Relationship between prediction accuracy and feature importance reliability: An empirical and theoretical study.

There is significant interest in using neuroimaging data to predict behavior. The predictive models are often interpreted by the computation of feature importance, which quantifies the predictive relevance of an imaging feature. Tian and Zalesky (2021) suggest that feature importance estimates exhibit low split-half reliability, as well as a trade-off between prediction accuracy and feature importance reliability across parcellation resolutions. However, it is unclear whether the trade-off between prediction accuracy and feature importance reliability is universal. Here, we demonstrate that, with a sufficient sample size, feature importance (operationalized as Haufe-transformed weights) can achieve fair to excellent split-half reliability. With a sample size of 2600 participants, Haufe-transformed weights achieve average intra-class correlation coefficients of 0.75, 0.57 and 0.53 for cognitive, personality and mental health measures respectively. Haufe-transformed weights are much more reliable than original regression weights and univariate FC-behavior correlations. Original regression weights are not reliable even with 2600 participants. Intriguingly, feature importance reliability is strongly positively correlated with prediction accuracy across phenotypes. Within a particular behavioral domain, there is no clear relationship between prediction performance and feature importance reliability across regression models. Furthermore, we show mathematically that feature importance reliability is necessary, but not sufficient, for low feature importance error. In the case of linear models, lower feature importance error is mathematically related to lower prediction error. Therefore, higher feature importance reliability might yield lower feature importance error and higher prediction accuracy. Finally, we discuss how our theoretical results relate with the reliability of imaging features and behavioral measures. Overall, the current study provides empirical and theoretical insights into the relationship between prediction accuracy and feature importance reliability.

Overlapping attentional networks yield divergent behavioral predictions across tasks: Neuromarkers for diffuse and focused attention?

Attention is a critical cognitive function, allowing humans to select, enhance, and sustain focus on information of behavioral relevance. Attention contains dissociable neural and psychological components. Nevertheless, some brain networks support multiple attentional functions. In this study, we used the visual attentional blink (VAB) as a test of the functional generalizability of the brain's attentional networks. In a VAB task, attention devoted to a target often causes a subsequent item to be missed. Although frequently attributed to limitations in attentional capacity or selection, VAB deficits attenuate when participants are distracted or deploy attention diffusely. The VAB is also behaviorally and theoretically dissociable from other attention tasks. Here we used Connectome-based Predictive Models (CPMs), which associate individual differences in task performance with functional connectivity patterns, to test their ability to predict performance for multiple attentional tasks. We constructed visual attentional blink (VAB) CPMs, and then used them and a sustained attention network model (saCPM; Rosenberg et al., 2016a) to predict performance. The latter model had been previously shown to successfully predict performance across tasks involving selective attention, inhibitory control, and even reading recall. Participants (n â€‹= â€‹73; 24 males) underwent fMRI while performing the VAB task and while resting. Outside the scanner, they completed other cognitive tasks over several days. A vabCPM constructed from VAB task data (behavior and fMRI) successfully predicted VAB performance. Strikingly, the network edges that predicted better VAB performance (positive edges) predicted worse performance for selective and sustained attention tasks, and vice versa. Predictions from applying the saCPM to the data mirrored these results, with the network's negative edges predicting better VAB performance. The vabCPM's positive edges partially yet significantly overlapped with the saCPM's negative edges, and vice versa. Many positive edges from the vabCPM involved the default mode network, whereas many negative edges involved the salience/ventral attention network. We conclude that the vabCPM and saCPM networks reflect general attentional functions that influence performance on many tasks. The networks may indicate an individual's propensity to deploy attention in a more diffuse or a more focused manner.

Stepwise increasing sequential offsets cannot be used to deliver high thermal intensities with little or no perception of pain.

Offset analgesia (OA) is the disproportionate decrease in pain experience following a slight decrease in noxious heat stimulus intensity. We tested whether sequential offsets would allow noxious temperatures to be reached with little or no perception of pain. Forty-eight participants continuously rated their pain experience during trials containing trains of heat stimuli delivered by Peltier thermode. Stimuli were adjusted through either stepwise sequential increases of 2°C and decreases of 1°C or direct step increases of 1°C up to a maximum of 46°C. Step durations (1, 2, 3, or 6 s) varied by trial. Pain ratings generally followed presented temperature, regardless of step condition or duration. For 6-s steps, OA was observed after each decrease, but the overall pain trajectory was unchanged. We found no evidence that sequential offsets could allow for little pain perception during noxious temperature presentation.NEW & NOTEWORTHY Offset analgesia is the disproportionate decrease in pain experience following a slight decrease in noxious heat stimulus intensity. We tested whether sequential offsets would allow noxious temperatures to be reached with little or no perception of pain. We found little evidence of such overall analgesia. In contrast, we observed analgesic effects after each offset with long-duration stimuli, even with relatively low-temperature noxious stimuli.

Functional dissociation of the inferior frontal junction from the dorsal attention network in top-down attentional control.

The posterior lateral prefrontal cortex-specifically, the inferior frontal junction (IFJ)-is thought to exert a key role in the control of attention. However, the precise nature of that role remains elusive. During the voluntary deployment and maintenance of visuospatial attention, the IFJ is typically coactivated with a core dorsal network consisting of the frontal eye field and superior parietal cortex. During stimulus-driven attention, IFJ instead couples with a ventrolateral network, suggesting that IFJ plays a role in attention distinct from the dorsal network. Because IFJ rapidly switches activation patterns to accommodate conditions of goal-directed and stimulus-driven attention (Asplund CL, Todd JJ, Snyder AP, Marois R. Nat Neurosci 13: 507-512, 2010), we hypothesized that IFJ's primary role is to dynamically reconfigure attention rather than to maintain attention under steady-state conditions. This hypothesis predicts that in a goal-directed visuospatial cuing paradigm IFJ would transiently deploy attention toward the cued location, whereas the dorsal attention network would maintain attentional weights during the delay between cue and target presentation. Here we tested this hypothesis with functional magnetic resonance imaging while subjects were engaged in a Posner cuing task with variable cue-target delays. Both IFJ and dorsal network regions were involved in transient processes, but sustained activity was far more evident in the dorsal network than in IFJ. These results support the account that IFJ primarily acts to shift attention whereas the dorsal network is the main locus for the maintenance of stable attentional states. NEW & NOTEWORTHY Goal-directed visuospatial attention is controlled by a dorsal fronto-parietal network and lateral prefrontal cortex. However, the relative roles of these regions in goal-directed attention are unknown. Here we present evidence for their dissociable roles in the transient reconfiguration and sustained maintenance of attentional settings: while maintenance of attentional settings is confined to the dorsal network, the configuration of these settings at the beginning of an attentional episode is a function of lateral prefrontal cortex.

Functional Specialization and Flexibility in Human Association Cortex.

The association cortex supports cognitive functions enabling flexible behavior. Here, we explored the organization of human association cortex by mathematically formalizing the notion that a behavioral task engages multiple cognitive components, which are in turn supported by multiple overlapping brain regions. Application of the model to a large data set of neuroimaging experiments (N = 10 449) identified complex zones of frontal and parietal regions that ranged from being highly specialized to highly flexible. The network organization of the specialized and flexible regions was explored with an independent resting-state fMRI data set (N = 1000). Cortical regions specialized for the same components were strongly coupled, suggesting that components function as partially isolated networks. Functionally flexible regions participated in multiple components to different degrees. This heterogeneous selectivity was predicted by the connectivity between flexible and specialized regions. Functionally flexible regions might support binding or integrating specialized brain networks that, in turn, contribute to the ability to execute multiple and varied tasks.

Amodal processing in human prefrontal cortex.

Information enters the cortex via modality-specific sensory regions, whereas actions are produced by modality-specific motor regions. Intervening central stages of information processing map sensation to behavior. Humans perform this central processing in a flexible, abstract manner such that sensory information in any modality can lead to response via any motor system. Cognitive theories account for such flexible behavior by positing amodal central information processing (e.g., "central executive," Baddeley and Hitch, 1974; "supervisory attentional system," Norman and Shallice, 1986; "response selection bottleneck," Pashler, 1994). However, the extent to which brain regions embodying central mechanisms of information processing are amodal remains unclear. Here we apply multivariate pattern analysis to functional magnetic resonance imaging (fMRI) data to compare response selection, a cognitive process widely believed to recruit an amodal central resource across sensory and motor modalities. We show that most frontal and parietal cortical areas known to activate across a wide variety of tasks code modality, casting doubt on the notion that these regions embody a central processor devoid of modality representation. Importantly, regions of anterior insula and dorsolateral prefrontal cortex consistently failed to code modality across four experiments. However, these areas code at least one other task dimension, process (instantiated as response selection vs response execution), ensuring that failure to find coding of modality is not driven by insensitivity of multivariate pattern analysis in these regions. We conclude that abstract encoding of information modality is primarily a property of subregions of the prefrontal cortex.

Sleep deprivation reduces the rate of rapid picture processing.

Object recognition becomes impaired at faster presentation rates and here we show the neuroanatomical foci of where this might be exacerbated by sleep deprivation (SD). Twenty healthy human participants were asked to detect a target house in serially presented house pictures that appeared at 1-15images/s. Temporal response profiles relating fMRI signal magnitude to presentation frequency were derived from task-responsive regions. Following SD, the inverted U-shaped response profile within parahippocampal place area was lower and peaked at a slower presentation rate than when participants slept normally. Contrastingly, SD did not shift the relatively monotonic early visual cortex responses. The intraparietal sulci but not the frontal eye fields or medial frontal region, showed similar shifts in temporal response profiles following SD, suggesting differential contribution of areas mediating attention control towards limiting rapid object processing. As nodes of the default mode network (DMN) continued to show monotonically increasing deactivation at higher presentation frequencies even following SD, the observed state modulations of temporal responses likely represent temporal limitations in object processing as opposed to task disengagement.

The attentional blink reveals the probabilistic nature of discrete conscious perception.

Attention and awareness are two tightly coupled processes that have been the subject of the same enduring debate: Are they allocated in a discrete or in a graded fashion? Using the attentional blink paradigm and mixture-modeling analysis, we show that awareness arises at central stages of information processing in an all-or-none manner. Manipulating the temporal delay between two targets affected subjects' likelihood of consciously perceiving the second target, but did not affect the precision of its representation. Furthermore, these results held across stimulus categories and paradigms, and they were dependent on attention having been allocated to the first target. The findings distinguish the fundamental contributions of attention and awareness at central stages of visual cognition: Conscious perception emerges in a quantal manner, with attention serving to modulate the probability that representations reach awareness.

A Unified attentional bottleneck in the human brain.

Human information processing is characterized by bottlenecks that constrain throughput. These bottlenecks limit both what we can perceive and what we can act on in multitask settings. Although perceptual and response limitations are often attributed to independent information processing bottlenecks, it has recently been suggested that a common attentional limitation may be responsible for both. To date, however, evidence supporting the existence of such a "unified" bottleneck has been mixed. Here, we tested the unified bottleneck hypothesis using time-resolved fMRI. Experiment 1 isolated brain regions involved in the response selection bottleneck that limits speeded dual-task performance. These same brain regions were not only engaged by a perceptual encoding task in Experiment 2, their activity also tracked delays to a speeded decision-making task caused by concurrent perceptual encoding (Experiment 3). We conclude that a unified attentional bottleneck, including the inferior frontal junction, superior medial frontal cortex, and bilateral insula, temporally limits operations as diverse as perceptual encoding and decision-making.

Time-on-task and sleep deprivation effects are evidenced in overlapping brain areas.

Both sleep deprivation and extended task engagement (time-on-task) have been shown to degrade performance in tasks evaluating sustained attention. Here we used pulsed arterial spin labeling (pASL) to study participants engaged in a demanding selective attention task. The participants were imaged twice, once after a normal night of sleep and once after approximately 24h of total sleep deprivation. We compared task-related changes in BOLD signal alongside ASL-based cerebral blood flow (CBF) changes. We also collected resting baseline CBF data prior to and following task performance. Both BOLD fMRI and ASL identified spatially congruent task activation in ventral visual cortex and fronto-parietal regions. Sleep deprivation and time-on-task caused a decline of both measures in ventral visual cortex. BOLD fMRI also revealed such declines in fronto-parietal cortex. Only early visual cortex showed a significant upward shift in resting baseline CBF following sleep deprivation, suggesting that the neural consequences of both SD and ToT are primarily evident in task-evoked signals. We conclude that BOLD fMRI is preferable to pASL in studies evaluating sleep deprivation given its better signal to noise characteristics and the relative paucity of state differences in baseline CBF.

Faster scaling of auditory neurons in cortical areas relative to subcortical structures in primate brains.

Allometric studies in primates have shown that the cerebral cortex, cerebellum, and remaining brain structures increase in size as a linear function of their numbers of neurons and nonneuronal cells across primates. Whether such scaling rules also apply to functionally related structures such as those of the auditory system is unknown. Here, we investigate the scaling of brain structures in the auditory pathway of six primate species and the closely related tree shrew. Using the isotropic fractionator method to estimate the numbers of neurons and nonneuronal cells in the inferior colliculus, medial geniculate nucleus, and auditory cortex (Ac), we assessed how they scaled across species and examined the relative scaling relationships among them. As expected, each auditory structure scales in mass as a linear function of its number of neurons, with no significant changes in neuronal density across species. The Ac scales proportionately with the cerebral cortex as a whole, maintaining a relative mass of approximately 1% and a relative number of neurons of 0.7%. However, the Ac gains neurons faster than both subcortical structures examined. As a result, larger primate brains have increased ratios of cortical to subcortical neurons involved in processing auditory information.

CURATE.AI COR-Tx platform as a digital therapy and digital diagnostic for cognitive function in patients with brain tumour postradiotherapy treatment: protocol for a prospective mixed-methods feasibility clinical trial.

INTRODUCTION: Conventional interventional modalities for preserving or improving cognitive function in patients with brain tumour undergoing radiotherapy usually involve pharmacological and/or cognitive rehabilitation therapy administered at fixed doses or intensities, often resulting in suboptimal or no response, due to the dynamically evolving patient state over the course of disease. The personalisation of interventions may result in more effective results for this population. We have developed the CURATE.AI COR-Tx platform, which combines a previously validated, artificial intelligence-derived personalised dosing technology with digital cognitive training. METHODS AND ANALYSIS: This is a prospective, single-centre, single-arm, mixed-methods feasibility clinical trial with the primary objective of testing the feasibility of the CURATE.AI COR-Tx platform intervention as both a digital intervention and digital diagnostic for cognitive function. Fifteen patient participants diagnosed with a brain tumour requiring radiotherapy will be recruited. Participants will undergo a remote, home-based 10-week personalised digital intervention using the CURATE.AI COR-Tx platform three times a week. Cognitive function will be assessed via a combined non-digital cognitive evaluation and a digital diagnostic session at five time points: preradiotherapy, preintervention and postintervention and 16-weeks and 32-weeks postintervention. Feasibility outcomes relating to acceptability, demand, implementation, practicality and limited efficacy testing as well as usability and user experience will be assessed at the end of the intervention through semistructured patient interviews and a study team focus group discussion at study completion. All outcomes will be analysed quantitatively and qualitatively. ETHICS AND DISSEMINATION: This study has been approved by the National Healthcare Group (NHG) DSRB (DSRB2020/00249). We will report our findings at scientific conferences and/or in peer-reviewed journals. TRIAL REGISTRATION NUMBER: NCT04848935.

Brief Report: Emotional Picture and Language Processing in Adults with Autism Spectrum Disorder.

There is currently limited research and a lack of consensus on emotional processing impairments among adults with autism spectrum disorder (ASD). The present pilot study sought to characterize the extent to which adults with ASD are impaired in processing emotions in both words and pictures. Ten adults with ASD rated word and picture stimuli on emotional valence and arousal. Their ratings were compared to normative data for both stimuli sets using item-level correlations. Adults with ASD rank-ordered stimuli similarly to typically developing individuals, demonstrating relatively typical understanding of emotional words and pictures. However, they used a narrower range of the scales which suggests more subtle impairments affecting emotion-processing. Future directions arising from the findings of this pilot study are discussed.

Shared and unique brain network features predict cognitive, personality, and mental health scores in the ABCD study.

How individual differences in brain network organization track behavioral variability is a fundamental question in systems neuroscience. Recent work suggests that resting-state and task-state functional connectivity can predict specific traits at the individual level. However, most studies focus on single behavioral traits, thus not capturing broader relationships across behaviors. In a large sample of 1858 typically developing children from the Adolescent Brain Cognitive Development (ABCD) study, we show that predictive network features are distinct across the domains of cognitive performance, personality scores and mental health assessments. On the other hand, traits within each behavioral domain are predicted by similar network features. Predictive network features and models generalize to other behavioral measures within the same behavioral domain. Although tasks are known to modulate the functional connectome, predictive network features are similar between resting and task states. Overall, our findings reveal shared brain network features that account for individual variation within broad domains of behavior in childhood.

Offset analgesia is reduced on the palm and increases with stimulus duration.

BACKGROUND: A noxious stimulus following a more intense stimulus often feels less painful than continuous noxious stimulation. This effect, known as offset analgesia (OA), may be due to descending inhibitory control, to changes in peripheral neural transmission or both. The timing and location of noxious thermal stimulation were manipulated to better understand the peripheral and central contributions to OA. METHODS: In a first experiment, participants (n = 29) provided continuous pain ratings as stimuli were delivered to the palm or dorsum of each hand. Offset trials included 44°C (T1), 45°C (T2) and 44°C (T3) stimulation periods. Baseline trials were identical except the T3 temperature fell to 35°C. Constant trials were 44°C throughout. The duration of T1 and T2 was either 1 s or 6 s, whereas T3 was always 12 s. In a second experiment, participants (n = 43) rated pain levels of noxious stimuli presented to the forearms with varying T1 and T2 durations (3, 6, 10 or 13 s) and a 20 s T3 period. RESULTS: OA effects became stronger with increasing inducing durations. OA, however, was not found on the palm even at longer durations. CONCLUSIONS: The increase in OA with duration suggests that accumulated nociceptive signalling is more important to triggering OA than is a decrease in nociceptors' instantaneous firing rates. The lack of OA on the palm, however, implies a key role for the rapidly adapting Type II AMH fibres that may be absent or not readily activated on the palm. Unravelling the relative central and peripheral contribution to OA requires further investigation. SIGNIFICANCE: Offset analgesia (OA) is a fundamentally temporal phenomenon dependent on dynamic changes in stimulus intensity. Here we demonstrate increased OA with increased stimulus duration. This finding implies the more slowly-responding AMH-I peripheral mechanoreceptors contribute to OA. The more rapidly responding AMH-II peripheral mechanoreceptors, however, may be absent or more difficult to activate in the palm where we did not observe OA. This finding implies that the AMH-II receptors are necessary for OA. Our studies suggest methods to unravel the different peripheral and central contributions to OA.

Isolation of a central bottleneck of information processing with time-resolved FMRI.

When humans attempt to perform two tasks at once, execution of the first task usually leads to postponement of the second one. This task delay is thought to result from a bottleneck occurring at a central, amodal stage of information processing that precludes two response selection or decision-making operations from being concurrently executed. Using time-resolved functional magnetic resonance imaging (fMRI), here we present a neural basis for such dual-task limitations, e.g. the inability of the posterior lateral prefrontal cortex, and possibly the superior medial frontal cortex, to process two decision-making operations at once. These results suggest that a neural network of frontal lobe areas acts as a central bottleneck of information processing that severely limits our ability to multitask.

What are the units of storage in visual working memory?

An influential theory suggests that integrated objects, rather than individual features, are the fundamental units that limit our capacity to temporarily store visual information (S. J. Luck & E. K. Vogel, 1997). Using a paradigm that independently estimates the number and precision of items stored in working memory (W. Zhang & S. J. Luck, 2008), here we show that the storage of features is not cost-free. The precision and number of objects held in working memory was estimated when observers had to remember either the color, the orientation, or both the color and orientation of simple objects. We found that while the quantity of stored objects was largely unaffected by increasing the number of features, the precision of these representations dramatically decreased. Moreover, this selective deterioration in object precision depended on the multiple features being contained within the same objects. Such fidelity costs were even observed with change detection paradigms when those paradigms placed demands on the precision of the stored visual representations. Taken together, these findings not only demonstrate that the maintenance of integrated features is costly; they also suggest that objects and features affect visual working memory capacity differently.

An attentional blink for sequentially presented targets: evidence in favor of resource depletion accounts.

Several accounts of the attentional blink (AB) have postulated that this dual-target deficit occurs because of limited-capacity attentional resources being devoted to processing the first target at the expense of the second (resource depletion accounts; e.g., Chun & Potter, 1995). Recent accounts have challenged this model (e.g., Di Lollo, Kawahara, Ghorashi, & Enns, 2005; Olivers, van der Stigchel, & Hulleman, 2007), proposing instead that the AB occurs because of subjects' inability to maintain appropriate levels of attentional controlwhen targets are separated by distractors. Accordingly, the AB is eliminated when three targets from the same attentional set are presented sequentially in a rapid serial visual presentation (RSVP) stream. However, under such conditions poorer identification of the first target is typically observed, hinting at a potential trade-off between the first and subsequent target performances. Consistent with this hypothesis, the present study shows that an AB is observed for successive targets from the same attentional set in an RSVP stream when the first target powerfully captures attention. These results suggest that resource depletion contributes significantly to the AB.

Increased Automaticity and Altered Temporal Preparation Following Sleep Deprivation.

STUDY OBJECTIVES: Temporal expectation enables us to focus limited processing resources, thereby optimizing perceptual and motor processing for critical upcoming events. We investigated the effects of total sleep deprivation (TSD) on temporal expectation by evaluating the foreperiod and sequential effects during a psychomotor vigilance task (PVT). We also examined how these two measures were modulated by vulnerability to TSD. DESIGN: Three 10-min visual PVT sessions using uniformly distributed foreperiods were conducted in the wake-maintenance zone the evening before sleep deprivation (ESD) and three more in the morning following approximately 22 h of TSD. TSD vulnerable and nonvulnerable groups were determined by a tertile split of participants based on the change in the number of behavioral lapses recorded during ESD and TSD. A subset of participants performed six additional 10-min modified auditory PVTs with exponentially distributed foreperiods during rested wakefulness (RW) and TSD to test the effect of temporal distribution on foreperiod and sequential effects. SETTING: Sleep laboratory. PARTICIPANTS: There were 172 young healthy participants (90 males) with regular sleep patterns. Nineteen of these participants performed the modified auditory PVT. MEASUREMENTS AND RESULTS: Despite behavioral lapses and slower response times, sleep deprived participants could still perceive the conditional probability of temporal events and modify their level of preparation accordingly. Both foreperiod and sequential effects were magnified following sleep deprivation in vulnerable individuals. Only the foreperiod effect increased in nonvulnerable individuals. CONCLUSIONS: The preservation of foreperiod and sequential effects suggests that implicit time perception and temporal preparedness are intact during total sleep deprivation. Individuals appear to reallocate their depleted preparatory resources to more probable event timings in ongoing trials, whereas vulnerable participants also rely more on automatic processes.