Signal Processing and machine learning contributions to rhythm analysis during cardiopulmonary resuscitation

• Autores: Iraia Isasi Liñero
• Directores de la Tesis: Elisabete Aramendi Ecenarro (dir. tes.), Unai Irusta Zarandona (dir. tes.)
• Lectura: En la Universidad del País Vasco - Euskal Herriko Unibertsitatea ( España ) en 2020
• Idioma: inglés
• Títulos paralelos:
  • Seinale prozesaketan eta ikasketa automatikoan oinarritutako ekarpenak bihotz-erritmoen analisirako bihotz-biriketako berpiztean
• Tribunal Calificador de la Tesis: Pablo Laguna Lasaosa (presid.), Eva Navas Cordón (secret.), Irena Llieva Jekova (voc.)
• Programa de doctorado: Programa de Doctorado en Electrónica y Telecomunicaciones por la Universidad del País Vasco/Euskal Herriko Unibertsitatea
• Materias:
  • Ciencias de la vida
    • Biofísica
      • Biomecánica
• Enlaces
  • Tesis en acceso abierto en: ADDI
• Resumen

  • Out-of-hospital cardiac arrest (OHCA ) is characterized by the sudden loss of the cardiac function, andcauses around 10% of the total mortality in developed countries. Survival from OHCA depends largelyon two factors: early defibrillation and early cardiopulmonary resuscitation (CPR). The electrical shock isdelivered using a shock advice algorithm (SAA) implemented in defibrillators. Unfortunately, CPR mustbe stopped for a reliable SAA analysis because chest compressions introduce artefacts in the ECG. Theseinterruptions in CPR have an adverse effect on OHCA survival. Since the early 1990s, many efforts havebeen made to reliably analyze the rhythm during CPR. Strategies have mainly focused on adaptive filtersto suppress the CPR artefact followed by SAAs of commercial defibrillators. However, these solutionsdid not meet the American Heart Association¿s (AHA) accuracy requirements for shock/no-shockdecisions. A recent approach, which replaces the commercial SAA by machine learning classifiers, hasdemonstrated that a reliable rhythm analysis during CPR is possible. However, defibrillation is not theonly treatment needed during OHCA, and depending on the clinical context a finer rhythm classificationis needed. Indeed, an optimal OHCA scenario would allow the classification of the five cardiac arrestrhythm types that may be present during resuscitation. Unfortunately, multiclass classifiers that allow areliable rhythm analysis during CPR have not yet been demonstrated. On all of these studies artefactsoriginate from manual compressions delivered by rescuers. Mechanical compression devices, such as theLUCAS or the AutoPulse, are increasingly used in resuscitation. Thus, a reliable rhythm analysis duringmechanical CPR is becoming critical. Unfortunately, no AHA compliant algorithms have yet beendemonstrated during mechanical CPR. The focus of this thesis work is to provide new or improvedsolutions for rhythm analysis during CPR, including shock/no-shock decision during manual andmechanical CPR and multiclass classification during manual CPR. // Out-of-hospital cardiac arrest (OHCA ) is characterized by the sudden loss of the cardiac function, andcauses around 10% of the total mortality in developed countries. Survival from OHCA depends largelyon two factors: early defibrillation and early cardiopulmonary resuscitation (CPR). The electrical shock isdelivered using a shock advice algorithm (SAA) implemented in defibrillators. Unfortunately, CPR mustbe stopped for a reliable SAA analysis because chest compressions introduce artefacts in the ECG. Theseinterruptions in CPR have an adverse effect on OHCA survival. Since the early 1990s, many efforts havebeen made to reliably analyze the rhythm during CPR. Strategies have mainly focused on adaptive filtersto suppress the CPR artefact followed by SAAs of commercial defibrillators. However, these solutionsdid not meet the American Heart Association¿s (AHA) accuracy requirements for shock/no-shockdecisions. A recent approach, which replaces the commercial SAA by machine learning classifiers, hasdemonstrated that a reliable rhythm analysis during CPR is possible. However, defibrillation is not theonly treatment needed during OHCA, and depending on the clinical context a finer rhythm classificationis needed. Indeed, an optimal OHCA scenario would allow the classification of the five cardiac arrestrhythm types that may be present during resuscitation. Unfortunately, multiclass classifiers that allow areliable rhythm analysis during CPR have not yet been demonstrated. On all of these studies artefactsoriginate from manual compressions delivered by rescuers. Mechanical compression devices, such as theLUCAS or the AutoPulse, are increasingly used in resuscitation. Thus, a reliable rhythm analysis duringmechanical CPR is becoming critical. Unfortunately, no AHA compliant algorithms have yet beendemonstrated during mechanical CPR. The focus of this thesis work is to provide new or improvedsolutions for rhythm analysis during CPR, including shock/no-shock decision during manual andmechanical CPR and multiclass classification during manual CPR.