Cailey Collins¹ (cpcollins@mun.ca), Blaire Dube²; ¹Memorial University of Newfoundland and Labrador

Detection of abnormalities in medical imaging is essential for diagnosis, yet a substantial proportion of radiological errors arise because observers fail to inspect target-containing regions. Recently, segmented rapid serial visual presentation (RSVP; dividing a full image into centrally presented segments) has been proposed as a method to improve diagnostic search. By forcing observers to view every region in a controlled sequence, segmented RSVP aims to ensure full image coverage and reduce premature search termination. Although prior work demonstrates significant improvements in target sensitivity, the methods used ensured that targets always appeared intact within a single segment. In realistic image segmentation, diagnostically relevant features may instead be split across segments, potentially impairing target detection. The present study addresses this gap across two experiments. Experiment 1 tested whether fragmentation reduces target sensitivity. Participants performed an RSVP search for either a square among diamonds or a diamond among squares (intermixed). Each trial comprised a 24-frame search stream, with each frame containing four stimuli that were spatially jittered so that stimuli could appear partially outside the RSVP frame (edge-occluded) but never overlapped. 50% of trials contained a target; on full-target trials (25%), the target appeared within a single frame, on partial-target trials (25%), only half of the target was visible, with the other half occluded by a frame edge. Results showed a clear detection cost for partial targets. Experiment 2 tested whether repeated exposure to fragmented features—an inherent property of segmented RSVP—would offset this cost. Participants viewed streams in which fragmented targets appeared across two different frames. Although repetition improved overall sensitivity, a fragmentation cost remained, indicating that repeated exposure only partially compensates for missing diagnostic information. Thus, target fragmentation imposes meaningful perceptual costs in segmented RSVP search, highlighting constraints on RSVP-based diagnostic aids that should be considered in their design and implementation.

Acknowledgements: Natural Sciences and Engineering Research Council of Canada Discovery Grant