Authors :
Presenting Author: Forouzan Farahani, Ph.D. – New York University Langone Health
Jiyun Shin, PhD – New York University Langone Health
Simon Henin, Ph.D. – New York University Langone Health
Charan Ranganath, Ph.D. – University of California, Davis
Anli Liu, M.D. – New York University Langone Health
Rationale:
The human brain segments continuous experience during shifts in events (boundaries) or visual features (scene cuts). Event segmentation is critical to the formation of episodic memories, with evidence of hippocampal involvement from prior fMRI and SEEG studies.
We presented short films to surgical epilepsy patients, measuring eye movements and pupil size at event boundaries and scene cuts to understand the behavioral basis of hippocampal processing.
Methods:
One hundred healthy subjects were recruited via an online platform to annotate event boundaries in 15 short films. Boundaries corresponded to peaks in the agreement ratio. Boundary strength was rated as high, mid and low based on the ratios. Scene cuts were identified with the SceneDec.
Sixteen surgical epilepsy patients, aged 18-65 years watched films in 3 blocks of 5 (2-5 min each). Eye tracking and pupil size were recorded during film watching. Time-locked analyses of saccade probability, amplitude, velocity and pupil size were performed around both boundaries and cuts.
To address timing uncertainty in the annotated boundaries, we re-centered events to the closest cut times. We compared saccade amplitude, velocity and fixation duration within a 0.5 s window around each transition using FDR-corrected paired t-tests. The saccade rate was computed by calculating the saccade occurrence in 25-ms bins within a -250-ms to 750-ms peri-boundary window and tested against surrogate data to find above-chance bins.
Results:
Raters annotated 27.4 boundaries per film (mean segment 5.8 s). SceneDec found 658 cuts (43.8 ± 28.5 per film), with 76.4% overlap with event boundaries.
Re-centering boundaries within ±0.5 s on cut times revealed a robust saccade-probability elevation at 250–325 ms. By contrast, boundaries exceeding ±0.5 s from any cut showed no significant bins. Peri‐cut saccade probability around cuts also exceeded chance in 250-375 ms post‐cut.
At event boundaries, saccade amplitude increased from 0.30 ± 0.33 to 0.35 ± 0.32 (p < 0.001); velocity from 7.5 ± 6.0 to 8.3 ± 6.0 (p < 0.001); fixation duration dropped from 0.3 ± 0.2 s to 0.2 ± 0.3 s (p < 0.001). High and mid agreement boundaries showed the largest amplitude increases.