Categorization is considered a high level visual-perceptual function. It allows individuals to make reasonable guesses based on experience about the function and desirability of a stimulus or object. Monkeys, like humans, can classify whether images blending cats and dogs, so-called morphing from cat to dog are more cat-like or dog-like. In our experiments, monkeys were required to identify whether the morph being presented is closer to a cat or a dog in a sequential two-alternative forced choice task. Single unit recordings in regions of inferior temporal cortex--specifically area TE and rhinal cortex (Rh)--have implicated these regions in late-stage visual processes, such as categorization and stimulus-reward association, respectively. We have previously found that TE-lesioned monkeys are only mildly impaired in a perceptually difficult test of visual categorization (employing dog-cat morphs), and that the performance of Rh-lesioned monkeys is indistinguishable from that of controls. To study how this categorization takes place, our monkeys were trained to touch a bar to initiate a trial, and release the bar during one of two intervals;early (during the presence of a red central target) if they identified the stimulus as more cat-like, or late (following the transition of the central target to green) if the stimulus was more dog-like. Rewards were only delivered for correctly identifying a dog. In summary, if cat, release on red to skip the trial and move on to the next, if dog, release on green to get a reward. We measured the percentage of trials and reaction times when the monkey correctly indicated cat. As the morph level changes from pure cat towards dog, the number of times the monkey indicated a cat decreased sigmoidally. The sigmoid was also shifted toward dog, indicating a bias toward choosing dog - when very unsure, guessing dog prevents missing a potential reward, with a penalty of having to wait through the whole trial for nothing if the stimulus should have been identified as cat. As the morph progresses from all cat to equal cat/dog, the monkeys take longer to make a decision, i.e., the reaction times become longer. We model this behavior as a drift diffusion process. A point starts midway between 2 boundaries, and takes a random walk until it hits a boundary, which represents a decision of cat or dog. The drift velocity depends on the morph level presented. The model has an analytical solution. Here we use the experimentally measured frequency with which the monkey identifies a given morph as dog to determine the drift velocity, which is then used to predict the mean reaction time. We find that for morph levels close to a pure cat the experimental mean reaction times are similar to the theoretical ones. When the morph level is close to a 50:50 mixture, the experimental mean reaction times are faster than the theoretical predictions. It appears that for the easy morph levels the monkey reaches the boundary quickly and decides. For more ambiguous morphs the monkey takes more time to accumulate evidence. If reaching a bound takes too long, the monkey stops accumulating evidence and indicates dog - not wanting to miss a potential reward. The question now is whether there is another process apart from drift diffusion, also stochastic, governing how long the monkey will wait to accumulate evidence before simply terminating the trial by guessing dog. Now, we asked whether decreasing the information available by shortening the stimulus duration would affect any of our study groups: controls, or monkeys with either TE or rhinal removals. We trained the monkeys to touch a bar to initiate a trial, and release the bar during one of two intervals;early (during the presence of a red central target) if they identified the stimulus as more cat-like, or late (following the transition of the central target to green) if the stimulus was more dog-like. A correct response resulted in the delivery of a fixed-size liquid reward, an incorrect response led to a punishment time-out. We presented the morphed stimuli for durations of 25, 50, 100, 250 or 500 ms in an interleaved design. The stimuli appeared on a background of black and white noise. When the stimuli were removed, the background immediately reappeared;the reappearance of the visual noise appeared to mask the after-image. The accuracy with which control monkeys categorized the stimuli decreased as the stimulus presentation became shorter. The reaction times of control monkeys were indistinguishable across the different stimulus durations. Bilateral TE and Rh cortex removals had different effects: TE-lesioned monkeys made more incorrect categorization judgments, but made their decisions as quickly as controls. Rh-lesioned monkeys categorized as accurately as controls, but took longer to respond, that is, their reaction times were longer. The mild impairment in the performance of TE-lesioned monkeys is consistent with our previous results, and may reflect compromised perceptual or associative processes. The slower reaction times of the Rh-lesioned monkeys could be related to the decision process in some as yet undefined manner.
Showing the most recent 10 out of 14 publications
