Tiffany Tai¹, Ning-Ting Hsu¹, Haley Frey², Katrina A. Wolters², Shao-Min (Sean) Hung³, Po-Jang (Brown) Hsieh¹, David Whitney²; ¹National Taiwan University
, ²University of California, Berkeley, ³Waseda Institute for Advanced Study, Waseda University

Dermatologists often scan across many lesions in rapid succession, forming quick impressions that guide which areas require closer examination. Such efficiency suggests that clinicians may depend on temporal ensemble perception—the ability to extract a quick overall sense of a feature from a sequence of stimuli rather than from detailed analysis of each item. Although this mechanism is well established for basic visual features, far less is known about whether observers can compute meaningful summaries for high-level categories that lack strict, explicit rules, such as skin lesion malignancy. Demonstrating temporal ensemble processing for malignancy would clarify how broad-area screening is achieved under tight time constraints and reveal both the strengths and limits of this strategy. This study examines whether observers form an ensemble malignancy from a brief, sequential stream of skin lesion images and whether their judgments reflect integration across all items rather than reliance on a randomly sampled subset. Participants (N = 5) viewed sequences of 1, 2, 4, or 6 lesions (600 ms total), sampled from image sets with known consensus malignancy ratings. After each sequence, they rated the average malignancy. Observers’ ratings in the whole-set condition (set size = 6) strongly correlated with the consensus average malignancy (mean Fisher’s z = 0.68), indicating reliable extraction of overall malignancy despite extremely limited exposure. Correlations increased systematically with set size, climbing from 1 to 6 images in a pattern well captured by a linear model (R² = 0.93). This improvement shows that observers integrated information across multiple images rather than basing judgments on a single randomly sampled lesion. These findings provide the first evidence that viewers can perceive average malignancy from rapidly presented skin lesions, suggesting that temporal summary mechanisms may contribute to efficient screening in dermatology and offering a foundation for understanding how such processes might aid—or bias—clinical decision-making.

Acknowledgements: supported in part by NIH R01CA236793