Biomedical signals obtained from the human body can be beneficial in a variety of scenarios in a healthcare setting. For example, physicians can use the noninvasive sensing, recording, and processing of a heart’s electrical activity in the form of electrocardiograms (ECGs) to help make informed decisions about a patient’s cardiovascular health. A typical biomedical signal acquisition system will consist of sensors, preamplifiers, filters, analog-to-digital conversion, processing and analysis using computers, and the visual display of the outputs. Given the digital nature of these signals, intelligent methods and computer algorithms can be developed for analysis of the signals. Such processing and analysis of signals might involve the removal of instrumentation noise, power line interference, and any artifacts that act as interference to the signal of interest. The analysis can be further enhanced into a computer-aided decision-making tool by incorporating digital signal processing methods and algorithms for feature extraction and pattern analysis. In many cases, the pattern analysis module is developed to reveal hidden parameters of clinical interest, and thereby improve the diagnostic and monitoring of clinical events.
The methods used for biomedical signal processing can be categorized into five generations. In the first generation, the techniques developed in the 1970s and 1980s were based on time-domain approaches for event analysis (e.g., using time-domain correlation approaches to detect arrhythmic events from ECGs). In the second generation, with the implementation of the Fast Fourier Transform (FFT) technique, many spectral domain approaches were developed to get a better representation of the biomedical signals for analysis. For example, the coherence analysis of the spectra of brain waves also known as electroencephalogram (EEG) signals have provided an enhanced understanding of certain neurological disorders, such as epilepsy. During the 1980s and 1990s, the third generation of techniques was developed to handle the time-varying dynamical behavior of biomedical signals (e.g., the characteristics of polysomnographic (PSG) signals recorded during sleep possess time-varying properties reflecting the subject’s different sleep stages). In these cases, Fourier-based techniques cannot be optimally used because by definition Fourier provides only the spectral information and doesn’t provide a time-varying representation of signals. Therefore, the third-generation algorithms were developed to process the biomedical signals to provide a time-varying representation, and clinical events can be temporally localized for many practical applications.
These algorithms were essentially developed for speech signals for telecommunications applications, and they were adapted and modified for biomedical applications. The nearby figure illustrates an example of knee vibration signal obtained from two different knee joints, their spectra, and joint time-frequency representations. With the advancement in computing technologies, for the past 15 years, many algorithms have been developed for machine learning and building intelligent systems. Therefore, the fourth generation of biomedical signal analysis involved the automatic quantification, classification, and recognition of time-varying biomedical signals by using advanced signal-processing concepts from time-frequency theory, neural networks, and nonlinear theory.
During the last five years, we’ve witnessed advancements in sensor technologies, wireless technologies, and material science. The development of wearable and ingestible electronic sensors mark the fifth generation of biomedical signal analysis. And as the Internet of Things (IoT) framework develops further, new opportunities will open up in the healthcare domain. For instance, the continuous and long-term monitoring of biomedical signals will soon become a reality. In addition, Internet-connected health applications will impact healthcare delivery in many positive ways. For example, it will become increasingly effective and advantageous to monitor elderly and chronically ill patients in their homes rather than hospitals.
These technological innovations will provide great opportunities for engineers to design devices from a systems perspective by taking into account patient safety, low power requirements, interoperability, and performance requirements. It will also provide computer and data scientists with a huge amount of data with variable characteristics.
The future of biomedical signal analysis looks very promising. We can expect innovative healthcare solutions that will improve everyone’s quality of life.
Sridhar (Sri) Krishnan earned a BE degree in Electronics and Communication Engineering at Anna University in Madras, India. He earned MSc and PhD degrees in Electrical and Computer Engineering at the University of Calgary. Sri is a Professor of Electrical and Computer Engineering at Ryerson University in Toronto, Ontario, Canada, and he holds the Canada Research Chair position in Biomedical Signal Analysis. Since July 2011, Sri has been an Associate Dean (Research and Development) for the Faculty of Engineering and Architectural Science. He is also the Founding Co-Director of the Institute for Biomedical Engineering, Science and Technology (iBEST). He is an Affiliate Scientist at the Keenan Research Centre at St. Michael’s Hospital in Toronto.