Computer-aided diagnosis system for cardiac disorders using variational mode decomposition and novel cepstral quinary patterns

• Published in: Biomedical signal processing and control Vol. 81; p. 104509
• Main Authors: Khan, Muhammad Umar, Aziz, Sumair, Iqtidar, Khushbakht, Fernandez-Rojas, Raul
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2023
• Subjects: Cepstral quinary patterns; Computer-aided diagnosis; Heart diseases; ReliefF; SVM; Variational mode decomposition; Variational mode decomposition; Cepstral quinary patterns; SVM; Heart diseases; Computer-aided diagnosis; ReliefF
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2022.104509

Cardiac disorders cause a large number of human mortalities every year. This raises a sheer need for an early and accurate diagnosis of cardiac disorders to provide early meaningful intervention. Ischemic heart disease (IHD) and rheumatic heart disease (RHD) are the leading causes of heart failure. In this article, we proposed a novel framework for the classification of IHD and RHD using Pulse Plethysmograph (PuPG) signals obtained from a subject’s fingertip. The presented framework comprises a combination of variational mode decomposition (VMD), cosine-based soft segmentation, and novel cepstral quinary patterns (CQPs). The PuPG signals were first preprocessed through VMD by decomposing them in various modes. After an extensive time–frequency analysis, only relevant modes were selected and combined to reconstruct a preprocessed PuPG signal. The preprocessed signals were segmented through developed cosine-based soft segmentation to eliminate similar content in various classes. Features were extracted from the preprocessed signal using novel CQPs. CQPs were able to extract the hidden discriminative information about the disease through cepstrum transformed representation. The extracted CQP features were further reduced through the ReliefF ranking algorithm. The extracted reduced features were exposed to a range of well-known classification methods such as Support Vector Machines (SVM) with linear and non-linear kernels, Ensemble classifiers, and K-nearest neighbors. SVM-Gaussian (SVMG) provides the best performance of 99% accuracy using 10-fold cross-validation. The proposed CQPs were also compared with time, frequency, and cepstral features. Comparative analysis confirms that the proposed method outperforms the existing renowned techniques for the diagnosis of cardiac disorders. [Display omitted] •Classification of multiple cardiac disorders using PuPG signals is proposed.•A new dataset of PuPG signals was recorded from healthy, IHD, and RHD subjects.•Cepstral Quinary Patterns (CQPs) are proposed as a novel feature extraction method.•CQPs outperforms conventional time, frequency, and cepstral features.•SVM classifier yielded 99% accuracy with only five CQP features.