Iranian Journal of Medical Physics ijmp.mums.ac.ir A New Method to Improve Automated Classification of Heart Sound Signals: Filter Bank Learning in Convolutional Neural Networks Seyed Vahab Shojaedini 1 *, Sajedeh Morabbi 1 1. Department of Biomedical Engineering, Iranian Research Organization for Science and Technology, Tehran, Iran A R T I C L E I N F O A B S T R A C T Article type: Original Article Introduction: Recent studies have acknowledged the potential of convolutional neural networks (CNNs) in distinguishing healthy and morbid samples by using heart sound analyses. Unfortunately the performance of CNNs is highly dependent on the filtering procedure which is applied to signal in their convolutional layer. The present study aimed to address this problem by applying filter bank learning concept in CNNs. Material and Methods: In proposed method, the filter bank of CNN is updated based on a cross-entropy minimization rule to extract higher-level features from spectral characteristics of the heart sound signal. The deeper level of the extracted features in parallel with their spectral-based nature leads to better discrimination between healthy and morbid heart sounds. The proposed method was applied to three different heart sound datasets of PASCAL-A, PASCAL-B, and Kaggle, including normal and abnormal categories. Results: The proposed method obtained a true positive rate (TPR) between minimally 86% and maximally 96% (if FPR=0%) among all the examined datasets. In addition, the false-positive rate (FPR) was obtained as 7-8% (if TPR=100%) among the mentioned datasets. Finally, the accuracy was achieved in the range of 93- 98% when the FPR was 0% and within the range of 96-96.5% when the TRP was 100%. Conclusion: Increased TPR in the proposed method (96% for the proposed method vs. 87% for CNN) in parallel with a decrease in its FPR (7% for the proposed method vs. 10% for CNN) showed the proposed method's superiority against its well-known alternative in automated self-assessment of the heart. Article history: Received: Apr 05, 2019 Accepted: Sep 28, 2019 Keywords: Heart Sound Classification Deep Learning Neural Networks Self-Assessment ►Please cite this article as: Shojaedini SV, Morabbi S. A New Method to Improve Automated Classification of Heart Sound Signals: Filter Bank Learning in Convolutional Neural Networks. Iran J Med Phys 2020; 17: 331-339. 10.22038/ijmp.2019.38169.1489. Introduction Cardiovascular diseases are one of the main causes of death worldwide. Frequent evaluation of the cardiovascular system is necessary for early detection of the heart pathologic conditions and physical examination is the simplest way for such diagnosis [1]. Heart sounds provide important information about the performance of the cardiovascular system; therefore, playing an important role in the early detection of heart pathologic conditions, such as arrhythmias, valve disease, and heart failure [2]. The heart sounds in the traditional methods are analyzed by experts. The progress of this method is hampered because of its time-consuming nature and human errors [3]. Therefore, based on the computerized detection and classification, automated methods have been substituted the traditional methods regarding the analysis of the heart sound. Automated methods help to improve patient care by eliminating the need for a highly-skilled examiner [4]. This feature makes these methods an ideal candidate for the self-evaluation of at-risk patients. Furthermore, advances in heart sound processing have resulted in considerable improvements in the results' quality. These advantages turned the automated methods into a useful, cost-effective, and non-invasive diagnostic modality for cardiac pathology. One of the major challenges of automatic methods is distinguishing healthy and morbid samples, which is a well-known classification problem. Several methods have been proposed for such a classification as grouped into the categories described below. In several studies, the instance-based learning concept was incorporated in the procedure of distinguishing healthy and pathologic cardiovascular systems. For instance, the k-nearest neighbors classifier has been utilized for detecting morbid heart systems based on their sound signal [5]. In other researches, the support vector machine (SVM) approach was applied to classify heart sounds [6]. For instance, the least square SVM (LSSVM) method was utilized to distinguish normal and abnormal heart sounds. In another type of SVM-based methods, growing-time SVM was examined to classify pathological and healthy heart sounds [7]. Hidden Markov model (HMM) was applied for the heart sound *Corresponding Author: Tel: +982156276311; Email: shojadini@irost.ir