Medical biometrics constitute physiological signals of the human body that are typically used by physicians. Examples of medical biometric signals include the electrocardiogram (ECG), phonocardiogram (PCG), photoplethysmogram (PPG), electroencephalogram (EEG), pupillometry and other. It has been demonstrated that certain physiological signals have highly personalized characteristics and can constitute strong biometric modalities.
Like any other biometric modality, there are bodily and environmental factors that affect the medical biometric signals. For instance, exercise increases the heart rate with direct effects on the ECG, PPG and PCG signals. In order to adequately evaluate the performance of medical biometric systems, it is important to take into consideration this variability. The Biometrics Security Laboratory is pleased to release ten datasets of physiological signals under various conditions that need to be considered in biometric security settings.
1. UofT ECG Database (UofTDB). Recordings of 1020 subjects.
Recordings from 1020 subjects captured from fingertips similar to Lead I configuration. A subset of the subjects (<100) were recorded for up to six sessions over a period of six months, under different postures (sit, stand, supine and tripod) and exercise condition. The Vernier EKG sensors were used with a sampling rate of 200Hz. Each recording is between 2-5 minutes long.
2. Short term rest ECG database. Recordings of 52 healthy volunteers, twice over a month period.
For this experimental setup, two recording sessions took place several weeks apart, in order to investigate the permanence of the characteristics of the signal with time. A Vernier sensor was used to record resting Lead I ECG (1mV body potential / 1V sensor output). During the collection, the subjects were given no special instructions. Each recording is 3 minutes long, with sampling rate of 200 Hz.
3. Short term exercise ECG database. Recordings of 52 healthy volunteers (same as rest) , twice over a month period.
Using the same apparatus and sensor orientation as in the short term rest eCG database, this database includes ECG signals after physical exercise of the same subjects as before. Every recording is 3 minutes long.
4. Long term rest ECG database. 2 hour recordings of 10 healthy volunteers in a working environment.
In this experiment each recording session lasted 2 hours. The Hidalgo Equivital EQ-01 monitor (10-bit resolution) was used for the collection. This portable device allows for long term two-channel chest ECG monitoring and unobtrusive wireless transmission. To account for the particular emotional changes, no special instructions were given to the participants. After connection establishment and testing of the equipment, subjects signed consent forms and returned to their personal working stations, to continue their daily activities for 2 hours. ECGs were digitized at 256 Hz.
5. Combined ECG and PCG. Short recordings of 21 healthy volunteers
In this experiment, the two physiological signals (ECG and pPG) were collected simultaneously but without synchronization of the devices. PCG signals were collected using the Littmann Electronic Stethoscope Model 4100WS with a sampling rate of 8KHz and a 16 bit resolution. Every subject was recorded at rest 6 times, each for 8 seconds. The exact location of the stethoscope on the chest was decided based on the strength of the heart sound, while the recording position was speci?ed for each subject, so that it was not changed for multiple recordings.
Lead I ECG signals of the same subjects were collected using a Vernier sensor (1 mV body potential / 1 V sensor output). Each recording was 3 minutes long and the signals were digitized at 200 Hz. In this expeirment every subject was recorded only once.
6. Transient Otoacoustic Emissions (TEOAE). Recordings from 54 volunteers
In this expeiment, the Vivosonic integrity system was used.To ensure the quality of recording without too much constraint on environment, earmu?was used for noise canceling but the experiment was setup in an office space with regular amount of noise. The participants were instructed to sit in a chair and relax. 54 sub jects were successfully recorded during two both sessions, which where sceduled at least one week apart.
The volunteers were between the ages 20 and 30 with no hearing problems. Dataset consists of one response of length 17.2ms per ear per session for each subject.
7. Photoplethysmograph (PPG) Dataset (Biosec1) Recordings in four different settings to evaluate permanence, uniqueness and robustness of PPG signals as biometric modality.
This database employs the different protocol for data collection, compared to the our old version (Biosec1). It is crucial to achieve the time-stable features for robust authentication system.
