A key challenge in the computational analysis of EEG (Electroencephalography) data lies in identifying and mitigating artifacts - distortions in the signals that span spatial, frequency, and temporal scales. Defining what constitutes an artifact and determining the threshold for normal data is inherently complex and often depends on the method of data collection. As a result, semi-automated, human-in-the-loop approaches are commonly employed. These methods pre-screen EEG data automatically for artifacts, and domain experts subsequently correct the output. While effective, this process is time-consuming and labor-intensive.
The ArtiPlex project sets out to revolutionize the current artifact detection paradigm by integrating the results of multiple detection methods running in parallel. The key innovation lies in proposing “Multiplex Analytics” - a concept designed to progressively visualize the stream of multiple, automated detection results. This means that human analysts get no longer just the results of one computational analysis, but can observe potentially dozens or even hundreds of computational analyses and how they evolve (e.g., converge or diverge) over time. After all, modern multi core CPUs allow for many physical threads in parallel – so why not use them to get “a second opinion” computed by a completely different approach at the same time? And since dozens or even hundreds of parallel threads are possible, we might as well run dozens of “What-if” analyses to provide the human expert with a broad and nuanced ensemble of results. This approach allows experts to monitor and steer the detection processes effectively. Work on this project will be conducted in close collaboration with EEG specialists at Aarhus University’s Department of Engineering, as well as with researchers from abroad working in the emerging field of Progressive Visual Analytics.
Danmarks Frie Forskningsfond
