Artificial Intelligence–Guided Behavioral Phenotyping in Epilepsy

Abstract

Abstract A major impediment to progress in basic epilepsy research is the fact that evidence-based, rigorous translational research is not only prohibitively time and labor-intensive, such as 24/7 video-electroencephalogram recordings, but rests on inherently subjective scoring by human observers, as exemplified by the Racine scale. Recent technical progress in machine learning and computer vision highlighted a variety of novel possibilities for quantifying behavior in animal models of epilepsies. This chapter briefly reviews the latest advances in artificial intelligence (AI)-guided animal motion tracking and segmentation of pose dynamics that bear great potential of revolutionizing behavioral phenotyping in basic epilepsy research. As an emerging field fueled by the recent successes of deep learning, AI-guided behavioral phenotyping will be discussed primarily in order to provide insights into the fundamentals of the field and at the same time raise awareness of potential pitfalls of the underlying technology. By concisely surveying the diverse and rapidly growing landscape of the relevant methods and toolboxes available in neuroscience research, this chapter aims to spark interest in AI-aided behavioral phenotyping in the epilepsy community.