Conventional biosensor technology is often based on the acquisition of a few measured variables that are technically easy to derive from the human body. From these measurement signals, the parameters that are interesting from a medical point of view are then either directly tapped by medical staff or determined with the help of signal processing. As a result, the information available for diagnostics or therapy planning is limited.
Intelligent biosignal acquisition aims to optimise conventional measurement procedures on a patient-specific basis with regard to information acquisition without restricting the patient or the medical staff. To achieve this, the measurement problems must first be analysed in detail and new concepts for signal acquisition must be developed. These concepts include both the physical derivation of signals and the electronic processing and subsequent digital signal processing.
Multi-channel systems, for example, offer a high clinical potential, as they are resistant against the failure of individual channels and can provide a comprehensive picture of the patient in the clinic. In addition to the technical difficulties that arise with this, the processing of high-dimensional multivariate/multimodal time signals also presents a challenge.
At the Fraunhofer IMTE, such measurement methods and systems are being developed with a focus on bioelectrical approaches such as electromyography or bioimpedance analysis. These are implemented by developing individual sensors and electronic measuring circuits, taking into account the respective standards. This is combined with innovative methods of signal processing and sensor fusion based on modern machine learning methods. These include probabilistic models, among others spatio-temporal Gaussian processes and deep convolutional networks (CNN). A special focus is the efficient execution of inference on edge devices, which is achieved through structuring and hardware acceleration.