Introduction

Schizophrenia (SZ) is a complex mental disorder characterized by symptoms that include positive symptoms (such as delusions and hallucinations), negative symptoms (such as emotional flatness and social withdrawal), and cognitive deficits [1]. Although antipsychotic medications have been widely used to alleviate these symptoms, the exact neurobiological mechanisms underlying the action of antipsychotic remain unclear. The complexity theory posits that the brain operates as a system characterized by self-organization, nonlinearity, and chaotic attributes [2, 3]. The characteristic disorganization and unpredictability of thought and behavior in schizophrenia mirror the intrinsic chaos of complex systems, characterized by heightened sensitivity and instability, underscoring the relevance of complexity theory in understanding psychosis [4]. Therefore, complexity theory could serve as a valuable clinical tool to assess symptom severity and predict treatment outcomes in SZ.

Neural complexity refers to the brain’s capacity to explore different states. Various resting-state models suggest that brain activity fluctuations balance between stable and synchronous states [5,6,7,8], with rapid transitions enabling the brain to reconfigure functional networks at a low energy barrier, optimizing responses to environmental demands [9]. Stochastic resonance models show that moderate amounts of noise help subthreshold neurons reach firing thresholds, increasing neuronal firing and enhancing information processing [10,11,12]. Low complexity indicates insufficient noise, failing to promote neuronal firing effectively, leading to reduced information processing capacity. Thus, higher complexity has greater potential and propensity for information processing [13, 14]. Both the disease state and deterioration have the potential to impact the complexity of brain activity [15], leading to degradation of their information processing, thereby making individuals less adaptable to the demands of frequently changing environments [16]. This ability is frequently compromised in individuals with SZ, resulting in abnormal behavioral patterns and impaired cognitive and psychosocial functions [2]. For example, consider a random number generation test where participants were instructed to choose numbers between 1 and 10 multiple times, aiming to create sequences devoid of any noticeable pattern. The findings indicated that individuals with SZ tended to produce sequences with a higher frequency of repetitions [17]. In another examination, using the Test of Random Rhythm Generation, it was observed that individuals with SZ tended to generate time sequences that were more structured and rhythmic compared to the control group [18]. These structured sequences reveal the degeneration of information processing abilities in SZ patients, who may be more susceptible to cognitive burdens when dealing with complex tasks, thus limiting their ability to cope when faced with complex, uncertain environments.

In the realm of information theory, entropy stands as a potent gauge of the intricacy within time series data [19] and can serve as an approximate estimate of neural complexity [20, 21]. Although several studies have reported increases or decreases in brain complexity in SZ using electroencephalography [22, 23], magnetoencephalography [24] and functional magnetic resonance imaging (fMRI) [25, 26], differences in complexity measures have not been adequately characterized in regional brain regions, and most studies have assumed temporal stationarity throughout the measurement period. However, our brain is an inherently dynamic system, and performing diverse cognitive functions requires the dynamic reconfiguration of networks that interact through complex and transient communication patterns, even at rest [27, 28]. In contrast to the well-explored dynamics of functional connectivity, which captures inter-regional synchrony, the time-varying patterns of the brain activity itself are less investigated. Spontaneous fluctuations of BOLD signal are supposed to be generated from mental processes [29, 30], especially during rest, where a variety of freely occurring mental activities significantly influence the brain’s transient states. The dynamic changes in these transient states reveal the brain’s adaptive mechanisms in response to cognitive task demands [31, 32]. Therefore, studying the dynamic changes in brain complexity can help us understand the brain’s flexibility and capacity in information processing and environmental adaptation, thus providing deeper insights into cognitive deficits in schizophrenia. We hypothesize that the complexity and its dynamic changes of resting-state brain activity are impaired in SZ patients and are associated with cognitive deficits, and that this complexity and its dynamic changes may be modified after antipsychotic treatment.

To validate this hypothesis, we used weighted permutation entropy (WPE) to measure the complexity of the BOLD signal. First, we compared the abnormalities of complexity between drug-naïve first episode (DNFE) individuals with SZ and healthy controls (HC), as well as the changes before and after treatment. Subsequently, we investigated the connectivity changes in regions with abnormal complexity and analyzed the relationship between the complexity of these regions and functional connectivity strength. Finally, we used the sliding window method to explore the dynamic characteristics of resting-state brain activity complexity and investigated the relationship between abnormal dynamic characteristics and cognitive deficits in SZ patients.

Methods

Participants

A total of forty-four DNFE patients diagnosed with schizophrenia were recruited from Beijing HuiLongGuan Hospital between March 2009 and January 2010. The inclusion criteria were as follows: (1) met the DSM-IV diagnostic criteria for schizophrenia, with the diagnosis established by two trained psychiatrists using the structured clinical interview for DSM-IV (SCID) [33]; (2) an illness duration of fewer than 5 years; and (3) no previous treatment with antipsychotics, ensuring that all patients were entirely medication naïve.

Thirty healthy control subjects without any family or personal history of mental disease were recruited from the local community during the same period through advertisements. The control group was meticulously matched with DNFE patients in terms of gender, age, and education.

Additionally, any participants with major medical illness, or a history of illegal drug or alcohol abuse/dependence were excluded. All participants were right-handed.

Medication and clinical assessments

In this study, all schizophrenia patients were inpatients, and the treatment was carried out in a hospital setting. The patients were treated with risperidone, the dosage began at 1 mg/day and gradually increased to 3–6 mg/day based on the individual’s response and tolerability within the first week. These dosages were consistently maintained throughout the subsequent 11-weeks period. In cases of short-term insomnia, patients were treated with either chloral hydrate or lorazepam. Furthermore, benzhexol hydrochloride was employed to manage extrapyramidal symptoms. No other antipsychotics or antidepressants were permitted during the study. Throughout the 12-week treatment, no serious adverse effects or substance use-related comorbidities were observed. The assessment of patients’ psychiatric symptoms and cognitive functions was undertaken using the Positive and Negative Syndrome Scale (PANSS) [34] and the MATRICS Consensus Cognitive Battery (MCCB) scales [35], respectively. These evaluations were performed both prior to the baseline and after the 12-week treatment period with risperidone.

Data acquisition and preprocessing

Resting-state fMRI data were acquired on a 3 T General Electric scanner with 8-channel brain phased array. Patients were scanned both at baseline and follow-up, whereas healthy volunteers underwent scanning only once. Functional data were obtained using a gradient-echo echo-planar imaging sequence [TR/TE = 2000 ms/30 ms, flip angle = 90°, field of view = 240 × 240 mm2, matrix size = 64 × 64; slice thickness = 4 mm, voxel size = 3.75 × 3.75 × 4 mm3, 190 time points collected].

Image preprocessing used the Data Processing Assistant for Resting-State fMRI (DPARSF, V5.4, http://rfmri.org/DPARSF). The first 10 volumes were discarded to avoid nonequilibrium effects of magnetization, and slice timing and realignment correction were performed for the remaining images. Functional images were further normalized to the EPI template (resampled voxel size of 3 mm × 3 mm × 3 mm). Then several covariates including Fristion 24 motion parameters, the cerebrospinal fluid, and white matter signals were regressed as nuisance variables to reduce spurious variance. Afterwards, spatial smoothing (FWHM = 8 mm), detrending, and band-pass filtering (0.01–0.1 Hz) were conducted. Participants were excluded in the following analyses if the translational and rotational displacement exceeded 3.0 mm or 3.0°, or micromovement quantified by mean frame-wise displacement larger than 0.3 mm. Regional Bold time series were extracted with the Brainnetome (BAN) atlas [36], which includes 246 cortical and subcortical regions of interest (ROIs). Using a winner-take-all strategy, we assigned each parcel to seven canonical resting-state networks (dorsal attention, default mode, frontoparietal, limbic, somatomotor, ventral attention, visual) [37] with highest overlap proportion. In addition, subcortical regions were summarized into an eight, subcortical network.

Analysis of the WPE of bold signals

WPE is an amplitude sensitive extension of permutation entropy that preserves the useful amplitude information of the signal. The calculation of WPE involves two three steps (Fig. 1A). First, regional Bold time series are divided into overlapping windows of reasonable size. Second, the time series within the windows are divided into segments of a fixed length. Within each segment, the relative ordering of the amplitudes is obtained, resulting in a permutation pattern. Third, the probabilities of these permutation patterns are computed, considering the weights assigned to each permutation based on the corresponding amplitudes. These probabilities are then used to quantify the complexity or entropy of the signal. WPE can be mathematically expressed using the following formula:

$${H}_{{WPE}}\left(m,\tau \right)=-\sum _{i:{\pi }_{i}^{m,\tau }\in \pi }\,{p}_{w}\left({\pi }_{i}^{m,\tau }\right)\mathrm{ln}{p}_{w}\left({\pi }_{i}^{m,\tau }\right)$$
Fig. 1: Overview of the analysis pipeline.
figure 1

A Construction of WPE matrix. The sliding window method was employed to partition the Bold time series, and the WPE was calculated for each brain region within each time window (Window length of 20 TRs, 95% overlap). B Time-resolved WPE on the neural signals from each ROI in an exemplary resting-state recording. Cascades are thresholded to at least 10 regions dropping simultaneously at peak (as shown by the orange lines). The WPE vectors corresponding to the four key moments of the cascades were recorded to present the spatial topology of the propagation patterns.

Full size image

\({H}_{{WPE}}\left(m,\tau \right)\) denotes the weighted permutation entropy with motif length \(m\) and lag parameter \(\tau .\) The sum is take over all motif indices \(i\) such that the corresponding symbolic motif \({\pi }_{i}^{m,\tau }\) belongs to the set of possible motifs. The term \({p}_{w}\left({\pi }_{i}^{m,\tau }\right)\) represents the variance-weighted relative frequency of motif \(i\), and \(\mathrm{ln}\) denotes the natural logarithm. Based on methodological considerations [38, 39] and previous applications of WPE in neural data analysis [40], we utilized a motif length of m = 3 and a lag parameter of τ = 1 to calculate the motif distribution across the target signal vector. Furthermore, the WPE values were normalized within the range of [0,1]. Here, the sliding window method was utilized. In the main text, we present the results obtained using a window length of 20TRs (40 s) with 95% overlap. This configuration provides a temporal resolution of 1TR (2 s) and generates a time-resolved signal complexity vector comprising 165 windows. In order to evaluate the influence of windowing parameters on the observed complexity dynamics, we additionally investigated window lengths of 20 samples and 30 samples, with overlap percentages of 95% and 90% respectively. Notably, these alternate setups produced remarkably consistent results, highlighting the robustness of our findings across various windowing configurations.

Functional connectivity analysis

We defined the brain regions with abnormal complexity as seed regions of interest (ROIs) and calculated the Pearson correlation coefficients between the ROIs and other brain regions. Subsequently, we computed a connectivity metric known as strength. The strength of node is defined as the mean value of the FC between the node and the rest of the brain’s nodes.

Complexity drop and propagation

As shown in Fig. 1B, by transforming the WPE matrix into the form of a line graph, we observed pattern of predominant high complexity with recurrent complexity drops. To study the propagation of these drops, we defined a threshold of drop engagement as the first percentile of the total WPE distribution (critical WPE = 0.335). The number of brain regions with complexity dropping in each time window was recorded, resulting in a line graph where the x-axis represents time windows and the y-axis represents the number of regions with complexity drop.

To investigate the propagation of complexity drops in the brain, we introduced the concept of a drop cascade (as shown by the orange lines in the Fig. 1B): all Bold windows with at least n = 10 regions simultaneously exhibiting a complexity drop in a signal time window, and if several consecutive time windows met this criterion, the window with the largest number of dropping regions was defined as the peak layer of the cascade. Starting from the peak layer, a search was conducted both forward and backward to identify adjacent windows where the number of dropping regions gradually decreased. The search continued until a window was reached where the number of dropping regions either remained the same or increased. To investigate the spread of complexity drops in the propagation phase, the WPE vectors of the start layer, half layer, peak layer, and end layer were extracted from all drop cascades. The node probability of each brain region in the respective layer was computed as the rate of occurrence in the respective window. Generate the node probability spatial topology at the group level for the HC, SZ0W, and SZ12W.

Statistical analysis

We performed group-level statistics on the mean WPE vector in healthy controls, patients at baseline, and follow-up. In this study, the normality of the data was assessed using the D’Agostino & Pearson omnibus normality test. Homogeneity of variances was checked with Levene’s test. Where assumptions of normality or homogeneity of variances were violated, appropriate non-parametric tests were employed. Two sample t test (two-tailed) was conducted to compare WPE and propagation properties between SZ0W (n = 40) and HC (n = 30) while controlling age, gender, and education. Paired-sample t test (two-tailed) was conducted to evaluate the longitudinal alteration of WPE and propagation properties between SZ0W (N = 23) and SZ12W (n = 23). The multiple comparison corrections were conducted using two methods for different spatial scales: network level and brain region level. For network level, Bonferroni correction was conducted with a significant level of P < 0.05. For brain region level, false positive effects were controlled using False Discover Rate (FDR, P < 0.05) correction and Bonferroni correction for baseline and pre-post treatment comparisons, respectively.

The mean WPE value of the regions, which showed significant differences either between SZ0W and HC, or between SZ0W andSZ12W, was extracted to a post hoc comparison. Additionally, the Pearson correlation analysis was conducted to confirm the relationship between either baseline WPE and clinical symptoms (MCCB scores and PANSS scores), or longitudinal alterations in WPE and improvement of clinical symptoms (PANSS scores of SZ0W -PANSS scores of SZ12W). Furthermore, we examined the relationship between baseline WPE and the rate of reduction in clinical symptoms ((PANSS scores of SZ0W -PANSS scores of SZ12W)/PANSS scores of SZ0W). Bonferroni correction was applied to the regions that displayed significant differences (4 regions) for the p value, with P < 0.05 as the significance threshold.

Results

Demographics

Participant demographic characteristics are presented in Table 1. Following a 12-week risperidone treatment, a total of twenty-seven patients fully participated in the experiment, engaging in both clinical assessments and the acquisition of resting-state brain imaging data post-treatment. All PANSS three subscale scores and total score were and total score were significantly reduced (all P < 0.05; Table 2). MRI data of 4 SZ0W and 1 SZ12W data were excluded due to the head motions. To perform paired-sample t test, we further excluded three post-treatment data points due to excessive head motion observed in these three cases prior to treatment. Consequently, we collected fMRI data from 30 HC, 40 SZ0W and 23 SZ12W. The demographic and clinical characteristics of participants included for analysis are shown in supplementary Table S1 and Table S2.

Table 1 Demographic, clinical characteristics, and cognitive data of participants at baseline (Mean ± SD).
Full size table
Table 2 Clinical characteristics before and after 12 weeks of treatment with risperidone (Mean ± SD).
Full size table

Spatial distribution of WPE and intergroup differences at network level

Spatial analysis revealed that higher WPE distributions are observed in the medial temporal lobe, anterior cingulate cortex, and subcortical regions, whereas cortical regions exhibit lower WPE values in comparison (Fig. 2A). The spatial topology of signal complexity was robust across functional parcellation granularity (Fig. S1, Supporting Information). To examine the abnormalities of WPE at network level, We partitioned the brain network using a winner-take-all strategy, and the results are shown in Fig. 2B. We averaged WPE within each network for each subject and then compared the intergroup differences in WPE within each network at baseline and pre-/post-treatment. Schizophrenia patients exhibited significantly lower mean WPE in the SC (t = 3.79, PBonf = 0.0003). Following a 12-week risperidone therapy, no significant differences were observed at the network level (Fig. 2C, Bonferroni correction, P < 0.05/8). This result remained consistent across different brain network parcellation methods (Table S3, Supporting Information).

Fig. 2: Spatial distribution of average WPE and inter-group differences at the network level.
figure 2

A Spatial distribution of average signal complexity. B Brain parcellation (246 parcels) and network affiliations (8 networks) used in the study. C Group differences in network-level WPE. Left: Network level differences between SZ0W and HC were accessed by two sample t test, and a significant difference was observed within the subcortical (Bonferroni correction, ***P < 0.001, 8 comparisons). Right: Network level differences between SZ12W and SZ0W were accessed by paired-sample t test, and no significant difference was observed. The shaded patches indicate 95% confidence intervals. DAN dorsal attention, DMN default mode, FPN frontoparietal, LIM limbic, SMN somatomotor, SC subcortical, VAN ventral attention, VIS visual network.

Full size image

Baseline comparisons between patients and healthy controls

At the brain region level, comparisons between groups showed that SZ0W had reduced WPE compared to HC in the left ventral caudate (t = −3.59, PFDR = 0.019), right ventral caudate (t = −4.17, PFDR = 0.008), left pallidum (t = −3.46, PFDR = 0.017) and left putamen (t = −3.74, PFDR = 0.021) (Fig. 3A, B). The result was robust across functional parcellation granularity and windowing parameters (Table S4 and S5, Supporting Information).

Fig. 3: Differences in WPE among brain areas at baseline and their correlation with cognitive performances.
figure 3

A Group differences in ROI level WPE between the HC group and the SZ0W group were accessed by two sample t test. Significant differences (PFDR < 0.05, 246 comparisons) were found in four subcortical regions. B Two groups WPE for the four ROIs (*PFDR < 0.05, **PFDR < 0.01). C Correlations between MCCB scores (y axis) and WPE of the left caudate (x axis) in SZ0W group (Bonferroni corrected, 4 comparisons).

Full size image

The WPE of left ventral caudate exhibited a significant positive correlation with the patients’ MCCB Continuous Performance Test (r = 0.48, Puncorr = 0.005, PBonf = 0.038) and MCCB Category Fluency Test(r = 0.37, Puncorr = 0.040, PBonf = 0.159). Additionally, a negative correlation was observed between the WPE of left ventral caudate and the MCCB Digital Sequence Test (r = −0.49, Puncorr = 0.009, PBonf = 0.021) (Fig. 3C).

Follow-up comparisons of WPE before and after treatment

After 12 weeks of risperidone therapy, patients showed increased WPE in the left ventral caudate (t = 3.78, PBonf = 0.004), but no difference was found in other areas (Fig. 4A, B).

Fig. 4: Differences in brain region-specific longitudinal WPE changes and their correlation with symptoms.
figure 4

A Group differences in ROI level WPE between the SZ0W group and the SZ12W group were accessed by paired-sample t test. Significant difference (Bonferroni corrected, 4 comparisons) was found only in left ventral caudate. B Two groups WPE for the four ROIs (**PBonf < 0.01). C Correlations between the reduction rate of the subscores and the total score of the PANSS scale (y axis) and WPE of the right ventral caudate before treatment (x axis) in SZ12W group (Bonferroni corrected, 4 comparisons).

Full size image

To test the relationship between the changes in WPE of the left ventral caudate and the reduction of clinical symptoms (measured by the changes in PANSS scores before and after treatment), Pearson correlation analysis was conducted. However, no significant correlation was observed. Further, after 12 weeks of risperidone treatment, the baseline WPE of right ventral caudate was associated with the reduction rate of PANSS positive score (r = 0.53, Puncorr = 0.01, PBonf = 0.040), PANSS negative score (r = 0.48, Puncorr = 0.032) and PANSS total score (r = 0.46, Puncorr = 0.025) in patients (Fig. 4C).

Widespread lower FC in patients and the relationship between FC strength and WPE

It is intuitively plausible that the complexity of individual time series might influence their correlation with other time series. Having identified the brain regions exhibiting abnormal complexity in patients, we investigated the relationship between the complexity of these regions and their functional connectivity.

In Fig. 5A, the spatial analysis revealed similar spatial distribution patterns in the connectivity difference maps of the four brain regions, with most areas showing decreased connectivity in patients compared to the control group. After FDR correction, widespread abnormal connectivity was observed with the right ventral caudate and left putamen as seed points. Compared to HC, SZ patients showed decreased connectivity between the right ventral caudate and regions such as the inferior frontal gyrus, superior temporal gyrus, thalamus, and inferior parietal lobule. Similarly, decreased connectivity was found between the left putamen and regions including the dorsolateral prefrontal cortex, inferior frontal gyrus, inferior parietal lobule, thalamus, precentral gyrus, and postcentral gyrus (Fig. 5A, Table S6, Supporting Information).

Fig. 5: Relationship between WPE and connectivity.
figure 5

A The patterns of functional connectivity differences between the SZ0W group and the HC group. Functional connectivity with the rest of the brain was computed separately for each of the four ROIs as seed points (two sample-test, top: uncorrected; bottom: PFDR < 0.05, 245 comparisons). Edges with cool and warm colors refer to significantly decreased and increased functional connectivity in SZ0W compared to HC, respectively. B Correlations between the WPE and the FC strength of the four ROIs (Uncorrected).

Full size image

Figure 5B showed that individuals with higher WPE values exhibit stronger FC strength in the right ventral caudate (r = 0.28, Puncorr = 0.021), left pallidum (r = 0.30, Puncorr = 0.012) and left putamen (r = 0.28, Puncorr = 0.021).

Complexity drop and propagation

The dynamic changes in brain activity complexity have the following characteristics: they usually exhibited high complexity activity, but this high complexity activity was interrupted by spontaneous and recurring complexity drops (Fig. 1B). This architecture of complexity was consistently present across all participants.

We investigated the propagation pattern of complexity drops across the whole brain over time windows by defining the concept of drop cascade. As shown in Fig. 6A, the propagation of brain activity is illustrated at four time points corresponding to the cascade stages: the star, the half, the peak, and the end, representing the initial phase, propagation phase, high peak phase, and recovery phase of complexity drops, respectively. The results revealed that, in the start and end layers, HC group showed a greater occurrence of node probability distribution in the left frontal lobe region, while the SZ0W group showed a greater occurrence in the right frontal lobe region. Following treatment, the node probability distribution in the patient group tended to approach that of the HC group. Compared to HC, in the half layer, the SZ0W group exhibited a noticeable contraction in the scope of node probability distribution. In the peak layer, the SZ0W group showed a decrease in probability values in the sensorimotor, occipital cortex and their peripheral cortex. After 12 weeks of risperidone treatment, a partial recovery was observed in this distribution. The topographic maps depicted the distribution of node probabilities in the HC, SZ0W and SZ12W groups, as illustrated in Fig. 6B. The patterns of the propagation of complexity drops was robust across windowing parameters (Figs. S2–S4, Supporting Information).

Fig. 6: Patterns of the propagation of complexity drops.
figure 6

A Patterns of the propagation of complexity drops in the HC group, SZ0W group and SZ12W group, respectively. B Peak landscape plots depicted the distribution of propagation probabilities (The x-axis represented propagation probabilities, while the y-axis represented the number of brain regions).

Full size image

Furthermore, the statistical analysis of the propagation properties is shown in Fig. 7. Compared to HC, the SZ0W group exhibited a significant increased in the number of drop cascades (t = 2.94, P = 0.005), indicating that the complexity drop phenomenon is more frequent. Additionally, the duration of cascades (Start-End, P = 0.001, Mann–Whitney test) and the first half of the cascades (Start-Peak, t = −2.78, P = 0.007) were significantly reduced (Fig. 7A). After 12 weeks of risperidone therapy, compared to the SZ0W group, the SZ12W group showed a decrease in the number of cascades (t = −3.07, P = 0.006) and an increase in the duration of cascades (P = 0.038, Mann–Whitney test), as shown in Fig. 7B. Additionally, we found that the duration of cascades in the SZ0W group was negatively correlated with continuous performance test scores (r = −0.41, P = 0.032) and positively correlated with both the general psychopathological symptom score (r = 0.36, P = 0.024) and total score (r = 0.37, P = 0.021) of the PANSS scale (Fig. 7C).

Fig. 7: Propagation properties of the complexity drops.
figure 7

A Group differences in propagation properties between the HC group and the SZ0W group were accessed by two sample t test. B Group differences in propagation properties between the SZ0W group and the SZ12W group were accessed by paired-sample t test. C Correlations between the duration of cascade and cognitive deficits and clincal sympotos (uncorrected). *P < 0.05, **P < 0.01. Start-End: overall duration of cascades; Strat-Peak: the first half of the cascade duration; Peak-End: The second half of the cascade duration. (Mean ± SD).

Full size image

Discussion

This study examined the effects of risperidone on brain complexity and its dynamic changes in DNFE patients with schizophrenia using resting state fMRI. At baseline, patients displayed reduced complexity primarily in striatal regions and left pallidum, with no significantly increased brain areas. Reduced complexity in the left caudate was positively correlated with cognitive performance. After treatment, complexity in the left caudate increased. Regions with abnormal complexity showed decreased functional connectivity, and complexity was positively correlated with connectivity strength. The dynamic analysis revealed recurring “complexity drops” in the brain’s resting state, which were less scope, less intensity, and shorter in duration in SZ patients but improved with treatment. The reduced duration of these complexity drops was negatively correlated with cognitive deficits and positively with clinical symptoms.

The caudate and putamen serve as the main input nuclei to the basal ganglia, receiving axons from various regions of the cortex. The reduced complexity in the caudate nucleus and the putamen reveals a novel form of functional impairment associated with schizophrenia, characterized by a degradation in the processing capabilities of the striatum. This finding is consistent with previous research investigating localized activity and connectivity within the striatum, where schizophrenia patients exhibited decreased temporal brain entropy in the caudate nucleus [41], as well as reduced sample entropy in the putamen [42] compared to healthy controls. Although employing different complexity metrics, we posit a strong interrelation between these measures. Neural complexity reflects the brain’s ability to dynamically transition between different states [43], and the complexity of neural signals can serve as an indicator of the brain’s information processing capability [13]. Higher complexity may suggest that the brain transitions between states or explores in a more diverse manner, thereby facilitating information processing [14]. Therefore, reduced complexity in the striatum may indicate a lack of diversity and flexibility in brain information processing, resulting in a decline in processing capability.

Intuitively, the complexity inherent in individual time series could influence their correlation with other time series, as regions characterized by higher complexity are likely to flexibly enhance or weaken information communication to maintain or alter mutual stimulation or inhibition with other functional areas [44, 45]. Our findings substantiate this perspective, as we observed that four brain regions exhibiting diminished complexity exhibited reduced functional connectivity with the majority of other brain regions. Moreover, the individual complexity of the striatal regions showed a positive correlation with the intensity of functional connections with other brain regions. This implies a connection between the decline in striatal information processing capacity and a weakening of its interactions with other regions of the brain. Studies with fMRI have shown reduced connectivity between the cortex and the striatum in patients with schizophrenia [46,47,48]. Furthermore, diffusion tensor imaging revealed reduced anatomical connectivity between the dorsolateral prefrontal cortex and the striatum in schizophrenia patients [49, 50]. The diminished functional connectivity could impact the overall integration of brain activity, subsequently influencing the manifestation of psychiatric symptoms and cognitive deficits [51, 52].

Some studies have suggested that the reduced complexity of brain activity in schizophrenia is related to cognitive performance and psychiatric symptoms [23, 53]. Our research results indicated that the reduced complexity in the left caudate nucleus of patients was positively correlated with Continuous Performance Test and Category Fluency, suggesting that the degradation of information processing capabilities might be a potential pathological mechanism underlying cognitive deficits in attention, response control, and information processing speed in SZ patients. Increasing evidence highlights the crucial role of the striatum in cognition [54, 55]. Progressive changes (reductions) in the volume of the striatum (caudate nucleus and putamen) in schizophrenia are associated with poorer performance in attention, problem-solving, and working memory tests [56]. The neural activity in striatal regions, such as the caudate nucleus, is central to attention [57]. The functional connectivity between the cortex and the striatum is closely associated with sustained attention or vigilance [58, 59]. Reduced structural connectivity is linked to poorer cognitive flexibility and slower information processing speed in patients with schizophrenia [60, 61]. Some studies suggest that the striatum may influence cortical information processing through the pathway involving the pallidum-thalamus-cortex. As the output nucleus of the striatum, the pallidum is responsible for transmitting signals from the striatum to the thalamus. Reduced complexity in the striatum and pallidum leads to a deterioration of their information processing capability, potentially resulting in negative effects on motor control, cognitive function, and other neural processes related to this pathway. In summary, the reduced complexity and weakened functional connectivity in the striatum are consistent with various impairments in higher-order brain functions observed in schizophrenia.

The increase in complexity of the left caudate nucleus after treatment indicates that short-term monotherapy with risperidone enhances spontaneous neural activity in this region. The enhancement in complexity not only reflects increased neural activity but also may indicate a restoration of brain function. We found that the WPE of the right caudate nucleus before treatment was associated with improvements in patients’ clinical symptoms, suggesting that greater complexity in the right caudate nucleus is related to a better treatment response. This is supported by related studies. A longitudinal study found that the amplitude of low-frequency fluctuations in the caudate nucleus increased after short-term antipsychotic treatment in first-episode schizophrenia patients, and this increase was associated with a reduction in clinical symptoms [62]. Additionally, task activation studies have shown that unmedicated patients exhibit lower BOLD activity in the caudate nucleus during the Stroop task, whereas this activity significantly increases after treatment [63]. These findings suggest that antipsychotic treatment can improve the functional state of the caudate nucleus. Complexity, as an important indicator of brain function, can reflect changes in the brain’s information processing capabilities and functional state. Therefore, the restoration of complexity induced by treatment may be a manifestation of brain function improvement.

In our study, we found a significant correlation between dynamic complexity abnormalities and cognitive deficits in patients with SZ. Specifically, patients exhibited more frequent complexity drop phenomena, which were shorter in duration, and had reduced propagation scope and intensity. Complexity drop phenomena may result from the brain’s reconfiguration when handling attention shifts or other cognitive tasks. For example, Krohn et al. found that complexity drop phenomena in resting-state brain activity might represent a neural activity state [64]. Furthermore, Chen et al.‘s research indicated that complexity temporarily decreases when subjects shift their attention, suggesting that when the brain state changes abruptly, the complexity of information transmission temporarily declines [65]. This phenomenon might be the brain’s response to reconfiguring resources and adapting to task demands, but in patients, this reorganization process appears more frequent and short-lived, unable to maintain stability. This situation may lead to insufficient processing and integration of information by the brain, thereby affecting overall cognitive performance. As our findings indicate, the reduced duration of complexity drops is associated with poor CPT performance, suggesting that patients have difficulty maintaining stable attention, resulting in slower reaction times and increased error rates [66]. Although risperidone treatment led to improvements in complexity indicators, we did not find a significant direct correlation between these improvements and cognitive enhancement. This may indicate that while risperidone positively affects brain complexity indicators, these changes may not directly translate into significant cognitive improvement. Additionally, this result may also be influenced by sample attrition, which limited our ability to statistically analyze the relationship between complexity indicators and cognitive function.

In a comprehensive view, the decline in complexity, abnormalities in the propagation of complexity drop phenomena, and post-treatment recovery may represent indirect mechanisms underlying psychiatric symptomatology and cognitive deficits in patients with schizophrenia, as well as improvements following treatment. Notably, the striatum emerges as the region with the most pronounced complexity abnormality in the current study. Studies have indicated that the most significant dopamine alterations in schizophrenia patients are localized in the dorsolateral striatum, especially the caudate nucleus [67, 68]. Hyperdopaminergia in the striatum could potentially disrupt signal transmission between the prefrontal cortex and the striatum, leading to imbalanced cortical dopamine and subsequently affecting information processing and cognitive integration [69, 70]. Notably, the left caudate nucleus is the only region that exhibited increased WPE values after treatment. Studies with PET have revealed strong correlations between the occupancy of dopamine D2receptors in the striatum and reductions in positive symptoms [71]. D2receptor effects of antipsychotic drugs on the striatum may serve to organize and modulate thalamocortical drive to alter neocortical function and reduce psychotic symptoms. As dopamine D2receptor antagonists, antipsychotic drugs increase glutamate levels via corticostriate projection fibers [72]. Since glutamatergic synapses are key excitatory synapses in the brain, this increased is expected to boost local neural activity, as indicated by the WPE increases observed after treatment in this study. Antipsychotic drugs also modulate γ-aminobutyric acid (GABA), with first-episode schizophrenia patients showing higher GABA levels in the caudate nucleus compared to HC, normalizing after 4 weeks of risperidone treatment [73]. GABA is an inhibitory neurotransmitter, and elevated GABA levels in schizophrenia patients may reduce local neural activity. Antipsychotic drugs may lower GABA levels through their action on dopamine D2 receptors [74], thereby enhancing local neural activity, consistent with the abnormal changes in caudate nucleus WPE observed in this study before and after treatment. Risperidone, as a typical antipsychotic medication, may address these complexities and their dynamic changes through multiple mechanisms. These include the regulation of dopamine D2 receptors, modulation of glutamate and GABA levels, and ultimately improving cognitive and psychiatric symptoms.

Limitation

Several limitations of the current study must be noted. First, the small sample size may have reduced statistical power, and a larger sample size is required to confirm the current results in future studies. Second, there is no patient control group receiving other types of antipsychotics, so we cannot determine whether the effects observed in this study were specific to risperidone. Third, WPE is an extension of PE. However, it does not take into account the complexity of BOLD signals at different time scales. Further exploration is need to investigate the impact of disease states on brain activity at different time scales. Forth, the research on the propagation patterns of complexity drop is based on the group level. Further exploration is need at the individual level to investigate the impact of SZ on the propagation patterns. Finally, we used univariate analysis to map imaging metrics to behavioral variables, helping to qualitatively understand cognitive mechanisms and explore the clinical application of imaging biomarkers. However, multiple factors influence this relationship, suggesting the need for multimodal analysis in future studies.

Conclusion

Despite the limitations, our study reveals reduced brain activity complexity and its abnormal dynamic changes in SZ, and a reversal of these abnormalities after antipsychotic treatment. The reduction in complexity and frequent and short-lived complexity drops may be potential mechanisms underlying information processing and cognitive deficits. The reversal of these changes may indicate the treatment effect on neurophysiology, leading to improvements in clinical symptoms and cognitive deficits. These findings offer novel insights into the pathophysiological mechanisms and therapeutic interventions for schizophrenia.