Detail publikace

Fully Automatic Detection of Strict Left Bundle Branch Block

Originální název

Fully Automatic Detection of Strict Left Bundle Branch Block

Anglický název

Fully Automatic Detection of Strict Left Bundle Branch Block

Jazyk

en

Originální abstrakt

Background: Strict (true) left bundle branch block (tLBBB) is a diagnostic marker that was proposed to predict cardiac resynchronization therapy responders. The criteria for tLBBB include the presence of QS- or rS- configurations of QRS in V1 and V2, the presence of mid-QRS notching or slurring in at least two of leads V1, V2, V5, V6, I and avL and finally a QRS duration (QRSd) of more than 130 ms (women) or 140 ms (men). Purpose: The main objective was to develop an automatic detector of tLBBB with the best possible success. This work was involved in the competition LBBB Initiative of the ICSE 2018. Methods: Ten-second six-lead ECG from the MADIT-CRT database were used to develop (300 signals – 96 tLBBB and 202 non-tLBBB) and test (302 signals) the detector. We detected all parts of the criteria separately. QRS notching and slurring were detected by thresholding several features - width and prominence of local extremes in QRS (for QRS notching detection) and the width and prominence of the local extremes in the first difference of QRS and standard deviation change in several different floating windows (for QRS slurring detection). The QRS configuration was evaluated according to the dominant deflection of the QRS. It must be negative without positive deflection at the end of the QRS. Publicly available algorithm (ECG SEEKER, Brno University of Technology) was used for measurement of the QRS duration. Results: Accuracy, sensitivity and specificity of tLBBB were 0.88, 0.86 and 0.90, respectively for a training dataset and 0.81, 0.69 and 0.87, respectively for a hidden testing dataset. Sensitivity and specificity of QRS configuration detection were 1.00 and 0.99, respectively for a training dataset. The average deviation of the reference values and measured QRSd is 9.8 ms. The accuracy of QRS notching/slurring detection is not determined due to the fact that the detector does not detect it near the QRS onset and offset, so the results cannot be compared with the references. Conclusion: The created detector was the best in the competition LBBB Initiative of the ICSE 2018, however the accuracy of the detector is low for use without verification. This is due to the fact that the QRS notching and slurring are not exactly defined. There is also no methodology for accurate determination the beginning and end of QRS notching/slurring. The accuracy of tLBBB detection is also highly dependent on the generally problematic QRS duration measurement.

Anglický abstrakt

Background: Strict (true) left bundle branch block (tLBBB) is a diagnostic marker that was proposed to predict cardiac resynchronization therapy responders. The criteria for tLBBB include the presence of QS- or rS- configurations of QRS in V1 and V2, the presence of mid-QRS notching or slurring in at least two of leads V1, V2, V5, V6, I and avL and finally a QRS duration (QRSd) of more than 130 ms (women) or 140 ms (men). Purpose: The main objective was to develop an automatic detector of tLBBB with the best possible success. This work was involved in the competition LBBB Initiative of the ICSE 2018. Methods: Ten-second six-lead ECG from the MADIT-CRT database were used to develop (300 signals – 96 tLBBB and 202 non-tLBBB) and test (302 signals) the detector. We detected all parts of the criteria separately. QRS notching and slurring were detected by thresholding several features - width and prominence of local extremes in QRS (for QRS notching detection) and the width and prominence of the local extremes in the first difference of QRS and standard deviation change in several different floating windows (for QRS slurring detection). The QRS configuration was evaluated according to the dominant deflection of the QRS. It must be negative without positive deflection at the end of the QRS. Publicly available algorithm (ECG SEEKER, Brno University of Technology) was used for measurement of the QRS duration. Results: Accuracy, sensitivity and specificity of tLBBB were 0.88, 0.86 and 0.90, respectively for a training dataset and 0.81, 0.69 and 0.87, respectively for a hidden testing dataset. Sensitivity and specificity of QRS configuration detection were 1.00 and 0.99, respectively for a training dataset. The average deviation of the reference values and measured QRSd is 9.8 ms. The accuracy of QRS notching/slurring detection is not determined due to the fact that the detector does not detect it near the QRS onset and offset, so the results cannot be compared with the references. Conclusion: The created detector was the best in the competition LBBB Initiative of the ICSE 2018, however the accuracy of the detector is low for use without verification. This is due to the fact that the QRS notching and slurring are not exactly defined. There is also no methodology for accurate determination the beginning and end of QRS notching/slurring. The accuracy of tLBBB detection is also highly dependent on the generally problematic QRS duration measurement.

BibTex


@misc{BUT147403,
  author="Radovan {Smíšek} and Pavel {Jurák} and Josef {Halámek} and Filip {Plešinger}",
  title="Fully Automatic Detection of Strict Left Bundle Branch Block",
  annote="Background: Strict (true) left bundle branch block (tLBBB) is a diagnostic marker that was proposed to predict cardiac resynchronization therapy responders. The criteria for tLBBB include the presence of QS- or rS- configurations of QRS in V1 and V2, the presence of mid-QRS notching or slurring in at least two of leads V1, V2, V5, V6, I and avL and finally a QRS duration (QRSd) of more than 130 ms (women) or 140 ms (men).
Purpose: The main objective was to develop an automatic detector of tLBBB with the best possible success. This work was involved in the competition LBBB Initiative of the ICSE 2018.
Methods: Ten-second six-lead ECG from the MADIT-CRT database were used to develop (300 signals – 96 tLBBB and 202 non-tLBBB) and test (302 signals) the detector. We detected all parts of the criteria separately. QRS notching and slurring were detected by thresholding several features - width and prominence of local extremes in QRS (for QRS notching detection) and the width and prominence of the local extremes in the first difference of QRS and standard deviation change in several different floating windows (for QRS slurring detection). The QRS configuration was evaluated according to the dominant deflection of the QRS. It must be negative without positive deflection at the end of the QRS. Publicly available algorithm (ECG SEEKER, Brno University of Technology) was used for measurement of the QRS duration.
Results: Accuracy, sensitivity and specificity of tLBBB were 0.88, 0.86 and 0.90, respectively for a training dataset and 0.81, 0.69 and 0.87, respectively for a hidden testing dataset. Sensitivity and specificity of QRS configuration detection were 1.00 and 0.99, respectively for a training dataset. The average deviation of the reference values and measured QRSd is 9.8 ms. The accuracy of QRS notching/slurring detection is not determined due to the fact that the detector does not detect it near the QRS onset and offset, so the results cannot be compared with the references.
Conclusion: The created detector was the best in the competition LBBB Initiative of the ICSE 2018, however the accuracy of the detector is low for use without verification. This is due to the fact that the QRS notching and slurring are not exactly defined. There is also no methodology for accurate determination the beginning and end of QRS notching/slurring. The accuracy of tLBBB detection is also highly dependent on the generally problematic QRS duration measurement.",
  address="International Society for Computerized Electrocardiology",
  booktitle="43rd Annual Conference of the ISCE",
  chapter="147403",
  howpublished="print",
  institution="International Society for Computerized Electrocardiology",
  number="1",
  year="2018",
  month="april",
  pages="1--1",
  publisher="International Society for Computerized Electrocardiology",
  type="abstract"
}