new P i FN for that lead will be higher than the former one.…”

Section: Decision Algorithmmentioning

confidence: 97%

“…4 we give some examples based on ECGs from the CSE library (11). It is important to notice that the large value of (Ͼ0.85) does not necessarily mean that we have false positive detections.…”

Section: Evaluation Proceduresmentioning

confidence: 98%