Functional Division and Synergy of Cognitive Control and Salience Processing in Category-Based Attentional Selection: Evidence from fMRI

WU Xia¹,²,³, LI Yiwei¹, SUN Xiaoya¹, CHEN Ying⁴, JIANG Yunpeng¹,²,³, CHEN Yan⁵,¹

¹ Faculty of Psychology, Tianjin Normal University; ² Key Research Base of Humanities and Social Sciences of the Ministry of Education, Academy of Psychology and Behavior; ³ Tianjin Social Science Laboratory of Students' Mental Development and Learning, Tianjin 300387, China

⁴ School of Vocational Education, Tianjin University of Technology and Education, Tianjin 300222, China

⁵ School of Psychology, Guizhou Normal University, Guiyang 550000, China

Abstract

Category-based attentional selection (CAS) represents a core process through which the brain optimizes information filtering via abstract category representations, yet the synergistic mechanisms by which cognitive control and salience processing jointly regulate CAS remain unclear. The present study combined a Majority Function Task (MFT)—which manipulates cognitive load through symbol category ratios (low load: 3:0, high load: 2:1)—with an Oddball paradigm that modulates salience level through stimulus probability (standard: 80%, novel: 20%), while orthogonally varying target relevance (task-relevant: altering category probabilities; task-irrelevant: altering color probabilities). This design systematically examined the behavioral and neural regulatory mechanisms of cognitive control and salience processing on CAS. Behavioral results demonstrated that high cognitive load significantly reduced CAS efficiency, with novel stimuli producing interference effects only when task-relevant. A three-way interaction revealed that under task-relevant conditions, the interference effect of novel stimuli was significantly greater under high load than low load. fMRI results showed that high cognitive load activated the dorsal attention network (DLPFC, SPL), whereas novel stimuli activated the ventral attention network (rTPJ, AIC). Conjunction activation analysis revealed co-activation within the cognitive control network (SPL, ACC, AIC). Multivoxel pattern analysis (MVPA) further demonstrated that the right parieto-occipital junction (rPOJ) and frontal eye fields (FEF) achieved 86.83% decoding accuracy for cognitive load and salience processing, indicating that these regions integrate dual-pathway information for dynamic resource allocation. Overall, cognitive control and salience processing consume resources through dorsal and ventral networks, respectively; when both processes co-occur, the cognitive control network determines CAS efficiency through conflict resolution and resource reallocation. This study elucidates the functional division and synergistic mechanisms of cognitive control and salience processing at the category level, proposes a dynamic pathway model, and provides novel neural evidence for refining dual-pathway models of attention.

Keywords: cognitive control, salience processing, category-based attention, cognitive control network, right parieto-occipital junction

Attentional selection constitutes a fundamental mechanism by which information-processing systems prioritize limited inputs for deeper processing in complex environments, serving as a gateway for cognitive activity (Desimone & Duncan, 1995). In the external world, the diversity of object features and dynamic environmental changes lead individuals to organize perceptual experience according to object collections—categories (Rosch et al., 1976). The abstract nature of categories integrates discrete features into unified representations, substantially reducing information redundancy and enhancing processing efficiency (Reeder & Peelen, 2013; Macé et al., 2009). Category-based attentional selection (CAS) directs attentional selection toward category representations, affording prioritized selection to stimuli matching target categories (Peelen & Kastner, 2014; Yang & Zelinsky, 2009; Wyble et al., 2013; Wu et al., 2016; Wu & Fu, 2017). Compared with feature-level attentional selection, CAS exhibits longer processing time windows and stronger involvement of object-selective cortices (Freedman et al., 2003; Ferrera et al., 2009), making it a crucial process for investigating dynamic interactions between high-level cognition and perception. However, previous research has primarily focused on the neural mechanisms of CAS itself (activated brain regions and temporal dynamics), leaving unclear how attention systems dynamically regulate cognitive resources during CAS and lacking theoretical foundations for applying CAS in complex environments.

On one hand, cognitive control can influence resource allocation to CAS. Cognitive control refers to the capacity to monitor, inhibit, and schedule information-processing flows according to current goals under resource-limited conditions, representing a core function of high-order attentional systems (Wu et al., 2020a; Fan, 2014; Lavie, 2005). The Majority Function Task (MFT) manipulates cognitive load by altering the entropy of to-be-processed information. The task requires participants to determine the majority orientation among a set of arrows; when the majority-to-minority ratio decreases from 3:0 to 2:1, information conflict increases substantially, necessitating recruitment of additional cognitive control resources to coordinate perceptual encoding and decision-making operations (Wu et al., 2016). Under elevated cognitive load, functional connectivity between the right temporoparietal junction (rTPJ) and dorsal attention network (DAN) strengthens significantly, facilitating real-time synergy between goal-directed top-down signals and stimulus-driven salience signals at rTPJ, thereby enhancing attentional priority for features such as color and location (Wu et al., 2015). However, how variations in cognitive load modulate selection efficiency and neural mechanisms specifically for category information remains unknown.

On the other hand, salience processing may also affect resource allocation to CAS. Salience refers to the conspicuousness of stimuli in physical or statistical features, rapidly guiding attention (Itti & Koch, 2001). According to whether it conforms to current task goals, salience can be distinguished as task-relevant versus task-irrelevant (Kim, 2014). The former denotes that novel stimuli overlap with features or categories upon which target decisions depend, which can be encoded as valid information and is accompanied by coordinated activity in frontoparietal networks (including lateral prefrontal cortex LPFC and intraparietal sulcus IPS) (Brass & von Cramon, 2004; Lerebourg et al., 2024). The latter refers to stimuli that, while prominent in probability or physical attributes, fail to satisfy current target conditions and constitute external interference only (Theeuwes, 2010). This interference effect correlates with enhanced rTPJ activation (Kucyi et al., 2012; Corbetta & Shulman, 2002). Task-relevant salience may conserve cognitive control resources and accelerate category filtering (Thayer et al., 2022), whereas task-irrelevant salience lacks active suppression of irrelevant features, generating resource competition (Oxner et al., 2023). Experimentally, the Oddball paradigm renders low-probability novel stimuli statistically salient, thereby triggering attentional capture (Näätänen, 2011). The mismatch negativity (MMN) and subsequent P3 components in EEG directly reflect the efficiency of such salience processing (Bekinschtein et al., 2009; Garrido et al., 2009). However, previous studies have predominantly focused on low-level features such as color and location (Chapman & Störmer, 2022; Oxner et al., 2023), leaving direct evidence lacking regarding whether salience processing influences resource allocation for high-level category information and whether differences between the two types of salience processing emerge in CAS contexts.

Both cognitive control and salience processing influence cognitive resource allocation, exhibiting functional division while simultaneously demonstrating synergy under specific conditions. On one hand, they show distinct neural mechanisms involving different brain networks: cognitive control primarily engages core regions of the cognitive control network (CCN) such as dorsolateral prefrontal cortex (DLPFC), frontal eye fields (FEF), and superior parietal lobule (SPL) (Li et al., 2010; Noudoost & Moore, 2011; Wang et al., 2020; De Fockert et al., 2004; Wu et al., 2015), whereas salience processing mainly involves the ventral attention network (VAN, e.g., rTPJ, anterior insula AIC) and salience network (SN, e.g., IPS, superior temporal sulcus STS) to rapidly detect external salient signals (Arcizet et al., 2011; Corbetta & Shulman, 2002; Theeuwes, 2010). This division becomes particularly evident in conflict tasks: when salient distractors conflict with task goals, the prefrontal cortex must enhance control over parietal regions to suppress interference (Broschard et al., 2024), while parietal cortex preferentially encodes salience signals (Kroner et al., 2023). In CAS, the abstract nature of category targets requires greater cognitive resource investment; under higher load conditions, stronger cognitive control is needed to continuously reinforce target templates, thereby better resisting irrelevant salient interference (Yang & Zelinsky, 2009). Salience processing, conversely, may rapidly capture salient features within categories through ventral visual pathways, competing with dorsal pathway spatial orienting (Chen et al., 2012), further accentuating division characteristics. On the other hand, cognitive control and salience processing also exhibit synergy in complex visual tasks. For instance, TPJ serves as a multimodal information integration hub, playing a central role in coordinating task goals and salient stimuli (Wu et al., 2015; Corbetta et al., 2008; Kucyi et al., 2012), suggesting that cognitive control and salience processing may achieve information synergy in specific brain regions to enable dynamic attentional resource allocation. Based on this, the present study investigates the boundaries and patterns of their synergy—namely, under which conditions stronger synergy emerges. These boundary conditions and synergistic patterns require further examination within the CAS framework.

To investigate the regulatory effects and neural mechanisms of cognitive control and salience processing on CAS, the present study employed functional magnetic resonance imaging (fMRI) technology, combining MFT and Oddball paradigms to simultaneously manipulate cognitive load, salience level, and salience relevance. By varying cognitive load (low load 3:0 vs. high load 2:1), salience level (standard stimuli 80% vs. novel stimuli 20%), and relevance to current task goals (task-relevant category vs. task-irrelevant color), we recorded participants' behavioral responses and brain activation patterns. Integrating whole-brain activation analysis and multivariate pattern analysis (MVPA) allowed multi-angle examination of the functional division and synergy between the two processes, providing systematic support for dynamic resource allocation. Experimental hypotheses were: (1) Behaviorally, high cognitive load would reduce CAS efficiency (decreased accuracy, prolonged reaction time); novel stimuli would interfere with CAS, with less interference when task-relevant; a three-way interaction would show that under task-relevant conditions, novel stimulus interference would be more pronounced under high load than low load. (2) Neuroimaging-wise, cognitive control would activate DAN (e.g., DLPFC, SPL), salience processing would activate VAN (e.g., rTPJ); CCN would participate in synergistic processing of both, with multivoxel patterns capable of distinguishing the two processes with high decoding accuracy.

2.1 Participants

Using MorePower 6.0 (Campbell & Thompson, 2012), we calculated the required sample size for a within-subjects 2 × 2 × 2 design with statistical power of 0.80, α level of 0.05, and medium effect size (ηp² = 0.3), yielding a required sample of 26 participants.

We recruited 29 university students from Tianjin Normal University (age range: 18–27 years; mean age = 20.83 ± 2.60 years; 24 women, 5 men). All participants were right-handed with normal or corrected-to-normal vision and no color blindness. The experiment was approved by the Tianjin Normal University Ethics Committee (approval number: 2022030702). Participants provided informed consent prior to the experiment and received compensation upon completion.

2.2 Experimental Design and Procedure

Stimuli were presented on a 17-inch CRT monitor with a resolution of 1024 × 768 pixels and a refresh rate of 60 Hz against a gray background (CIE x/y coordinates: 0.313/0.329). Participants viewed the screen from approximately 57 cm inside a shielded room. The experimental program was compiled using E-Prime 3.0 software.

The experiment employed a 2 (cognitive load: high, low) × 2 (salience level: novel, standard) × 2 (salience relevance: task-relevant, task-irrelevant) within-subjects design. The task combined MFT and Oddball paradigms (Figure 1 [FIGURE:1]). Each trial began with a central fixation cross presented for 100–600 ms, followed by a 400 ms search display containing three symbols (0.31° × 0.31°) equidistant from central fixation (0.86° eccentricity). Symbols could be digits (2–9) or letters (A, B, C, D, E, F, G, H). Symbol positions were selected from 12 possible locations around central fixation, with the three symbols appearing at equal angular distances (e.g., positions 1, 5, 9) to avoid uneven spatial distribution. Participants judged the category (digit or letter) of the majority of symbols and responded via button press: right index finger for digit-majority, right middle finger for letter-majority. Response mapping was counterbalanced across participants. After a response window of 900–1400 ms and a 100 ms blank screen, the next trial commenced.

Cognitive load was manipulated by altering category ratios: low load trials presented three symbols from the same category (3:0 ratio, all digits or all letters), whereas high load trials presented two symbols from one category and one from the other (2:1 ratio, e.g., two digits and one letter). Salience level was manipulated via stimulus frequency: standard stimuli comprised 80% of trials, while novel oddball stimuli comprised 20%. Salience relevance was manipulated by varying whether the oddball property aligned with task goals: task-irrelevant salience was implemented by altering color proportions (red: CIE x/y 0.640/0.330; green: CIE x/y 0.300/0.600), with one color serving as standard (80%) and the other as novel (20%), counterbalanced across blocks and participants. Task-relevant salience was implemented by altering the proportion of the majority symbol category, with one category (digits or letters) serving as the majority in 80% of trials and the other in 20%, counterbalanced across blocks and participants.

The experiment comprised four blocks of 200 trials each (160 standard, 40 novel). Participants completed thorough practice before the formal experiment. A 2-minute break was provided after each block, with total experiment duration approximately 35 minutes.

Figure 1 Schematic illustration of experimental design. Note: Cognitive load (low vs. high) was manipulated by altering the majority category ratio among three symbols (3:0 vs. 2:1). Stimulus salience level (standard vs. novel) was manipulated by altering presentation frequency (standard 80%, novel 20%). Salience relevance (relevant vs. irrelevant) was manipulated by alignment with task goals: under task-relevant conditions, salience was implemented by altering the proportion of the task-relevant majority category; under task-irrelevant conditions, salience was implemented by altering the proportion of task-irrelevant stimulus colors.

2.3 fMRI Data Acquisition and Preprocessing

Brain imaging data were acquired on a Siemens Prisma 3.0T scanner using a 64-channel head coil. Blood oxygen level-dependent (BOLD) signals were collected using echo-planar imaging (EPI) with the following parameters: resolution = 3.8 × 3.8 × 4.0 mm³, repetition time (TR) = 2 s, echo time (TE) = 27 ms, echo spacing = 0.4 ms, field of view = 240 × 240 mm², flip angle = 77°, slice thickness = 4.0 mm, slice gap = 0, number of slices = 40, EPI factor = 64, bandwidth = 3126 Hz/Px. Following functional scans, participants completed a 5-minute structural scan with resolution = 0.9 × 0.9 × 0.9 mm³.

Preprocessing included slice-timing correction, spatial normalization, coregistration, segmentation, motion correction, and smoothing. Functional images underwent slice-timing correction to align acquisition times across slices, followed by motion correction by realigning all images to the first volume. All images were normalized to MNI (Montreal Neurological Institute) template space within a 3 × 3 × 3 mm³ Talairach framework (Talairach & Tournoux, 1988) using bilinear interpolation. Functional images were spatially smoothed using a Gaussian filter with full-width at half maximum (FWHM) of 6 mm.

2.4 fMRI Statistical Analysis

Brain imaging data were analyzed using SPM12 in MATLAB. At the individual level, first-level statistical maps of BOLD activation were computed via the general linear model (GLM) with eight regressors corresponding to correct responses in each of the eight experimental conditions: 2 (cognitive load: high, low) × 2 (salience level: novel, standard) × 2 (salience relevance: task-relevant, task-irrelevant). These regressors were convolved with the canonical hemodynamic response function (HRF). The model also included six motion parameters from motion correction as nuisance covariates. Each voxel's time series was high-pass filtered (1/128 Hz) to remove low-frequency noise and signal drift. Linear contrast values were defined for each condition, and after GLM parameter estimation, contrast images were generated for main effects of cognitive load (high vs. low), salience level (novel vs. standard), salience relevance (relevant vs. irrelevant), and their interactions. These contrast images entered second-level group analyses using GLM.

Cluster thresholds were determined via Monte Carlo simulation using AlphaSim (http://afni.nimh.nih.gov/pub/dist/doc/manual/AlphaSim.pdf). Assuming a per-voxel Type I error of p < 0.005, a cluster extent of 46 contiguous voxels was required for correction at p < 0.05.

Multivoxel pattern analysis (MVPA) was implemented in Python 3.11 using Nilearn, Numpy, and Sklearn. Based on second-level GLM results, voxels showing high activation in both cognitive load and salience level conditions were selected as feature voxels for each condition. At the individual level, data structure comprised fMRI scans, conditions, and groups. Decoding aimed to classify which condition a participant was performing based on brain activation patterns. fMRI data were processed using SPM12's ImCalc function to generate: (1) cognitive load contrast images by subtracting each low-load standard trial from each high-load standard trial, and each low-load novel trial from each high-load novel trial; (2) salience level contrast images by subtracting each high-load standard trial from each high-load novel trial, and each low-load standard trial from each low-load novel trial. Condition and group labels were retained for model input. A leave-one-out cross-validation procedure was employed, with three blocks as training data (75%) and the remaining block as test data (25%). Model performance was tested on the held-out block to obtain decoding accuracy. Voxel weights were computed to generate individual participant maps of decoding accuracy across voxels, which were then spatially smoothed (FWHM = 6 mm Gaussian filter) and entered into group-level GLM for differential testing of decoding accuracy across voxels.

3 Results

3.1 Behavioral Results

Accuracy (ACC), reaction time (RT) for correct responses, and inverse efficiency scores (IES = RT/ACC, reflecting overall performance) were analyzed using a 2 (cognitive load: high, low) × 2 (salience level: novel, standard) × 2 (salience relevance: task-relevant, task-irrelevant) repeated-measures ANOVA.

3.1.1 Accuracy

The main effect of cognitive load was significant, F(1, 28) = 209.34, p < 0.001, ηp² = 0.88, with higher accuracy under low load than high load. The main effect of salience level was significant, F(1, 28) = 22.06, p < 0.001, ηp² = 0.44, with higher accuracy for standard than novel stimuli.

The cognitive load × salience level interaction was significant, F(1, 28) = 24.98, p < 0.001, ηp² = 0.47. Simple effects analysis revealed that accuracy for standard stimuli was significantly higher than for novel stimuli under both low load [F(1, 28) = 4.35, p = 0.046, ηp² = 0.13] and high load [F(1, 28) = 38.33, p < 0.001, ηp² = 0.58]. A paired-samples t-test on the salience effect (ACC_standard − ACC_novel) across load conditions showed a larger effect under high load than low load, t(28) = 5.00, p < 0.001, Cohen's d = 0.93, 95% CI [0.49, 1.36]. Additionally, low-load accuracy was significantly higher than high-load accuracy for both standard [F(1, 28) = 193.28, p < 0.001, ηp² = 0.87] and novel stimuli [F(1, 28) = 148.29, p < 0.001, ηp² = 0.84]. A paired-samples t-test on the load effect (ACC_low − ACC_high) across salience levels revealed a larger effect for novel than standard stimuli, t(28) = 5.00, p < 0.001, Cohen's d = 0.93, 95% CI [0.49, 1.36].

The salience level × salience relevance interaction was significant, F(1, 28) = 24.01, p < 0.001, ηp² = 0.46. Simple effects analysis showed that when task-relevant, standard stimuli accuracy was significantly higher than novel stimuli accuracy [F(1, 28) = 28.06, p < 0.001, ηp² = 0.50]; when task-irrelevant, no significant difference emerged between standard and novel stimuli [F(1, 28) = 0.14, p = 0.710].

The three-way interaction of cognitive load × salience level × salience relevance was significant, F(1, 28) = 12.22, p = 0.002, ηp² = 0.30. To clarify this interaction pattern, separate 2 × 2 ANOVAs were conducted for task-relevant and task-irrelevant conditions. Under task-relevant conditions, the cognitive load × salience level interaction was significant, F(1, 28) = 37.03, p < 0.001, ηp² = 0.57. Standard stimuli accuracy exceeded novel stimuli accuracy under both low load [F(1, 28) = 6.34, p = 0.018, ηp² = 0.19] and high load [F(1, 28) = 51.49, p < 0.001, ηp² = 0.65]. The salience effect was larger under high load than low load, t(28) = 6.09, p < 0.001, Cohen's d = 1.13, 95% CI [0.66, 1.59]. Low-load accuracy surpassed high-load accuracy for both standard [F(1, 28) = 73.71, p < 0.001, ηp² = 0.73] and novel stimuli [F(1, 28) = 72.50, p < 0.001, ηp² = 0.72]. The load effect was larger for novel than standard stimuli, t(28) = 6.09, p < 0.001, Cohen's d = 1.13, 95% CI [0.66, 1.59]. In contrast, under task-irrelevant conditions, the cognitive load × salience level interaction was not significant, F(1, 28) = 0.04, p = 0.844.

Figure 2 [FIGURE:2] Accuracy (%) across cognitive load, salience level, and salience relevance conditions. Note: **p < 0.001, p < 0.05, ns = p ≥ 0.05. The same notation applies below.

3.1.2 Reaction Time

The main effect of cognitive load was significant, F(1, 28) = 404.96, p < 0.001, ηp² = 0.94, with faster RTs under low load than high load. The main effect of salience level was significant, F(1, 28) = 72.53, p < 0.001, ηp² = 0.72, with faster RTs for standard than novel stimuli. The main effect of salience relevance was significant, F(1, 28) = 22.16, p < 0.001, ηp² = 0.44, with faster RTs for task-relevant than task-irrelevant conditions.

The cognitive load × salience relevance interaction was significant, F(1, 28) = 6.54, p = 0.016, ηp² = 0.19. Simple effects analysis showed that low-load RTs were significantly faster than high-load RTs for both task-relevant [F(1, 28) = 298.04, p < 0.001, ηp² = 0.91] and task-irrelevant conditions [F(1, 28) = 390.17, p < 0.001, ηp² = 0.93]. The load effect (RT_high − RT_low) was larger for task-irrelevant than task-relevant conditions, t(28) = 2.56, p = 0.016, Cohen's d = 0.48, 95% CI [0.09, 0.86].

The salience level × salience relevance interaction was significant, F(1, 28) = 88.90, p < 0.001, ηp² = 0.76. Simple effects analysis revealed that when task-relevant, standard stimuli RTs were significantly faster than novel stimuli RTs [F(1, 28) = 100.78, p < 0.001, ηp² = 0.78]; when task-irrelevant, no significant difference emerged between standard and novel stimuli RTs [F(1, 28) = 0.00, p = 0.959].

The three-way interaction of cognitive load × salience level × salience relevance was not significant, F(1, 28) = 0.71, p = 0.405.

Figure 3 [FIGURE:3] Reaction time (ms) across cognitive load, salience level, and salience relevance conditions.

3.1.3 Inverse Efficiency Scores

IES = RT/ACC was computed to correct for speed-accuracy trade-offs, reflecting overall performance across conditions (Townsend & Ashby, 1983). Lower IES values indicate better performance.

The main effect of cognitive load was significant, F(1, 28) = 250.56, p < 0.001, ηp² = 0.90, with better performance under low load than high load. The main effect of salience level was significant, F(1, 28) = 35.11, p < 0.001, ηp² = 0.56, with better performance for standard than novel stimuli. The main effect of salience relevance was significant, F(1, 28) = 6.94, p = 0.014, ηp² = 0.20, with better performance for task-relevant than task-irrelevant conditions.

The cognitive load × salience level interaction was significant, F(1, 28) = 21.72, p < 0.001, ηp² = 0.44. Simple effects analysis showed that standard stimuli performance exceeded novel stimuli performance under both low load [F(1, 28) = 20.75, p < 0.001, ηp² = 0.43] and high load [F(1, 28) = 36.51, p < 0.001, ηp² = 0.57]. The salience effect (IES_novel − IES_standard) was larger under high load than low load, t(28) = 4.66, p < 0.001, Cohen's d = 0.87, 95% CI [0.43, 1.28]. Low-load performance was significantly better than high-load performance for both standard [F(1, 28) = 353.98, p < 0.001, ηp² = 0.93] and novel stimuli [F(1, 28) = 164.35, p < 0.001, ηp² = 0.85]. The load effect (IES_high − IES_low) was larger for novel than standard stimuli, t(28) = 4.66, p < 0.001, Cohen's d = 0.87, 95% CI [0.43, 1.28].

The salience level × salience relevance interaction was significant, F(1, 28) = 40.27, p < 0.001, ηp² = 0.59. Simple effects analysis revealed that when task-relevant, standard stimuli performance was better than novel stimuli performance [F(1, 28) = 43.48, p < 0.001, ηp² = 0.61]; when task-irrelevant, no significant difference emerged between standard and novel stimuli performance [F(1, 28) = 0.00, p = 0.973].

The three-way interaction of cognitive load × salience level × salience relevance was significant, F(1, 28) = 10.48, p = 0.003, ηp² = 0.27. To clarify this interaction, separate 2 × 2 ANOVAs were conducted for task-relevant and task-irrelevant conditions. Under task-relevant conditions, the cognitive load × salience level interaction was significant, F(1, 28) = 19.07, p < 0.001, ηp² = 0.41. Simple effects analysis showed that standard stimuli performance was significantly better than novel stimuli performance under both low load [F(1, 28) = 28.12, p < 0.001, ηp² = 0.50] and high load [F(1, 28) = 39.90, p < 0.001, ηp² = 0.59]. The salience effect (IES_novel − IES_standard) was larger under high load than low load, t(28) = 4.37, p < 0.001, Cohen's d = 0.81, 95% CI [0.39, 1.23]. Low-load performance was significantly better than high-load performance for both standard [F(1, 28) = 285.55, p < 0.001, ηp² = 0.91] and novel stimuli [F(1, 28) = 73.54, p < 0.001, ηp² = 0.72]. The load effect (IES_high − IES_low) was larger for novel than standard stimuli, t(28) = 4.37, p < 0.001, Cohen's d = 0.81, 95% CI [0.39, 1.23]. In contrast, under task-irrelevant conditions, the cognitive load × salience level interaction was not significant, F(1, 28) = 0.00, p = 0.965.

Figure 4 [FIGURE:4] Inverse efficiency scores (IES, ms) across cognitive load, salience level, and salience relevance conditions.

3.2 fMRI Results

3.2.1 Brain Activation Results

Whole-brain contrast analysis (Figure 5 [FIGURE:5] and Table 1 [TABLE:1]) revealed no significant brain regions for the main effect of salience relevance. Extracting task-relevant regions showed significant main effects of cognitive load: high load > low load activated bilateral superior parietal lobule, bilateral insula, left middle frontal gyrus, and bilateral calcarine fissure; low load > high load activated right cuneus, right lingual gyrus, right superior frontal gyrus, left angular gyrus, right inferior parietal lobule, right medial superior frontal gyrus, left middle temporal gyrus, left parahippocampal cortex, and left inferior frontal gyrus. The main effect of salience level was significant: novel > standard stimuli activated left precentral gyrus, right angular gyrus, bilateral insula, and bilateral caudate nucleus.

Figure 5 Main effects of cognitive load and salience level. (a) Brain regions associated with cognitive load (high load > low load). (b) Brain regions associated with salience level (novel > standard). Red indicates voxels with increased activation; blue indicates voxels with decreased activation.

Table 1 Brain regions showing main effects of cognitive load and salience level (FDR, p < 0.05, k ≥ 46)

The salience level × salience relevance interaction primarily manifested in significant positive activation of left postcentral gyrus, bilateral precentral gyrus, bilateral insula, and right superior parietal lobule (Figure 6 [FIGURE:6] and Table 2 [TABLE:2]). Under task-relevant conditions, novel > standard stimuli activated left precentral gyrus, left pars opercularis of inferior frontal gyrus, right inferior marginal angular gyrus, and left insula; standard > novel stimuli activated left posterior cingulate cortex, right precuneus, left anterior cingulate and paracingulate gyri, left dorsolateral superior frontal gyrus, right middle occipital gyrus, left central operculum, left middle temporal gyrus, and left orbital inferior frontal gyrus. Under task-irrelevant conditions, novel > standard stimuli only activated left fusiform gyrus and right lingual gyrus. No significant activation clusters were observed for cognitive load × salience level, cognitive load × salience relevance, or three-way interactions.

Figure 6 Salience level × salience relevance interaction. (a) Brain regions showing the interaction, (novel − standard)_task-relevant > (novel − standard)_task-irrelevant. (b) Simple effects for task-relevant conditions: brain regions activated by salience level contrast (novel task-relevant > standard task-relevant). (c) Simple effects for task-irrelevant conditions: brain regions activated by salience level contrast (novel task-irrelevant > standard task-irrelevant). Red indicates voxels with increased activation; blue indicates voxels with decreased activation.

Table 2 Salience level × salience relevance interaction and simple effects analysis (uncorrected, p < 0.001, k ≥ 46)

3.2.2 Joint Activation of Cognitive Control and Salience Processing

To investigate synergy between cognitive control and salience processing, conjunction activation analysis was performed on main effects of cognitive load and salience level. As shown in Figure 7 [FIGURE:7] and Table 3 [TABLE:3], left superior parietal lobule, left inferior occipital gyrus, right angular gyrus, right precentral gyrus, bilateral insula, and right caudate nucleus were jointly engaged in both cognitive control and salience processing.

Figure 7 Brain regions showing conjunction activation for cognitive control and salience processing. Red indicates voxels with increased activation.

Table 3 Brain regions showing conjunction activation for cognitive control and salience processing (FDR, p < 0.05, k ≥ 46)

3.2.3 Predictive Brain Activation Patterns for Cognitive Control and Salience Processing

To identify key hubs for synergy between cognitive control and salience processing, a classifier was trained on regions from the conjunction analysis to discriminate between cognitive load and salience level conditions. Results showed that the classifier achieved mean decoding accuracy of 86.83% for distinguishing cognitive load and salience level conditions, significantly above chance (50%) [t(28) = 73.57, p < 0.001, Cohen's d = 27.81]. Further analysis of voxel weights revealed that the right parieto-occipital junction (rPOJ) and right precentral gyrus/FEF contributed significantly to classifier predictions (Figure 8 [FIGURE:8] and Table 4 [TABLE:4]).

Figure 8 MVPA results discriminating cognitive control and salience processing. Red indicates brain regions with significant contribution to decoding.

Table 4 Brain regions discriminating cognitive control and salience processing (uncorrected, p < 0.005, k ≥ 46)

4 Discussion

By combining MFT and Oddball tasks to manipulate cognitive load, salience level, and salience relevance, this study systematically investigated the functional division and synergistic mechanisms of cognitive control and salience processing in category-based attentional selection (CAS). Behavioral results demonstrated that high load significantly impaired CAS efficiency (increased IES); task-relevant novel stimuli showed facilitatory effects under low load but switched to interference under high load, while task-irrelevant novel stimuli produced interference that amplified with increasing load. This pattern supports a dynamically updated category priority map wherein target enhancement dominates when resources are ample, but salience weighting exceeds target enhancement when resources are scarce. Neuroimaging results revealed that the dorsal attention network (DAN, including DLPFC, SPL) activated under high load, whereas the ventral attention network (VAN, including rTPJ, AIC) activated for novel stimuli, demonstrating clear functional division. However, conjunction activation analysis showed that core CCN regions (SPL, ACC, AIC) participated in both processes, suggesting that CCN coordinates cognitive control and salience processing through global resource regulation. rPOJ and FEF decoded both processes with ~87% accuracy, indicating they serve as synergistic hubs. These findings extend category priority maps to higher-order parieto-occipital regions and validate the applicability of multi-resource competition frameworks at the semantic level, injecting a dynamic perspective into dual-pathway models of attention.

Behavioral results revealed a dual-modulation pattern of cognitive load and salience processing on CAS, providing critical evidence for understanding the dynamics of attentional resource allocation (Lavie, 2005). Specifically, high load increased information processing costs (worse performance under high load), while task-irrelevant novel stimuli only significantly amplified interference under high load. In contrast, task-relevant novel stimuli showed slight facilitatory effects under low load (smaller difference from standard stimuli) but switched to pronounced interference under high load (larger difference from standard stimuli). This pattern supports priority map theory (Fecteau & Munoz, 2006; Itti & Koch, 2001; Mo et al., 2018), wherein priority values for resource allocation are flexibly modulated by both stimulus-driven salience and goal-driven relevance. When cognitive resources are abundant (low load), target templates (abstract semantic representations of categories) can be adequately maintained, allowing the system to actively utilize task-relevant salience (e.g., novel stimuli) as additional cues to enhance selection efficiency through positive modulation of the priority map. At this stage, task-relevant novel stimuli, matching the target template, are interpreted as beneficial signals, producing facilitatory effects. However, under high load, target template maintenance is weakened, forcing the attentional system to rely on externally driven stimulus processing (Geng & Mangun, 2011). Consequently, any novel stimulus (regardless of relevance) consumes residual resources, particularly task-relevant novel stimuli that partially overlap with target features and are amplified in the priority map, causing their effect to switch from facilitation under low load to significant interference under high load. By comparison, task-irrelevant novel stimuli consistently lack target template constraints, with interference magnitude determined primarily by remaining resources: under low load, ample resources produce minimal interference; under high load, resource scarcity exacerbates interference through intensified competition. This difference further validates selective amplification mechanisms (Lavie, 2005), whereby when attentional resources are limited, the cognitive system does not uniformly suppress all salience processing but rather sacrifices fine-tuned modulation of target-matching features first, creating differential interference patterns across relevance conditions (task-relevant vs. task-irrelevant). Overall, cognitive load dynamically alters neural resource allocation strategies for salience processing by modulating target template maintenance strength: under low load, the system actively processes relevant salience to optimize selection; under high load, it passively responds to resource competition from all novel stimuli. This mechanism provides direct behavioral evidence for dynamic resource allocation in CAS and extends priority map theory into the semantic domain and dynamic resource frameworks.

The roles of cognitive control and salience processing in CAS reflect a dialectical unity of functional division and neural synergy. Their division manifests as separable neural pathways: cognitive control primarily relies on DAN (including DLPFC, FEF, SPL) to reinforce target templates (e.g., semantic categories of digits) and suppress interference (Li et al., 2010; Broschard et al., 2024), whereas salience processing automatically captures salient features (e.g., color contrast of green letters) through SN (including ACC and AIC) (Harsay et al., 2012; Kumaran et al., 2009; Corbetta & Shulman, 2002; Kroner et al., 2023). This separation is particularly pronounced in conflict tasks: when task-irrelevant salient distractors appear under high cognitive load, DLPFC activation increases to maintain target representations while TPJ preferentially responds to salience signals, impairing behavioral efficiency (Bouvier et al., 2023; Geng & Mangun, 2011). This resource competition phenomenon indicates that the two processes differ fundamentally in resource allocation: cognitive control is a capacity-limited, goal-directed process, whereas salience processing is a stimulus-driven, automatic process (Theeuwes, 2010).

Conversely, synergy between the two processes in CAS is achieved through dynamic functional coupling. fMRI conjunction analysis revealed that left superior parietal lobule, left inferior occipital gyrus, right precentral gyrus, right ACC, bilateral AIC, and right caudate nucleus participated in both processes, indicating that CCN coordinates goals and salience through global resource regulation to accomplish cognitive control processing (Keller et al., 2022; Li et al., 2010; Noudoost & Moore, 2011; Wang et al., 2020; Wu et al., 2015). For example, when task-relevant salience appears, functional connectivity between DLPFC and LOC strengthens, synergistically encoding salience signals into target templates to enhance search efficiency (Oxner et al., 2023). This synergistic mechanism received further support from neural decoding results: multivoxel activation patterns in rPOJ and FEF significantly distinguished cognitive control from salience processing. rPOJ, located at the high-order visual-spatial junction of occipital and parietal cortices, participates in attentional selection and integrates visual and motor information for visuomotor coordination (Chen et al., 2012; Collignon et al., 2011; Zhang et al., 2024). Thus, rPOJ may serve as a critical conflict-monitoring region that dynamically adjusts dorsal-ventral pathway resource allocation by real-time evaluation of conflict strength between task demands and stimulus salience. Meanwhile, FEF, situated in the premotor-attentional hub of the precentral gyrus, is a key region for translating the synergistically encoded priority map into concrete orienting commands (Noudoost & Moore, 2011). Therefore, FEF may act as an output conversion hub from priority map to action commands, feeding back selective gain to visual cortex and synchronously modulating eye-hand orienting networks to encode goal-driven enhancement and salience-driven conspicuous signals into executable orienting and response plans. Overall, these results align with Katsuki and Constantinidis's (2014) adaptive control model, which posits that parietal-frontal networks dynamically regulate goal-directed enhancement and stimulus salience processing according to priority maps. The present study further speculates that CAS efficiency depends on CCN's capacity for processing complex information and flexible allocation, with rPOJ and FEF serving as key nodes for synergistic processing. In summary, this study pioneeringly reveals the division-synergy balance between cognitive control and salience processing within a category-attention framework. Unlike simple feature-based attention (Chapman & Störmer, 2022), the abstract nature of category targets in CAS demands higher-level semantic processing, amplifying CCN's regulatory requirements. This suggests that rapid target identification depends not only on explicit feature definition but also on dynamic coordination of conflicts between task goals and environmental salience.

Based on evidence that cognitive control and salience processing are functionally divided yet interconnected, we propose a concise dynamic pathway model. First, division is reflected in relatively distinct neural pathways and functional roles: cognitive control is dominated by DAN (e.g., DLPFC, FEF, SPL), prioritizing and maintaining target categories through capacity-limited, goal-directed processing (Li et al., 2010; Awh et al., 2012); salience processing is dominated by VAN and SN (e.g., rTPJ, ACC, AIC), automatically detecting and capturing physically or statistically salient features, with the two pathways anatomically and initiatorily separable (Corbetta et al., 2002; Arcizet et al., 2011). Second, under complex task contexts, they exhibit conditional synergy: when task-irrelevant salient distractors appear under high cognitive load, the attentional system often manifests resource competition with coupled reallocation, wherein DAN enhances target representations via DLPFC to maintain control while VAN preferentially responds to salience signals via TPJ (Geng & Mangun, 2011; Wu et al., 2015). Ultimately, hubs such as ACC/AIC coordinate resource allocation between the two processes with assistance from rPOJ and FEF, achieving inhibition of irrelevant salience and enhancement of relevant salience to enable dynamic attentional configuration (Chen et al., 2012; Katsuki & Constantinidis, 2014). This resource reallocation process also explains the behavioral pattern of facilitation under low load and interference under high load, consistent with priority map theory's emphasis on salience-goal weight competition (Fecteau & Munoz, 2006). Overall, the dynamic pathway model emphasizes that cognitive control and salience processing are not mutually opposed but achieve functional division through separable pathways while synergizing under specific boundary conditions to simultaneously satisfy goal-directed and stimulus-driven demands in complex environments.

Notably, CCN's interaction with salience processing depends on task relevance. The study found that only when salient stimuli were task-relevant did core CCN regions (e.g., DLPFC, SPL) show significantly enhanced activation, with functional connectivity strength to visual cortex (e.g., LOC) positively correlating with behavioral efficiency. This phenomenon can be explained by a goal-salience synergy mechanism: CCN not only maintains goal direction but also dynamically evaluates the task relevance of salience signals and synergistically encodes qualifying salient features into target representations. This synergy is crucial in conflict tasks where multiple stimuli share category attributes, requiring CCN to prioritize salient stimuli spatially or semantically consistent with task goals via SPL's spatial orienting function (Kroner et al., 2023). This finding challenges the traditional view that CCN is responsible for pure cognitive control (Desimone & Duncan, 1995), indicating greater flexibility: in task-relevant salience processing, CCN acts as an adaptive filter that both suppresses irrelevant interference and actively enhances salient signals matching target templates. For example, Wu et al. (2015) proposed that after TPJ (SN core) detects salient stimuli, it communicates their task relevance to CCN via functional connectivity with prefrontal cortex, determining whether to allocate attentional resources. In this study, task-relevant novel stimuli evoked co-activation of rPOJ and DLPFC, further supporting this cross-network collaboration model. Thus, CCN's involvement in task-relevant salience processing is essentially goal-directed adaptive selection rather than passive response. In complex category processing, this dynamic monitoring mechanism can balance conflicts between abstract semantic representations and concrete perceptual features.

Interestingly, reviewers noted that while behavioral results showed a main effect of salience relevance, no significant brain activation was observed for this factor. This discrepancy may be explained in two ways. First, behavioral main effects reflect macroscopic performance after cross-regional synergy, whereas fMRI captures only voxel-level average signal intensity (Haxby et al., 2001). Neural representation of task relevance may not rely on activation magnitude changes in single brain regions but rather on distributed activity patterns across multiple regions or network-level synergy (Bressler & Menon, 2010). Second, task relevance modulation may occur during pre-stimulus activation phases, while GLM analysis only captures average BOLD signals 4–6 seconds post-stimulus (Logothetis, 2008). For instance, target template (category semantics) matching may influence subsequent processing through pre-activation of visual cortex before stimulus onset (Miao et al., 2023), whereas task-irrelevant stimuli, lacking attentional templates, elicit more stimulus-driven responses. If core effects of task relevance concentrate in pre-stimulus windows, fMRI may fail to detect differences that manifest behaviorally due to incomplete temporal coverage. Therefore, the behavioral-fMRI discrepancy does not reflect measurement conflict but rather the combined product of different measurement dimensions (macroscopic synergy vs. local averaging) and analysis time windows (pre-activation vs. post-stimulus response). Future studies should combine high temporal resolution techniques (e.g., EEG/MEG) to capture full-timeline neural dynamics from pre- to post-stimulus periods and employ MVPA to uncover distributed neural coding features, thereby comprehensively revealing how task relevance modulates attentional processing through spatiotemporal synergy mechanisms.

Several unresolved issues remain. First, the semantic clarity of digits/letters may underestimate processing difficulty for ambiguous natural categories (e.g., diverse tool categories); future work should introduce prototype gradient materials (Miao et al., 2023) to test category boundary effects. Second, fMRI's low temporal resolution limits parsing of early perceptual windows (e.g., N1/P1); combining EEG (Wu & Fu, 2017) could more precisely capture the temporal dynamics of category representation. Additionally, individual differences (e.g., prefrontal decline in older adults) may amplify salience interference effects (Zhang et al., 2024), necessitating inclusion of age and clinical variables in future studies. Clinically, TMS targeting rPOJ and FEF activity may improve target-distractor inhibition capacity in patients (Kucyi et al., 2012). In summary, optimizing CAS efficiency requires balancing task goal clarity with dynamic environmental salience—a principle valuable for both theoretical construction and practical applications.

This study combined MFT and Oddball paradigms to systematically examine the roles of cognitive control and salience processing in category-based attentional selection (CAS). Behavioral results demonstrated independent effects of cognitive control and salience processing on attentional selection, with synergistic effects emerging under high load and task-relevant novel conditions. Neuroimaging results further revealed that cognitive control primarily activated DAN, while salience processing depended on VAN/SN. Within CCN, rPOJ and FEF synergistically connected the two processes: rPOJ evaluated the match between task demands and stimulus salience, while FEF translated this into executable orienting commands. Overall, CAS involves both functional division and conditional synergy between cognitive control and salience processing.

References

Arcizet, F., Mirpour, K., & Bisley, J. W. (2011). A pure salience response in posterior parietal cortex. Cerebral Cortex, 21(11), 2498–2506.

Awh, E., Belopolsky, A. V., & Theeuwes, J. (2012). Top-down versus bottom-up attentional control: a failed theoretical dichotomy. Trends in cognitive sciences, 16(8), 437–443.

Bekinschtein, T. A., Dehaene, S., Rohaut, B., Tadel, F., Cohen, L., & Naccache, L. (2009). Neural signature of the conscious processing of auditory regularities. Proceedings of the National Academy of Sciences, 106(5), 1672–1677.

Bouvier, B., Susini, P., Marquis-Favre, C., Misdariis, N. (2023). Revealing the stimulus-driven component of attention through modulations of auditory salience by timbre attributes. Scientific Reports, 13, 6842.

Bressler, S. L., & Menon, V. (2010). Large-scale brain networks in cognition: emerging methods and principles. Trends in cognitive sciences, 14(6), 277–290.

Broschard, M. B., Turner, B. M., Tranel, D., & Freeman, J. H. (2024). Dissociable roles of the dorsolateral and ventromedial prefrontal cortex in human categorization. Journal of Neuroscience, 44(34), e2343232024.

Brass, M., & von Cramon, D. Y. (2004). Selection for cognitive control: a functional magnetic resonance imaging study on the selection of task-relevant information. Journal of Neuroscience, 24(40), 8847–8852.

Campbell, J. I. D., & Thompson, V. A. (2012). MorePower 6.0 for ANOVA with relational confidence intervals and Bayesian analysis. Behavior Research Methods, 44(4), 1255–1265.

Chapman, A. F., & Störmer, V. S. (2022). Feature similarity is non-linearly related to attentional selection: evidence from visual search and sustained attention tasks. Journal of Vision, 22(8), 4, 1–19.

Chen, Q., Weidner, R., Weiss, P. H., Marshall, J. C., & Fink, G. R. (2012). Neural interaction between spatial domain and spatial reference frame in parietal–occipital junction. Journal of cognitive neuroscience, 24(11), 2223–2236.

Collignon, O., Vandewalle, G., Voss, P., Albouy, G., Charbonneau, G., Lassonde, M., & Lepore, F. (2011). Functional specialization for auditory–spatial processing in the occipital cortex of congenitally blind humans. Proceedings of the National Academy of Sciences, 108(11), 4435–4440.

Corbetta, M., Kincade, J. M., & Shulman, G. L. (2002). Neural systems for visual orienting and their relationships to spatial working memory. Journal of Cognitive Neuroscience, 14(3), 508–523.

Corbetta, M., Patel, G., & Shulman, G. L. (2008). The reorienting system of the human brain: from environment to theory of mind. Neuron, 58(3), 306–324.

Desimone, R., & Duncan, J. (1995). Neural mechanisms of selective visual attention. Annual Review of Neuroscience, 18(1), 193–222.

De Fockert, J., Rees, G., Frith, C., & Lavie, N. (2004). Neural correlates of attentional capture in visual search. Journal of Cognitive Neuroscience, 16(5), 751–759.

Fan, J. (2014). An information theory account of cognitive control. Frontiers in human neuroscience, 8, 680.

Fan, J., Guise, K. G., Liu, X., & Wang, H. (2008). Searching for the Majority: Algorithms of Voluntary Control. PLOS ONE, 3(10), e3522.

Fecteau, J. H., & Munoz, D. P. (2006). Salience, relevance, and firing: a priority map for target selection. Trends in cognitive sciences, 10(8), 382–390.

Ferrera, V. P., Yanike, M., & Cassanello, C. (2009). Frontal eye field neurons signal changes in decision criteria. Nature neuroscience, 12(11), 1458–1462.

Freedman, D. J., Riesenhuber, M., Poggio, T., & Miller, E. K. (2003). A comparison of primate prefrontal and inferior temporal cortices during visual categorization. Journal of Neuroscience, 23(12), 5235–5246.

Garrido, M. I., Kilner, J. M., Stephan, K. E., & Friston, K. J. (2009). The mismatch negativity: a review of underlying mechanisms. Clinical neurophysiology, 120(3), 453–463.

Geng, J. J., & Mangun, G. R. (2011). Right temporoparietal junction activation by a salient contextual cue facilitates target discrimination. Neuroimage, 54(1), 594–601.

Harsay, H. A., Spaan, M., Wijnen, J. G., & Ridderinkhof, K. R. (2012). Error awareness and salience processing in the oddball task: shared neural mechanisms. Frontiers in human neuroscience, 6, 246.

Haxby, J. V., Gobbini, M. I., Furey, M. L., Ishai, A., Schouten, J. L., & Pietrini, P. (2001). Distributed and overlapping representations of faces and objects in ventral temporal cortex. Science, 293(5539), 2425–2430.

Itti, L., & Koch, C. (2001). Computational modelling of visual attention. Nature reviews neuroscience, 2(3), 194–203.

Katsuki, F., & Constantinidis, C. (2014). Bottom-up and top-down attention: different processes and overlapping neural systems. The Neuroscientist, 20(5), 509–521.

Keller, A. S., Jagadeesh, A. V., Bugatus, L., Williams, L. M., & Grill-Spector, K. (2022). Attention enhances category representations across the brain with strengthened residual correlations to ventral temporal cortex. Neuroimage, 249, 118900.

Kim, H. (2014). Involvement of the dorsal and ventral attention networks in oddball stimulus processing: a meta-analysis. Human brain mapping, 35(5), 2265–2284.

Kroner, A., Senden, M., & Goebel, R. (2023). Neural correlates of high-level visual saliency models. bioRxiv.

Kucyi, A., Hodaie, M., & Davis, K. D. (2012). Lateralization in intrinsic functional connectivity of the temporoparietal junction with salience- and attention-related brain networks. Journal of neurophysiology, 108(12), 3382–3392.

Kumaran, D., Summerfield, J. J., Hassabis, D., & Maguire, E. A. (2009). Tracking the emergence of conceptual knowledge during human decision making. Neuron, 63(6), 889–901.

Lerebourg, M., de Lange, F. P., & Peelen, M. V. (2024). Attentional guidance through object associations in visual cortex. Science Advances, 10(41), eado6226.

Lavie, N. (2005). Distracted and confused?: selective attention under load. Trends in Cognitive Sciences, 9(2), 75–82.

Li, L., Gratton, C., Yao, D., & Knight, R. T. (2010). Role of frontal and parietal cortices in the control of bottom-up and top-down attention in humans. Brain research, 1344, 173–184.

Logothetis, N. K. (2008). What we can do and what we cannot do with fMRI. Nature, 453(7197), 869–878.

Macé, M. J.-M., Joubert, O. R., Nespoulous, J.-L., & Fabre-Thorpe, M. (2009). The time-course of visual categorizations: you spot the animal faster than the bird. PLoS One, 4(6), e5927.

Miao, Z., Wang, J., Wang, Y., Jiang, Y., Chen, Y., & Wu, X. (2023). The time course of category-based attentional template pre-activation depends on the category framework. Neuropsychologia, 189, 108667.

Mo, C., He, D., & Fang, F. (2018). Attention priority map of face images in human early visual cortex. Journal of Neuroscience, 38(1), 149–157.

Näätänen, R., Kujala, T., & Winkler, I. (2011). Auditory processing that leads to conscious perception: a unique window to central auditory processing opened by the mismatch negativity and related responses. Psychophysiology, 48(1), 4–22.

Noudoost, B., & Moore, T. (2011). Control of visual cortical signals by prefrontal dopamine. Nature, 474(7351), 372–375.

Oxner, M., Martinovic, J., Forschack, N., Lempe, R., Gundlach, C., & Müller, M. (2023). Global enhancement of target color—not proactive suppression—explains attentional deployment during visual search. Journal of Experimental Psychology: General, 152(6), 1705–1722.

Peelen, M. V., & Kastner, S. (2014). Attention in the real world: toward understanding its neural basis. Trends in Cognitive Sciences, 18, 242–250.

Reeder, R. R., & Peelen, M. V. (2013). The contents of the search template for category-level search in natural scenes. Journal of Vision, 13(3), 1–13.

Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive psychology, 8(3), 382–439.

Talairach, J., & Tournoux, P. (1988). Co-planar stereotaxic atlas of the human brain: 3-dimensional proportional system – an approach to cerebral imaging. Thieme Medical Publishers.

Thayer, D. D., Bahle, B., & Hollingworth, A. (2022). Guidance of attention from visual working memory is feature-based, not object-based: implications for models of feature binding. Journal of Experimental Psychology: General, 151(5), 1018–1034.

Theeuwes, J. (2010). Top-down and bottom-up control of visual selection. Acta Psychologica, 135(2), 77–99.

Townsend, J. T., & Ashby, F. G. (1983). Stochastic modeling of elementary psychological processes. Cambridge University Press.

Wang, H., Liu, X., & Fan, J. (2011). Cognitive control in majority search: a computational modeling approach. Frontiers in Human Neuroscience, 5, 16.

Wang, M., Yu, B., Luo, C., Fogelson, N., Zhang, J., Jin, Z., & Li, L. (2020). Evaluating the causal contribution of fronto-parietal cortices to the control of the bottom-up and top-down visual attention using fMRI-guided TMS. Cortex, 126, 200–212.

Wu, Q., Chang, C.-F., Xi, S., Huang, I.-W., Liu, Z., Juan, C.-H., . . . Fan, J. (2015). A critical role of temporoparietal junction in the integration of top-down and bottom-up attentional control. Human Brain Mapping, 36(11), 4317–4333.

Wu, T., Chen, C., Spagna, A., Wu, X., Mackie, M.-A., Russell-Giller, S., Xu, P., Luo, Y., Liu, X., Hof, P. R., & Fan, J. (2020a). The functional anatomy of cognitive control: a domain-general brain network for uncertainty processing. Journal of Comparative Neurology, 528(8), 1265–1292.

Wu, T., Dufford, A. J., Mackie, M.-A., Egan, L. J., & Fan, J. (2016). The capacity of cognitive control estimated from a perceptual decision making task. Scientific Reports, 6(1), 34025.

Wu, T., Spagna, A., Chen, C., Schulz, K. P., Hof, P. R., & Fan, J. (2020b). Supramodal mechanisms of the cognitive control network in uncertainty processing. Cerebral Cortex, 30(12), 6336–6349.

Wu, X., & Fu, S. (2017). The different roles of category and feature specific attentional control settings on attentional enhancement and inhibition. Attention, Perception, & Psychophysics, 79, 1968–1978.

Wu, X., Liu, X., & Fu, S. (2016). Feature and category specific attentional control settings are differently affected by attentional engagement in contingent attentional capture. Biological Psychology, 118, 8–16.

Wu, Y., Wang, J., Zhang, Y., Zheng, D., Zhang, J., Rong, M., . . . Jiang, T. (2016). The neuroanatomical basis for posterior superior parietal lobule control lateralization of visuospatial attention. Frontiers in neuroanatomy, 10, 32.

Wyble, B., Folk, C., & Potter, M. C. (2013). Contingent attentional capture by conceptually relevant images. Journal of Experimental Psychology: Human Perception and Performance, 39(3), 861–871.

Yang, H., & Zelinsky, G. J. (2009). Visual search is guided to categorically-defined targets. Vision research, 49(16), 2095–2103.

Zhang, L., Bai, L., Guo, Z., Gao, J., Wu, J., Huang, J., & Liu, Z. (2024). Abnormal functional connectivity of the occipital thalamus with the superior occipital gyrus is associated with mild cognitive impairment in elderly individuals with primary insomnia. Brain and Behavior, 14(2), e3411.