Dr. Prima Dewi Purnamasari, S.T., M.Sc.
All signals originating from and/or acquired from living organisms can be classified as physiological signals or biomedical signals. These signals may originate at the molecular, cellular, or systemic/organ level. Examples of biomedical signals include the electrocardiogram (ECG), which reflects the electrical activity of the heart; speech signals; the electroencephalogram (EEG), which represents the electrical activity of the brain; the electromyogram (EMG), which measures the electrical activity of muscles; and the electroretinogram (ERG), which captures electrical responses from the retina, among others.
The potential applications of physiological signals remain extensive for both medical and non-medical purposes. Medical applications include the detection of conditions such as sleep apnea and cardiac arrhythmias using ECG signals. Non-medical applications include the detection of stress, meditation states, and driver drowsiness using ECG data, as well as emotion recognition and alcohol impairment detection using EEG signals.
Machine learning methods are increasingly being developed for a wide range of applications, particularly in the healthcare sector. Their primary objective is to assist healthcare professionals in patient diagnosis, monitoring, and decision-making. One of the major challenges in applying machine learning to physiological signals is identifying appropriate and informative features. This requires advanced signal processing techniques to extract meaningful characteristics from the raw data. Although physiological signals vary in their sources and applications, they share a common fundamental structure as time-varying signals with amplitudes that change over time. Consequently, feature extraction methods developed for one type of physiological signal may also be adapted and applied to other signal modalities.
This research focuses on the development of machine learning algorithms and feature extraction techniques for human physiological signals. To support these efforts, access to high-quality physiological signal datasets with reliable ground-truth labels is essential. Such datasets are typically provided by physicians, hospitals, healthcare institutions, or other trusted data owners. In addition, physiological signals can be integrated with tabular and contextual data to enrich the information available for analysis, improve model performance, and enable more comprehensive data-driven insights.