One of the most pressing challenges in mining information from biological signals is dealing with the large amount of noise inherently present within them. Certain types of noise can be the results of predictable phenomena, such as powerline or sensor interference, and due to their predictability, they can be modeled and their effect remedied by various methods. Other noise, however, such as the motion artifact consistently present in bio-signal data, is completely random and unpredictable, and therefore much more difficult to remove. Because many complex feature extraction methods are highly sensitive to noise, the volume of data that can be extracted from the signal and its usefulness are greatly diminished by the presence of motion artifact and other stochastic noise.
The Automated Signal Quality Validation Framework is able to remedy this issue in a biosignal-agnostic way. A high-quality template that represents a discrete piece of the signal (a QRS complex in ECG, for example) is chosen and used as a model for the signal without noise. Using a matched filter, the template is used to evaluate its corresponding input biosignal, removing regions it identifies as pure noise and keeping regions that contain usable signal. The Framework has shown effectiveness on ECG and ABP signals, and due to its signal-agnostic nature, can be applied to almost any signal with repeating patterns. Removal of such noise opens the door to complex feature extraction, in turn producing more robust information and more impactful results.