Conventional biosensor technology is often based on the acquisition of a few measured variables that are easy to acquire from the human body from a technical perspective. Based on these measurement signals, the parameters of interest from a medical point of view are then derived either directly by the medical personnel or with the help of signal processing steps. As a result, the information available for diagnostics or therapy assessment is limited.
The aim of intelligent biosensor technology is to optimize conventional measurement methods regarding patient-individualized information acquisition without causing discomforts for the patient or the medical staff. To achieve this, the measurement problems must first be analyzed in detail and new concepts for signal acquisition must be developed. These concepts include the physical acquisition of signals as well as electronic and subsequent digital signal processing.
Multi-channel sensors, for example, have high clinical potential as they are redundant against failure of individual channels and can provide a comprehensive view of the patient in the clinic. However, in addition to the technical difficulties of their implementation, the processing of high-dimensional multivariate/multimodal time signals from these sensors also poses a challenge.
At Fraunhofer IMTE, various measurement techniques and systems are currently under development, with a focus on bioelectrical approaches such as electromyography or bioimpedance analysis. These approaches are realized by designing and testing individual sensors and electronic measurement circuits in consideration of the respective standards. This also includes innovative methods of signal processing and sensor fusion based on modern methods of machine learning, for example probabilistic models such as spatio-temporal Gaussian processes, and deep convolutional neural networks (CNNs). A special focus is the efficient execution of inference on edge devices, which is achieved by structure exploitation and hardware acceleration.