Quantifying individual differences in eye-gaze dynamics
We know some people, such as those with ADHD and autism, tend to gaze at images differently. But at the individual level, could everyone have their “own” way of looking? Can we use this to improve diagnoses and better understand visual search?
Eye-tracking data is generally analyzed by dividing trajectories into periods of relative stillness (fixations) and sudden jumps to new visual targets (saccades). The most common algorithms for this use ad-hoc and approximate thresholds for velocity and acceleration. Recent results indicate that eye trajectories are well described by Markovian (i.e. memoryless) probabilistic switching between saccades and fixations (Lencastre et al., in preparation). In this case, thresholds can be empirically estimated by optimising the transition matrix between fixations and saccades (Krivan 2022). This project, part of the Virtual Eye project at Simula and OsloMet, will investigate the potential of the parameters obtained using Markov model fitting to help identify conditions such as autism and ADHD (Lencastre et al. 2024, Papanikolaou et al. 2024). Another potential use of this technique may be to control for physiological variation between subjects, allowing for clearer identification of remaining behavioural and strategic variation in visual search (Szorkovszky et al., in preparation).
Goal
The goal of the thesis is to answer one or more of the following questions, depending on the length of the thesis:
- Does the transition matrix (k-ratio) method produce a reliable measure of individual variation? This will be verified using multiple saccade classification algorithms as well as multiple time-periods per individual.
- Can individual thresholds contribute to diagnoses of ADHD and autism?
- Do individual thresholds improve the characterization of visual search strategy?
Learning outcome
The student will gain experience working with experimental data, mathematical modelling and model fitting, providing several components of the basic toolkit of data science. Many of the statistical techniques used are well suited to clinical applications.
Qualifications
The student should be comfortable coding in Python with numpy and have general knowledge of statistical inference. Experience with mixed-effect modelling is an advantage.
Supervisors
- Alexander Szorkovszky
- Pedro Lind
Collaboration partners
- OsloMet AI Lab
References
- Krivan, J.J. (2022). Dynamic Classification of Eye-Tracking Movements: Usage and Ranking. Master thesis. Oslo Metropolitan University.
- Lencastre, P., Lotfigolian, M., & Lind, P. G. (2024). Identifying Autism Gaze Patterns in Five-Second Data Records. Diagnostics, 14(10), 1047.
- Papanikolaou, C., Sharma, A., Lind, P. G., & Lencastre, P. (2024). Lévy Flight Model of Gaze Trajectories to Assist in ADHD Diagnoses. Entropy, 26(5), 392.
