Abstract:Atrial fibrillation (AF) is the most common cardiac disorder. It was reported that about 2.3 × 106 American adults have AF in 2001, and this number would increase to 5.6 × 106 by the year of 2050 [1]. AF can lead to stroke and other heart problems. The risk of stroke is increased fivefold in AF patients [2].There are mainly two types of AF: paroxysmal, where the arrhythmia terminates spontaneously after two episodes and persistent, when it sustains beyond 7 days [3]. Radiofrequency isolation of the pulmonary veins (PVs) has been successfully applied to the interventional treatment of paroxysmal AF [4]. However, efficacy of PV isolation is limited in persistent AF [5]. This suggests different mechanisms behind these two types of AF.To target ablation sites, a variety of methods have been proposed [6,7]. Among them, entropy is shown to have the potential of identifying the pivot based on intracardiac electrograms [8] and bipolar electrograms [9]. These findings suggest a potential role of entropy analysis in noninvasive cardiac imaging.Noninvasive cardiac imaging visualizes the electrical activity of the heart from body surface potential maps (BSPMs). The inverse solution of equivalent current density estimated from BSPMs can be used to obtain the activation sequence of the heart [10]. The equivalent current density includes the spatiotemporal information of AF. Entropy analysis on the equivalent current density may provide quantification of the degree of complexity spatiotemporally and detect the possible pivot locations in atria. In this original work, we apply Shannon entropy (SE) to the inverse problem in a paroxysmal AF patient.BSPMs were performed in the AF patient. 208 conductive-carbon surface electrodes arranged in 20 strips (ActiveTwo system, BioSemi V.O.F., Amsterdam) were placed over the body surface with 144 electrodes on the chest and 64 electrodes on the back. All the body-surface electrodes were referenced to the Wilson central terminal. An electromagnetic 3D digitizer (Fastrak, Polhemus Inc., Colchester, VT) was used to locate the body-surface electrode positions on the chest when the patient lies in the supine position and on the back when the patient sat up after the BSPM recording.The heart–torso geometries were constructed from a thoracic computer tomography (CT) scan and a contrast-enhanced cardiac CT angiography scan. A boundary element model including left atrium (LA) and right atrium (RA) was built based on the cardiac CT angiography scan, which contains 3000–4000 boundary elements. A torso model including the torso, lungs, and heart was also constructed.The atrial signal was contaminated with three types of noise: baseline wander noise, high frequency noise, and ventricular far field noise. Passing the signal through a high-pass filter and a low-pass filter eliminated the first two. Ventricular far field noise was removed by QRST (the time from the beginning of the Q wave to the end of the T wave) subtraction in BSPMs. A QRST template of one channel was calculated by averaging all the QRST segments in this channel. We obtained the cleaned BSPMs by subtracting QRST templates from the original electrocardiogram (ECG) recording in each channel, which was used to estimate the equivalent current density afterward.Based on bidomain theory [11], the discrete cellular architecture can be represented by a macroscopic continuum model which consists of the intracellular and extracellular domains. These two domains are coupled by the continuity of transmembrane currents flowing from one domain to the other. The electrical field within the torso model is governed by(1)∇·[(Gi+Ge)∇∅⁣e]=∇·(-Gi∇∅⁣m)where Gi and Ge are the intracellular and extracellular effective conductivity tensors, Øe is the extracellular potential, and Øm is the transmembrane potential confined to a volume. Define equivalent current density j→eq as shown in the below equation:(2)j→eq=-Gi∇∅⁣mso Eq. (1) can be rewritten as follows:(3)∇·[(Gi+Ge)∇∅⁣e]=∇·j→eqEquation (3) demonstrates that j→eq can serve as the equivalent current sources for the generation of field potential. So body-surface potentials are linearly related to the equivalent current density distribution. By combining the cleaned BSPMs and boundary element models of atria and torso, the current density distribution was solved using the curry software package (Compumedics, Charlotte, NC).The degree of complexity of atrial electrical activity was measured by SE [12]. The SE is defined as the average value of logarithms of the probability density function(4)SE=-∑i=1Mp(i)lnp(i)where M is the number of bins that current density falls in, and p(i) is the probability density of each bin.Higher entropy values are positively associated with the complexity of the atrial electrical activity. For normal tissues, they are potentially influenced by multiple rotors. However, inside a rotor, the electrical activity is regionally stable and regular. Following this reasoning, we expect lowest SE values in pivot locations surrounded by high entropy value regions. So the location with lowest SE value is defined as the pivot location.We evaluated the entropy-based noninvasive imaging method qualitatively by plotting the SE map and identifying potential pivot locations in a paroxysmal AF patient.Figure 1 shows the SE map for a paroxysmal AF patient. The length of data used for analysis is 5 s. The detected pivot location is the area between left PVs and right PVs.The detected pivot location is in concordance with clinical observations that isolating PVs terminated the paroxysmal AF [4]. This suggests that the entropy-based noninvasive cardiac imaging has the potential of identifying regions of complex electrical activities which may be related with waves pivoting and be potential target for ablation procedure in the paroxysmal AF patients. In the future, we will apply our method to more paroxysmal AF patients and persistent AF patients. Also, computing SE maps in a moving time window would provide the dynamic information about the pivot locations.The authors would like to thank Zhaoye Zhou for her help in data collection and analysis. This work was supported in part by NIH HL080093.

An Entropy-Based Noninvasive Cardiac Imaging of Atrial Fibrillation1