| OCR Text |
Show MANAGING LIMITED ECG ACCESS IN A CLINICAL CARDIAC ELECTROPHYSIOLOGY SETTING by Minna Wang A Senior Honors Thesis Submitted to the Faculty of The University of Utah In Partial Fulfillment of the Requirements for the Honors Degree in Bachelor of Science In Department of Bioengineering Approved: ______________________________ Dr. Rob MacLeod Thesis Faculty Supervisor _____________________________ Dr. Patrick Tresco Chair, Department of Bioengineering _______________________________ Dr. Kelly Broadhead Honors Faculty Advisor _____________________________ Sylvia D. Torti, PhD Dean, Honors College May 2016 Copyright © 2016 All Rights Reserved ABSTRACT Electrocardiographic Imaging (ECGI) is a computational approach that seeks to reconstruct cardiac electrical activity with high precision from body surface ECGs by solving a numerical inverse problem. Though there have been great advancements in ECGI, there is still a need for progress in specific applications, including discernment of ectopic foci and reconstruction of other arrhythmias. In order to improve and validate the localization of ectopic regions using ECGI, it is necessary to record high resolution ECGs known as “body surface potential maps” during stimulation of the heart from known ectopic foci. Such measurements typically occur in the catheterization lab during invasive procedures such as device placement or during diagnostic evaluation of cardiac electrical stability using catheter based stimulation electrodes. However, there are spatial constraints on electrode placement on the body, such as defibrillator patches (for patient safety) and sterile fields (for device implantation). To overcome the resulting limits in body surface electrode placement and coverage, one can use estimation algorithms to utilize spatial redundancy and draw equivalent data from reduced numbers of strategically place electrodes. In this study, we applied a body surface estimation and limited lead selection algorithm to body surface mapping data and determined a patient specific lead placement strategy to acquire body surface maps in ECGI verification studies. ii TABLE OF CONTENTS ABSTRACT ii INTRODUCTION 1 METHODS 2 RESULTS 4 DISCUSSION 7 REFERENCES 9 iii 1 INTRODUCTION It is important for physicians to pinpoint the abnormal regions of the heart in diagnostic assessments of cardiovascular symptoms such as myocardiac infarction (also known as heart attack) and cardiac rhythm disturbances (arrhythmias). Conventional electrocardiograms (ECGs) utilize only nine electrodes combined to form twelve leads and are thereby limited in their resolution and localization of problematic areas. This limitation is the rationale for body surface potentials mapping (BSPM), in which potentials are measured across the entire torso. BSPM also provides input data for a reconstruction approach known as ECG Imaging (ECGI) in which cardiac activity is derived from body surface potentials by solving what is known as the electrocardiographic inverse problem. The goal of the ECGI is to noninvasively identify abnormal electrical activity in the heart for clinical applications such as predicting sites of pre-excitation in Wolff-Parkinson-White patients [1] [2]. ECGI uses BPSM combined with geometric models of the thorax to compute the bioelectric sources and describe macroscopic cardiac electrical activity [3]. One specific and important example of a clinical application of ECGI is detecting point sources of excitation or “ectopic foci”, which cause an abnormal, premature heart beat. In order to improve the accuracy of ECG to localize ectopic focis, it is necessary to acquire body surface potential maps from known artificially stimulated regions of the heart, a procedure carried out in the safety of the cardiac catheterization laboratory, also known as the cath-lab. However, spatial constraints in cath-lab settings, e.g., sterile fields and defibrillation pads, restrict electrode placement during BSPM. 2 Previous studies have shown that a relatively small number of electrodes can produce body surface maps using spatial redundancy [4]. We used this approach, developed by Lux et al. to predict ECG values from unsampled regions for clinical electrophysiology from small numbers of strategically selected ECG electrodes. Successful application of this technology in cath-lab settings will allow for validation and improvement of ECGI and more precise localization of abnormal electrical activity (i.e. ectopic foci) than standard ECG. METHODS In order to validate the accuracy of the simulation, potential maps generated during ICD testing were acquired and compared to simulated values. To perform this validation, we first applied the limited lead selection and body surface estimation algorithm derived by Lux, et al. to determine the ideal lead set for surface recordings and to calculate a transformation to estimate the full potential map [4]. Then the resulting lead set and transformation were used to obtain estimated maps from recorded ICD testing data. Finally, transformed potential maps were compared to full recordings. Data used for transformation and testing came from one hundred of Dr. Bob Lux’s patients. One hundred 192-lead body surface potential maps (BSPMs) generated during ICD testing were obtained from Dr. Lux at the Nora Eccles Harrison Cardiovascular Research and Training Institute (CVRTI) at the University of Utah. Only the ventricular segments of the datasets were used. The 100 patient datasets were split into two groups: a training dataset (70) and a testing dataset (30). In order to mathematically identify and quantify the existing spatial 3 redundancies, the limited lead selection algorithm was performed on the training dataset with spatial constraints representative of cath-lab settings to yield the optimal lead set of 32 leads. The same training data was used to calculate the transformation matrix relating the lead location to the corresponding potential map. The Lux algorithm was applied to the data from the optimal 32 leads for each patient in the testing dataset, utilizing spatial redundancy to generate a full body surface map from a subset of leads [4]. In doing so, full BSPMs (e.g. 192-lead BSPMs) were simulated from the data given by the optimal 32 leads. Fig. 1. Application of the limited lead selection and the body surface estimation algorithm The estimated potential maps were compared to the simulated maps for each patient and evaluated using absolute error, correlation relative error, and relative root-mean squared (RMS) error. Qualitative comparisons were also made by placing images of simulated 4 maps and actual maps side by side in order to identify regions with high error. Values are presented as mean ± standard deviation. RESULTS Correlation and error metrics were recorded for increments of five between 5 and 50 leads, inclusive. (Fig. 2) For lead selection with theoretical constraints on the patient’s back, error decreased significantly from 5 to 10 leads (60% to 22% percent) and decreased approximately 10% percent error. Correlation increased from 0.13 from 5 to 10 leads, but increased only 0.06 from 10 to 50 leads. For each of the three metrics, there was a noticeable dip in correlation and a noticeable peak in RMS error and percent error at 20 leads. For lead selection without constraints, the slope of the error and correlation graphs changes significantly at 20 leads. (Fig. 2) Correlation increases 0.10 from 5 leads to 20 leads, but only increases 0.02 from 20 leads to 50 leads. A similar pattern is seen in percent error, where the error drops from 30% to 10% from 5 to 20 leads and decreases to only 8% at 50 leads. 5 Fig. 2. Metrics for varying numbers of leads used in reconstruction with constraints (left) and without constraints (right). The results show an exponentially decreasing error for an increased number of leads, although significant decreases were not present over 30 electrodes. The body surface estimation algorithm effectively generated qualitatively and quantitatively similar potential maps from the thirty-two chosen leads. A visual comparison of body surface maps created from the original 192-lead recordings and resulting torso maps from 32-lead testing datasets show error is localized at the center of the torso where proximity to leads is the lowest (Fig. 3) The thirty-two leads chosen by the algorithm were placed primarily on the front center of the torso, with seven on the left and right shoulders. The highest mean absolute error over the thirty testing recordings was on the posterior side of the torso directly behind the heart and on the central anterior side where electrodes were not placed. (Fig 4.) 6 Constrained data displayed lower correlation and higher error than unconstrained data. In constrained data, percentage error was 7.3% higher and RMS error was 5.2 uV higher, while correlation was 2% lower. (Table 1) Fig. 3. Body surface potential maps of 4 patients at the peak of the QRS complex of original (left) and reconstructed (right) potentials from the selected 32 leads. Optimal 32 lead locations, calculated from 70 BSPM recordings, are indicated by red markers on map. 7 Fig. 4. Lead location and mean absolute error map. The lead locations were chosen by the algorithm based on information index to achieve full reconstruction of surface potentials without leads that are inaccessible for clinical reasons. Correlation % Error RMS Error Constraints 0.919±0.015 20.4±5% 49.7±0.5uV No constraints 0.938±0.015 13.1±5% 44.5±0.5 uV Table 1. Comparison of metrics for 32 leads with highest information index. Values are mean± standard deviation. DISCUSSION This study utilized a body surface estimation algorithm adapted from Lux et al. in order to predict ECG signals from unsampled regions based on a small subset of the original BSPM electrode set. BSPM consisted of 192 leads worth of data that we could reconstruct from 32 leads identified using the algorithm. The results of this study will help guide research for more precise localization of ectopic foci than standard ECG in settings such as the catheterization lab or whenever access to the body surface is limited. 8 Our results demonstrate that the limited lead selection algorithm can be applied to ECG data with constraints and result in comparable body surface potential maps. Error is localized to the areas constrained by, in our case, the defibrillator patch and sterile field (Fig. 3) but is also acceptable for many applications, including ECGI. This localization can be explained by intrinsic properties of the algorithm, as any high information leads that may have fallen within the constrained areas were excluded from the final lead set. Not surprisingly, our results also showed that achieving the same levels of reconstruction accuracy required more electrodes in the spatially constrained setting then when the entire torso was accessible. (Fig. 2). The rate of data quality improvement overall was more rapid for the case of unconstrained access, but then reached comparable error metrics within 30 leads. The results of this study were similar to those of the original algorithm by Lux et al. In the findings of Lux et al. and also Barr et al., 30 leads were required to produce BSPMs with an accuracy of 32 μV [4] [5]. By comparison, RMS error for this study was approximately 10 μV higher for a similar number of selected leads (Table 1). The current algorithm is limited in its ability to account for the body shape of the patient and its effect on lead data collection. Future work includes research into the creation of an algorithm that generates a personalized lead set for a patient. By inputting the patient’s torso shape and spatial constraints, the program would be able to output a customized lead set resulting in maximum accuracy for BSPM. We also plan to generate generalized lead sets for certain subsets of the population who have larger risk of cardiac disease. In clinical applications, the identification capabilities of the algorithm will allow for more precise localization of electrically abnormal regions of the heart and more accurate 9 diagnoses using ECGI. The decreased number of leads necessary for an accurate body surface map due to the algorithm can also lead to decreased data costs and setup time. The successful use of the adapted algorithm in this application also supports its use in future studies to verify the accuracy of the cardiac inverse problem in identifying ectopic foci. REFERENCES [1] B. M. Burton et al., "A toolkit for forward/inverse problems in electrocardiography within the SCIRun problem solving environment," IEEE Eng Med Bio Soc, pp. 267-270, 2011. [2] R. S. MacLeod and D. H. Brooks, "Recent progress in inverse problems in electrocardiology," IEEE Eng Med Bio, vol. 17, no. 1, pp. 73-83, 1998. [3] R. M. Gulrojani, “The forward and inverse problems of electrocardiography,” IEEE Eng Med Bio, vol. 17, no. 5, pp. 84-101, Sep./Oct. 1998. [4] R. L. Lux et al., “Limited lead selection for estimation of body surface potential maps in electrocardiography,” IEEE Trans. Biomed. Eng., vol. BME-25, no. 3, pp. 270-276, May 1978. [5] R. C. Barr, et al., "Selection of the number and positions of measuring locations for electrocardiography." IEEE Trans. Biomed. Eng., vol. BME-18, No. 2, pp. 125-138, March 1971. |
