Early detection of coronary artery disease in patients studied with magnetocardiography: An automatic classification system based on signal entropy Martin Steinisch a,f , Paul R. Torke a,b , Jens Haueisen b , Birgit Hailer c , Dietrich Gr ¨ onemeyer d , Peter Van Leeuwen d , Silvia Comani a,e,f,n a BIND—Behavioral Imaging and Neural Dynamics Center, ‘‘G. d’Annunzio’’ University, Via dei Vestini 33, 66013 Chieti, Italy b BMTI—Institute of Biomedical Engineering and Informatics, Ilmenau University of Technology, PF 100565, 98684 Ilmenau, Germany c Department of Medicine, Philippusstift, University Witten/Herdecke, H¨ ulsmannstr. 17, 43455 Essen, Germany d Department of Biomagnetism, Gr¨ onemeyer Institute of Microtherapy, University of Witten/Herdecke, Universit¨ atsstr. 142, 44799 Bochum, Germany e Department of Neuroscience and Imaging, ‘‘G. d’Annunzio’’ University, Via dei Vestini 33, 66013 Chieti, Italy f Casa di Cura Privata Villa Serena, Viale L. Petruzzi 42, 65013 Citt  a S. Angelo, Italy article info Article history: Received 16 June 2012 Accepted 22 November 2012 Keywords: Automatic classification system Magnetocardiography Entropy Coronary artery disease Early diagnosis abstract We propose an automatic system for the classification of coronary artery disease (CAD) based on entropy measures of MCG recordings. Ten patients with coronary artery narrowing Zor r50% were categorized by a multilayer perceptron (MLP) neural network based on Linear Discriminant Analysis (LDA). Best results were obtained with MCG at rest: 99% sensitivity, 97% specificity, 98% accuracy, 96% and 99% positive and negative predictive values for single heartbeats. At patient level, these results correspond to a correct classification of all patients. The classifier’s suitability to detect CAD-induced changes on the MCG at rest was validated with surrogate data. & 2012 Elsevier Ltd. All rights reserved. 1. Introduction One of the primary causes of death in adults in industrial countries is ischemic heart disease. The management of patients with unspecified chest pain requires that the cause be reliably identified while trying to avoid unnecessary invasive investiga- tion. Therefore, coronary angiography, which is the gold standard for the identification of a luminal narrowing of the coronary arteries, tends to be performed only in those cases when inter- vention is needed, to avoid serious side effects and high costs. As a consequence, the availability of non-invasive imaging techniques for the management of those patients is very attrac- tive and of increasing clinical importance. At present, the pre- dictive value of 12-lead ECG at rest, or exercise-ECG, stress echocardiography and nuclear imaging, is still low, whereas other techniques such as electronbeam computed tomography, mag- netic resonance imaging, or spiral computed tomography can be very discriminative. This is especially so when a combination of methods is used, such as exercise ECG combined with coronary calcium score and cardiac CT angiography, that reaches a 0.91 area under the curve (AUC) in ROC-analysis [1], or coronary CT angiography performed with ECG-gating [2], which obtains posi- tive and negative predictive values of 94% and 99%, respectively. However, these techniques tend to be time-consuming, expensive and in part associated with X-ray exposure or contrast agents. Body surface potential mapping (BSPM) definitely enhances the predictive power of ECG because of the increased number of registration sites on the thorax [3], but its extensive application in the clinic has been hindered by the duration and discomfort due to the positioning of 128 or more leads. During the last years several studies have underlined the potential of Magnetocardiography (MCG) as a non-invasive techni- que to detect coronary artery disease (CAD) in patients with chest pain (for an overview refer to [4]). MCG is a safe technique based on multiple superconducting sensors that permits the monitoring of the magnetic field variations associated with the spontaneous electrical activity of the heart from multiple sites above the chest. Current MCG systems work in magnetically shielded rooms, are able to detect very weak magnetic fields (of the order of 10 15 T) with no contact with the patient’s chest, and allow for very rapid exams (5–10 min). In contrast to ECG signals, MCG recordings are unaltered by the tissues surrounding the heart and are not affected by the problems related to bad lead contact, hence providing more Contents lists available at SciVerse ScienceDirect journal homepage: www.elsevier.com/locate/cbm Computers in Biology and Medicine 0010-4825/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.compbiomed.2012.11.014 n Corresponding author at: BIND - Behavioral Imaging and Neural Dynamics Center, ‘‘G. d’Annunzio’’ University, Via dei Vestini 33, 66013 Chieti, Italy. Tel.: þ3908713556901; fax: þ3908713556930. E-mail address: comani@unich.it (S. Comani). Please cite this article as: M. Steinisch, et al., Early detection of coronary artery disease in patients studied with magnetocardiography: An automatic classification system based on..., Comput. Biol. Med. (2012), http://dx.doi.org/10.1016/j.compbiomed.2012.11.014i Computers in Biology and Medicine ] (]]]]) ]]]–]]]