Spectrotemporal Modeling of Biomedical Signals: Theoretical Foundation and Applications

Abstract

Biomedical signal analysis allows us to extract meaningful information from biological processes, thus enabling the assessment, characterization, and understanding of their originating mechanisms. Biomedical signals, however, are nonstationary and have statistics that change over time. As a result, conventional frequency-domain signal analyses have their limitations and a more powerful analysis technique is needed, particularly one capable of characterizing the changes in spectral content over time. In this article, we propose one such spectrotemporal representation—the modulation spectrogram. We start by presenting the theoretical foundation behind the technique and compare three of the most utilized approaches to calculate the spectrotemporal representation, namely the short-time Fourier transform, the continuous wavelet transform, and the Hilbert transform. An open-source amplitude modulation analysis toolbox is presented to allow the reader to explore amplitude modulation analysis of different biomedical signals. Lastly, to illustrate the advantages of the modulation spectrum analysis over conventional frequency-domain tools, several biomedical applications are described, ranging from detecting breathing rate from ECG signal, to improved Alzheimer's disease diagnosis using novel features extracted from electroencephalograms, to artifact removal of wearable electrocardiograms. It is hoped that this article and its companion open-source toolkit will allow readers to quickly witness the advantages of the described spectrotemporal representation and explore novel biomedical signal analysis applications.