We investigated the neural networks involved in decision making under uncertainty during the so-called “Jumping to Conclusions” (JTC) paradigm. Functional imaging showed an extended executive cognition network of sustained activation during reasoning, while ventral striatal regions, which have been associated with saliency 
, were activated more strongly during the final stage of the particular decision as compared to its initiation.
In a behavioral perspective, participants asked for a comparable amount of evidence before coming to a conclusion to what has been observed in healthy subjects in previous studies 
. As expected, DTD depended strongly on the order of stimulus presentation. There was a trend towards fewer DTD in the monetary incentive version possibly due to the fact that subjects had to “pay” for each additional fish drawn, but the difference to subjects performing the classical version of the test was not significant. Together with the absence of group differences in confidence ratings, this pattern of performance supports the assumption that subjects did not “jump to conclusions” out of lack of motivation and to shorten the experiment. Furthermore, it may indicate that incentive motivation contributes only marginally to the activation patterns.
The task-specific recruitment of neural structures involved an extended network associated with executive functioning: right ventrolateral and bilateral dorsolateral prefrontal cortex, superior and inferior parietal lobule and precuneus, premotor regions and pre-SMA 
Although it cannot be excluded that the network activated also reflects other processes that differ between the beads task and the control task but is independent of the decision making process (e.g. increased visual processing of additional information, increased attentional demands or reading of the lake proportions etc. ), it is interesting that these regions were also found to be involved in the rewarded beads task used by Furl and Averbek 
. However some of these regions (ventrolateral prefrontal cortex, precuneus, pre–SMA were found to be more activated during draws than during decision in this study while others were found to be more activated during decision (posterior regions, insula). In addition, we confirmed activation in regions previously described by Blackwood and colleagues 
, medial occipital cortex and cerebellum. This previous study might have masked activation in the executive functioning network, in particular the DLPFC, because of differences in the control condition, where Blackwood and colleagues required participants to monitor the frequencies of stimuli and remember them during the whole block. This introduced a working memory component not present in our control condition which was to reach a decision under certainty (color of each stimulus) after each stimulus separately. Even if all information about previously collected data is visible at all times during the reasoning process, our data therefore support executive functions and working memory as a core component of this specific probabilistic reasoning test. The fact that none of these main regions of the executive functioning network were differentially activated at the time of the last versus all other stimuli indicates that executive control is required throughout the process of reaching a decision. Our findings show that differentiating between sustained and transient processes during the beads reveals an important network involved in the problem solving part of the task which cannot be clearly assigned to a particular process when using a block or event related design exclusively.
When analyzing the classical and the monetary incentive versions of the task separately, we did not find activation differences in the block design analysis. This corresponds to the absence of significant differences at the behavioral level, and supports the view that the experimental situation and the task by itself is motivating enough, at least for healthy controls, and subjects did not prematurely discontinue the trials to shorten the experiment. Usually, we would have expected increased activation in reward related brain regions such as ventral striatum and orbitofrontal cortex in the monetary incentive versus the nonincentive version of the task 
. The reason why we did not see such activation might be that participants did not receive feedback regarding the correctness of their decisions and therefore did not experience anticipation of reward during the presentation of task stimuli. In addition, differences might be covered by the reduced group size. Therefore it might have been preferable to expose all subjects to all conditions in a fully factorized design increasing the chance to detect effects of this manipulation.
In our second analysis, aimed at investigating activation at the moment of decision making, we found activation in structures linked to salience and dopaminergic neurotransmission nicely replicating the findings reported by Furls and Averbek using their rewarded version of the beads task 
. Regions that were more activated during the last stimulus that led to a decision compared to all preceding stimuli were bilateral striatum and midbrain including the ventral tegmental area, brain areas that are reliably activated in salience processing 
. This finding is in line with our hypothesis that salience processing might play a prominent role in decisions under uncertainty. The ventral striatal regions found here, were not only associated with salience but also with other processes like reward anticipation 1(e.g. 
). However, since we could not find differences between our rewarded and our unrewarded version in these regions, we conclude that the increased activation to the last fish is mainly driven by the acquired salience of this stimulus rather than its rewarding value. This conclusion is also in accordance with studies demonstrating the modulation of ventral striatal activation by salience, even in the context of monetary reward 
or aversive context conditioning 
Additionally, we found right VLPFC, pre-SMA, bilateral insula, medial occipital regions and bilateral thalamus to be transiently active. Interestingly and further supporting our conclusion of the role of saliency in our task, anterior insula is also implicated in the salience network 
, and seems to play a role in encoding uncertainty 
. Thalamic, especially medial regions (dorsomedial nucleus) have tight connections with prefrontal cortex and are relevant for the dopaminergic control of processing of sensory information 
Taken together, the increased activation in brain regions known to support dopaminergic and salience related cognitive functions, we speculate that the last fish before decision might constitute a highly salient signal with a marked subjective importance attributed to the provided color information, which would then ultimately trigger the a response to stop gathering evidence an come to a conclusion. In contrast, the block design analysis would capture activation related to the preparation of the decision during the whole block by a more cognitive process that recruits the executive functions network. The decision for on or the other lake would then not exclusively be based on ongoing cognitive calculations of probabilities as represented by prefrontal activation, but would at least partly be driven by a salience signal from the ventral striatum preceding the actual decision.
Menon and colleagues 
failed to see the typical behavioral JTC response pattern in schizophrenia patients in a version of the task where they showed a memory aid and postulated a possible influence of memory load on the distortion of the stimulus salience, although other authors have seen JTC bias even with no memory load 
. Here, he we found salience regions clearly activated at decisions without memory load. We found no significant correlations between brain activation and probabilistic reasoning styles, assessed by numbers of stimuli viewed before reaching a decision (DTD) and confidence ratings. Again this might be because we studied only healthy subjects without a broad enough range of behavioral differences.
When comparing the two versions of the task in the event-related analysis, we found more activation in the VTA in the classical version. Because at the moment of decision, the participants did not receive feedback about whether they actually gained or lost money, it cannot be excluded that this finding reflects relatively reduced importance of the salience system at the moment of decision in the monetary incentive version because of a sustained elevated level of reward-related activation of this system during the whole block of the JTC condition.
Our data show that, at least in healthy individuals, the JTC test activates regions implicated in salience processing and might provide a neural mechanism that could form a link between aberrant salience processing and the formation of metacognitive biases such as the JTC bias 
. To speculate further, aberrant and untimely spiking dopaminergic neurons might be a neurobiological correlate of false attribution of salience to stimuli relevant for decisions. In schizophrenia, because the dopamine system is dysregulated 
, this process might be chaotically upregulated which could be inferred from the altered activation of the dopaminergic midbrain and striatum during reward learning found in psychosis 
, and which might explain both the tendency for JTC and the marked heterogeneity of patients’ performance in the JTC test.
Our data leave the question open whether the metacognitive bias in JTC is a cognitive link from aberrant salience processing in schizophrenia to delusion formation 
. Behavioral evidence showing higher rates of JTC bias in delusional schizophrenia patients than in patients without delusions 
indicate that delusions and the JTC bias share variance. In addition, since our block design analysis found a tonic activation of prefrontal cortex during the decision process, it is possible that impairments in executive functioning and working memory, commonly found in schizophrenia, contribute to the JTC bias in this disorder, either independently or interacting with aberrant salience processing. Behaviorally, it has been shown that the JTC bias in schizophrenia is related to, but not completely dependent on, executive functioning, in particular mental flexibility 
. To test these relationships, additional studies with schizophrenia patients before and during antipsychotic treatment will have to be performed. The proposed mechanism might be specifically important in at-risk-mental states (ARMS), when first delusional symptoms are reported. The JTC neuroimaging paradigm and the differentiated analysis reported here should be an appropriate experimental approach to further our knowledge about the neurobiological underpinnings of this specific metacognitive deficit in schizophrenia.