ABroAD: A Machine Learning Based Approach to Detect Broadband NIRS Artefacts

Artefacts are a common and unwanted aspect of any measurement process, especially in a clinical environment, with multiple causes such as environmental changes or motion. In near-infrared spectroscopy (NIRS), there are several existing methods that can be used to identify and remove artefacts to improve the quality of collected data. We have developed a novel Automatic Broadband Artefact Detection (ABroAD) process, using machine learning methods alongside broadband NIRS data to detect common measurement artefacts using the broadband intensity spectrum. Data were collected from eight subjects, using a broadband NIRS monitoring over the frontal lobe with two sensors. Six different artificial artefacts – vertical head movement, horizontal head movement, frowning, pressure, ambient light, torch light – were simulated using movement and light changes on eight subjects in a block test design. It was possible to identify both light artefacts to a good degree, as well as pressure artefacts. This is promising and, by expanding this work to larger datasets, it may be possible to create and train a machine learning pipeline to automate the detection of various artefacts, making the analysis of collected data more reliable.