Noninvasive Study of the Human Heart using Independent Component Analysis Yi Zhu, Tong Lee Chen, Wanping Zhang, Tzyy-Ping Jung, Jeng-Ren Duann, and Chung-Kuan Cheng, Fellow, IEEE Abstract—We develop a new approach to study the hu- man heart using independent component analysis. Electro- cardiogram (ECG) is an important tool for heart disease diagnosis. However, the normal 12-lead ECG can only ob- tain limited heart signals and mostly rely on trained and experienced medical doctors to perform the analysis. We have designed experiments so that high spatial of electronic signals can be recorded from human subjects. Indepen- dent component analysis is applied to the recorded signals to separate different components from the recorded waves. The various separated components are further analyzed by back-projecting their activities to the surface montage to ex- amine the properties of components. Experimental results show that this is a promising approach and is able to be extended to perform more sophisticated heart simulations. I. Introduction A computer model that can simulate the heart is a com- plex topic that medical researchers have been working for on decades. Being able to model a heart provides the abil- ity to diagnose heart diseases efficiently, allowing doctors to easily locate the problem or perhaps point out which part of heart wall might be failing. One ideal for the model is its ability to take measurements in a noninvasive manner. Not only is it more cost effective for patients, it is also much simpler and faster to prepare, setup and take mea- surements than the invasive counterpart in going through surgery. Although electrocardiogram (ECG) can take mea- surements of the heart noninvasively to record the amount of electrical force at a given point on the body, a healthy subject could have abnormal heart rhythm while a known heart diseased subject could have normal heart rhythm. Thus merely studying ECG readings is not enough in di- agnosis. To get a better understanding of how the heart works would require the ability to separate out the sources of the heart and understand how each source contributes to a pulsating heart. In this paper, we propose an approach to make use of Independent Component Analysis (ICA) techniques to an- alyze electronic heart signals, which are obtained by our experiments. ICA refers to a family of related algorithms that performs blind source separation when statistical in- dependence are taken into account[1]. Although multiple Yi Zhu, Wanping Zhang and Chung-Kuan Cheng are with the Department of Computer Science and Engineering, University of California, San Diego, La Jolla, CA, U.S.A. 92093–0404. Phone: +1 858 534 6184, e-mail: {y2zhu,w7zhang,kuan}@ucsd.edu Tong Lee Chen is with Intuit, Inc. Tzyy-Ping Jung and Jeng-Ren Duann are with the Swartz Center for Computational Neuroscience, Institute for Neural Computation, University of California, San Diego, 4150 Regents Park Row, Suite 320, La Jolla, CA, U.S.A. 92037. Phone: + 858 458 1927, email: {jung,duann}@sccn.ucsd.edu channel recordings were used by other researchers, such as [2][3][4], to study human heart activities, there are not many previous works trying to separate and study the dif- ferent independent components embedded in the recorded heart waves by high density electrodes. As we mentioned above, it is of importance to identify different underlying sources that generate heart waves, which is approached by ICA algorithms in this work. The experimental results show that different components of the P-wave, QRS com- plex and T-wave, can be clearly identified and their cor- responding back projections demonstrate different proper- ties, which is a clear sign to show that it is a more effective way to locate the underlying problems of human hearts than normal ECG. The rest of the paper is organized as follows: in next sec- tion, the ICA concept and algorithm will be presented; Sec- tion 3 will introduce how our experiments are conducted and their results; future work will be discussed in Section 4. II. Independent Component Analysis Our methods are based on the theory of ICA, which was originally proposed to solve the blind source separa- tion problem [5]. At that time, Comon was hoping that a defined mathematical framework would allow a baseline in further development of the ICA concept. The research ef- fort in ICA was considered small scale until the mid 1990s where it gained more attraction and popularity from Bell and Sejnowski’s infomax principle [1]. Jung et al. applied the ICA technique on many different types of biomedical signals including ECG, electroencephalogram(EEG), MEG (the magnetic counterpart to EEG), and functional mag- netic resonance imaging (fMRI) scans [6][7]. In the ECG study, Jung et al. took eight channels of measurements on the surface of a mother’s chest and abdomen. The channels were then processed using ICA to separate the maternal and fetal heart beats. According to their study, Jung et al. discovered that, although biomedical signals provide abundant information about the physiological pro- cess, they are often contaminated by distortions caused by small movements of the electrical contacts known as arti- facts. ICA, on the other hand, shows promise in separating artifacts from source signals, and perhaps can further sep- arate components into more fine grained subcomponents. There are also other previous works [8][9][10][11] that tried to use ICA to remove artifacts in the ECG recording. In this work, instead, we design more elaborate experiments to collect high density heart signals and use ICA to an- alyze underlying heart activities. We shall first introduce