Anomaly detection concept for a non-invasive blood pressure measurement method in the ear

In this paper, a concept for automated anomaly detection for a new method of blood pressure measurement in the ear is presented. When the external auditory canal is closed off airtight, the enlargement of the arteries during the heartbeat causes a volume change of the closed air chamber and thus a pressure fluctuation. Pressure measurement in the ear results in a signal waveform of very small amplitude with respect to the pulse wave and in relation to the sensor noise of currently available absolute pressure sensors. Under real conditions, the useful signals are always exposed to interfering influences such as superimposed motion and environmental artifacts. This results in the necessity of an automatic artifact detection as an important requirement for the analysis of the biosignals in a non-laboratory environment. Different concepts for automated anomaly detection were investigated using a standardized test protocol with test subjects and evaluated regarding their suitability. Context signals were included in the analysis as well as statistical methods were applied to the signal itself. The approach using a one-dimensional convolutional neural network (1DCNN) achieved the best results with an average recognition rate of 79 %. However, the inclusion of acceleration data was identified as a promising addition in specific motion scenarios.