We developed a continuous integration task to measure visual temporal integration in the human visual system. Over the past decades, many tasks have been developed to this purpose. However, our novel task stands on its own for four main reasons: it has no requirements for fixation, it is based on continuous dynamic stimuli, it relies on continuous rather than dichotomous forced responses, and it is short in overall duration. We argue that for these reasons this continuous temporal integration task has overcome at least some of the limitations of the classical psychophysics procedures and is better suited for measurement of temporal integration windows in different populations.
Research on visual temporal integration has focused mostly on integration performance with respect to few brief stimuli rather than continuous dynamic ones (
Chota et al., 2021;
Deodato et al., 2024;
Lahkar et al., 2023;
Sharp, Melcher, & Hickey, 2018). Generally, integration tasks are characterized by presentation of two brief stimuli (<50 ms) separated by a variable inter stimulus interval (ISI) during fixation and participants are asked, more or less directly, if they were able to integrate them. For example, in the two-flash fusion task participants are presented with two consecutive flashes and they're asked to report if they were able to see both of them (segregation) or just one (integration) (
Deodato et al., 2024;
Deodato & Melcher, 2023b;
Samaha & Postle, 2015). Similarly, in temporal order and simultaneity judgement tasks, participants have to report the order of appearance of two stimuli or their simultaneity, under the assumption that incorrect responses are an indication of integration (
Chota et al., 2021;
Lahkar et al., 2023). Some studies have implemented a variant of the missing-element paradigm that consist of the presentation of two portions of a grid of stimuli divided between two different frames (
Di Lollo, 1977). In this case participants are asked to report either an odd element in the grid or a missing one, measuring respectively segregation or integration of the two frames across ISIs (
Santoni et al., 2024;
Sharp et al., 2018;
Sharp et al., 2019;
Wutz et al., 2018). However, the mechanisms of temporal integration operate continuously and over multiple time scales that are still under investigation (e.g.,
Vogelsang, Drissi-Daoudi, & Herzog, 2023). Thus it is not sufficient nor necessary to measure performance with respect to few stimuli presented for short periods of time to quantify the extent of the integration window. Accordingly, a recent study pointed that using only short ISIs (<100 ms) could bias interpretation of the results (
Menétrey, Roinishvili, Chkonia, Herzog, & Pascucci, 2024).
Additionally, in all these tasks performance is affected by state-dependent properties at the moment of stimulus onset, as well as evoked responses linked to an abrupt onset and then offset of a stimulus and also by spatial and temporal attentional factors such has momentary lapses and scanning of stimulus-irrelevant locations. A striking example of these issues is the difference between integration thresholds obtained with two flashes (i.e., two-flash fusion thresholds) and train of flashes (i.e., continuous flicker fusion) (
Maley, 1967), which led some to propose that they measure separate functions (
King, 1962). Moreover, early processing of a stimulus can interrupt the processing of subsequent or even precedent stimuli (
Herzog & Brand, 2015). Although these effects of visual masking have been extensively studied, their influence on visual temporal integration tasks is often not taken into account or considered as a mechanism of integration itself (
Menétrey et al., 2024). Finally, the presence of a gap between the two stimuli used in these tasks makes the process of integration more complex as It involves not only the integration of the stimuli but also consideration of the gap itself. This presents an additional challenge: whether the visual system should integrate the gap or exclude it from the integration process, and calls into question whether the nature of these tasks is only visual integration or also “gap detection.” Previous research has shown that introduction of a temporal gap in a plotting sequence actively interferes with integration of different stimuli to increase the perception of clear and distinct perceptual events (
Eriksen & Collins, 1968;
Hogben & Lollo, 1974). Indeed, it has been suggested that worse performance in the same task with gaps, versus without, could be a consequence of discontinuity detectors (
Hogben & Lollo, 1974). Typically, visual integration mechanisms operate on subsequent stimuli that are not separated by gaps, with some exceptions (e.g., occlusions or eye blinks). Therefore the natural tendency of the visual system to integrate contiguous stimuli is disrupted by artificial separations, potentially leading to less accurate and ecological valid measurements of temporal integration windows.
A notable exception to this issue is the Sequential Metacontrast task (SQM). In the SQM paradigm, a series of vertical lines is displayed, generating the illusion of two diverging motion streams. When one line has a horizontal vernier offset, it causes all the other lines to be perceived as having the same offset, despite being perfectly straight. However, if a vernier with an opposite offset is presented later in the stream the lines appear to be straight again. This is taken as evidence that the two offset are integrated and cancel each other (
Menétrey et al., 2023;
Vogelsang et al., 2023). However, integration in this task is based on moving stimuli and cannot be easily generalized since integration mechanisms responsible for the perception of motion have been shown to rely on at least partially different neural mechanisms and longer integration periods with respect to static integration (
Ronconi, Oosterhof, Bonmassar, & Melcher, 2017a).
In the continuous temporal integration task, a dynamic stimulus is presented until response and without any temporal gaps or blank frames. By avoiding fixation and the artificial segmentation introduced by ISIs, our task enables a more naturalistic and precise assessment of temporal integration windows. To compute the temporal integration function, we manipulated the amount of information available over a fixed period of time by updating bits of information about the stimulus more or less frequently (i.e., shorter or longer frame duration). This relies on the assumption that less information available in a given temporal integration window is characterized by worse detection performance. In other words, when the information is updated with a period that is longer than the temporal integration window, there is no integration. Consistent with our reasoning we found that performance reaches a negative plateau, which was accurately modelled by an exponential function.
A popular theory suggests that these integration mechanisms are implemented in the brain through neural oscillations. Specifically, the alpha rhythm (∼10 Hz) could represent cycles of integration which last ∼100 ms, such that stimuli that fall in the same cycle are integrated and otherwise segregated. Indeed, it has been found that integration/segregation of two flashes depends on the phase of alpha oscillations at stimulus onset (
Ronconi, Oosterhof, Bonmassar, & Melcher, 2017b;
Ronconi, Balestrieri, Baldauf, & Melcher, 2024) and individuals with faster alpha rhythms have shorter two-flash fusion thresholds (
Deodato & Melcher, 2023b;
Samaha & Postle, 2015). More generally, each alpha cycle could represent a window of evidence accumulation, consistent with the idea that temporal integration could serve as a mechanism for the summation of sensory evidence (
Tarasi & Romei, 2024). However, most of the studies testing this integration theory have used brief presentations of stimuli, with the limitations discussed above. Additionally, oscillations may interact with other neural mechanisms (
Deodato & Melcher, 2023a) and may not be directly responsible for the windows of integration, as integration windows larger than the alpha cycle have been reported (
Menétrey et al., 2024). For example, paradigms involving rapid serial visual presentation (RSVP), visual masking or missing element tasks consistently report different integration thresholds (
Karabay & Akyürek, 2017). Additionally, alternative accounts of performance in temporal integration tasks have postulated a decaying sensory trace that makes stimuli available in memory for a short time after their offset (
Di Lollo, 1977;
Eriksen & Collins, 1968;
Hogben & Lollo, 1974). Future research should confirm the relationship between continuous visual integration and rhythmic processes or decaying perceptual traces using a continuous task that does not include blank gaps.
In conclusion, the novel approach of our continuous integration task overcomes several limitations of traditional tasks that rely on presentation of discrete stimuli separated by gaps during enforced fixation. By using continuous dynamic stimuli and continuous response measures, our task is better suited for capturing the nature of temporal integration in the human visual system. Generally, abandoning the rigid trial structure of psychophysics allows for data collection in less-controlled settings. In this regard, our task shows promise for certain populations but also for online studies because the continuous presentation has fewer timing requirements, and it may be less influenced by dropped frames. Overall, this makes it a valuable tool for investigating visual temporal integration across different populations and conditions.