RT Journal Article SR Electronic T1 Definition, modeling and detection of saccades in the face of post-saccadic oscillations JF bioRxiv FD Cold Spring Harbor Laboratory SP 2021.03.24.436800 DO 10.1101/2021.03.24.436800 A1 Richard Schweitzer A1 Martin Rolfs YR 2021 UL http://biorxiv.org/content/early/2021/03/25/2021.03.24.436800.abstract AB When analyzing eye tracking data, one of the central tasks is the detection of saccades. Although many automatic saccade detection algorithms exist, the field still debates how to deal with brief periods of instability around saccade offset, so-called post-saccadic oscillations (PSOs), which are especially prominent in today’s widely used video-based eye tracking techniques. There is good evidence that PSOs are caused by inertial forces that act on the elastic components of the eye, such as the iris or the lens. As this relative movement can greatly distort estimates of saccade metrics, especially saccade duration and peak velocity, video-based eye tracking has recurrently been considered unsuitable for measuring saccade kinematics. In this chapter, we review recent biophysical models that describe the relationship between pupil motion and eyeball motion. We found that these models were well capable of accurately reproducing saccade trajectories and implemented a framework for the simulation of saccades, PSOs, and fixations, which can be used – just like datasets hand-labelled by human experts – to evaluate detection algorithms and train statistical models. Moreover, as only pupil and corneal-reflection signals are observable in video-based eye tracking, one may also be able to use these models to predict the unobservable motion of the eyeball. Testing these predictions by analyzing saccade data that was registered with video-based and search-coil eye tracking techniques revealed strong relationships between the two types of measurements, especially when saccade offset is defined as the onset of the PSO. To enable eye tracking researchers to make use of this definition, we present and evaluate two novel algorithms – one based on eye-movement direction inversion, one based on linear classifiers previously trained on simulation data. These algorithms allow for the detection of PSO onset with high fidelity. Even though PSOs may still pose problems for a range of eye tracking applications, the techniques described here may help to alleviate these.Competing Interest StatementThe authors have declared no competing interest.