Biomedical Signal Processing and Control 69 (2021) 102861 Available online 18 June 2021 1746-8094/© 2021 Elsevier Ltd. All rights reserved. Estimated ECG Subtraction method for removing ECG artifacts in esophageal recordings of diaphragm EMG Annemijn H. Jonkman a, b, * , Ricardo Juffermans a, b , Jonne Doorduin d, e , Leo M.A. Heunks a, b , Jaap Harlaar c, f a Department of Intensive Care Medicine, Amsterdam University Medical Centers, location VUmc, Amsterdam, the Netherlands b Amsterdam Cardiovascular Sciences Research Institute, Amsterdam, the Netherlands c Department of Biomechanical Engineering, Delft University of Technology, Delft, the Netherlands d Donders Institute for Brain, Cognition and Behavior, Radboudumc, Nijmegen, the Netherlands e Department of Neurology, Radboudumc, Nijmegen, the Netherlands f Department of Orthopedics, Erasmus Medical Center, Rotterdam, the Netherlands A R T I C L E INFO Keywords: Diaphragm EMG ECG contamination Template subtraction Wavelet flter ABSTRACT The accuracy of diaphragm electromyogram (EMGdi) derived parameters, as used in critically ill intensive care unit (ICU) patients, can be compromised due to electrocardiographic (ECG) interference in the EMGdi signal. Removal of ECG contamination from the esophageal recordings of the EMGdi is challenging due to spectral overlapping of EMG and ECG signals and because of variability in ECG shape and amplitude. Therefore, we designed an Estimated ECG Subtraction (EES) method, based on three steps: (1) identifcation of the timing of the ECG artifact without an ECG reference channel, (2) estimation of the normalized ECG, considering the EMGdi as noise, and (3) subtraction of the denormalized ECG estimate from the EMGdi recordings. We evaluated the EES method against the use of a single wavelet-based adaptive flter. Using EMGdi signals of ten ICU patients and simulated contaminated EMG, we demonstrated that the EES method yields uncontaminated EMGdi, and showed that it is more effective than a wavelet-based adaptive flter only. Implementation of this technique may offer means to improve diaphragm activity monitoring and control in clinical practice. 1. Introduction The diaphragm is the most important respiratory muscle. Monitoring diaphragm activity in mechanically ventilated intensive care unit (ICU) patients is performed to facilitate diaphragm-protective ventilation, to assess patient-ventilator interaction and work of breathing, as well as to identify neuromuscular dysfunctions [1–4]. Bedside monitoring of dia- phragm electrical activity (EAdi) is available on a specifc ventilator (Getinge, Sweden) via a dedicated nasogastric (feeding) tube embedded with multiple ring-shaped electrodes positioned at the level of the dia- phragm [5,6]. EAdi refects the spatial and temporal recruitment of the crural diaphragm motor units, and is the closest available signal to the neural respiratory center output [5,7]. The EAdi catheter was originally designed to control the timing and level of ventilator pressurization in neurally adjusted ventilatory assist (NAVA) mode [5], but can also be used to monitor diaphragm activity in other ventilator modes or with unassisted breathing [1]. Signal processing algorithms within the ventilator continuously select the electrode pair closest to the dia- phragm and flter out interferences, such as cardiac electrical activity (ECG) and motion artifacts due to cardiac contractions and esophageal peristalsis. However, we found that the reliability of EAdi-derived pa- rameters to monitor diaphragm activity is compromised by these ventilator signal processing algorithms [8,9]. Furthermore, we demon- strated that ineffective fltering and inadequate removal of QRS com- plexes that interfere with the raw diaphragm electromyogram (EMGdi) limits interpretation of patient-ventilator interaction and detection of neural inspiration onset [10]. Improved fltering methods are needed for optimal use of the EMGdi – and its processed EAdi signal – in clinical decision-making and research. Removal of ECG contamination from any type of EMG is a major challenge because of spectral overlapping of the ECG (0–100 Hz) and the EMG (20–250 Hz, but mostly <150 Hz) [11], causing an increase in power content of the EMG and distortion of its frequency information. Different methods for removal of ECG interference from the EMGdi have * Corresponding author at: Amsterdam UMC, Location VUmc, de Boelelaan 1117, 1081 HV Amsterdam, The Netherlands. E-mail address: ah.jonkman@amsterdamumc.nl (A.H. Jonkman). Contents lists available at ScienceDirect Biomedical Signal Processing and Control journal homepage: www.elsevier.com/locate/bspc https://doi.org/10.1016/j.bspc.2021.102861 Received 4 January 2021; Received in revised form 18 May 2021; Accepted 7 June 2021