The flow diagram of study selection is shown in Figure 1. Four English databases (Cochrane, Embase, Web of Science, and PubMed) and four Chinese databases [Chinese Biomedical Literature Database (CBM), China National Knowledge Infrastructure (CNKI), Wanfang databases, and Chinese Scientific Journal Database (VIP)] were systematically searched from inceptions to April 10th, 2023. We retrieved trials which explored the neuro-responses of depressive patients in response to explicit emotion recognition tasks. The search strategies were combined with Medical Subject Headings (MeSH) and free text words related to the following terms: depression, emotion recognition, and neuroimaging. We also reviewed the reference lists of relevant articles to identify additional literature. Detailed search strategies are provided in Supplementary Table 1.

The pre-specified inclusion criteria were as follows:

Studies were excluded if they met any of the following criteria: (1) protocol, abstracts, review, or case report; (2) patients with bipolar I disorder, sub-threshold depression, postpartum depression or depression secondary to brain injury, Parkinson's Disease, Alzheimer's Disease, etc.; (3) no details of diagnostic criteria; (4) duplicated articles or overlapping subjects; (5) imaging was performed via structural imaging, magnetic resonance spectroscopy, or functional near-infrared spectroscopy; (6) resting-state, brain functional connectivity, or brain activation-functional correlations studies; (7) studies focused on regions of interest analysis; or (8) data could not be extracted.

We imported retrieved records into Endnote software (X9), and duplicates were removed. Two reviewers (JZ and XYG) independently screened titles and abstracts to exclude irrelevant records. Then, the remaining records in the full text were scrutinized according to pre-specified eligible criteria. In case of overlapping samples between studies, we retained the study with a larger sample size. Any disagreement was resolved by team discussion until a consensus was reached.

Two independent reviewers (XBL and DLZ) extracted data with a pre-defined data extraction form. The extracted data included study characteristics (author, publication year, sites), participant information (sample size, sex ratio, age, diagnostic criteria, and subtypes and severity of depression), task performance (task paradigms and stimulus types), fMRI acquisition and analysis (fMRI acquisition, MRI processing, and software version) and results of imaging (spatial coordinates of vertices and three-dimensional coordinates system).

After extraction, two reviewers cross-checked data to ensure accuracy. Any discrepancy was arbitrated by the third reviewer (JL).

Two reviewers (YXL and HSX) independently assessed the risk of bias of included studies using the modified version of the Newcastle–Ottawa scale (NOS) (35, 36) adapted to fMRI. The tool contains 11 methodological items which are categorized into 4 domains (selection, comparability, exposure, and statistical analysis). If the item was addressed properly, it was rated as “+”; “–” indicated that the condition was not met; and “?” meant that the information was not clear or missing. Referring to the study of Gentili et al. (36), the specific criteria is shown in Supplementary Table 2. A third author (RJJ) supervised the entire process to resolve discrepancies.

This ALE analysis was performed with Ginger-ALE 2.3.6 software (www.brainmap.org/ale). Talairach coordinates were converted into MNI coordinates using Ginger-ALE 2.3.6; all coordinates were presented in the MNI space. We used cluster-level family wise error (FWE) (cluster p = 0.05, 1,000 permutations, uncorrected p = 0.001) to explore the spatial consistency of overall increased activity and decreased activity, respectively. Additionally, we used the Ginger-ALE users' manual as the reference and conducted uncorrected p-value with a conservative threshold (uncorrected p < 0.001), and removed clusters under a user-chosen size by setting the Min. Volume as 250 mm³ to find additional overlap and decrease the possibility of false-negatives. MRIcron and a common anatomical template in MNI space (ch2bet.nii.gz) were used for visualizing the results. We conducted two overall analyses of emotion processing based on the brain activation direction across negative and positive stimuli.

Subgroup analyses were conducted based on different stimulus types (negative and positive), age of the participant (age ≦ 18 and age > 18), and severity of depression.

A whole-brain voxel-based jackknife sensitivity analysis was performed to test the robustness of the results by removing one dataset each time.

The significant differential clusters of explicit tasks between patients with depression and healthy participants were reported in all studies. Each abnormal brain region was extracted to calculate the number of replicates in the overall analysis, negative stimuli analysis, and positive stimuli analysis, separately.

In total, 11,365 articles were retrieved from eight databases. We identified 8,383 records after removing duplicates, and reference tracking yielded five additional records. After screening the titles and abstracts, 403 articles were retained for full-text reading. As a result, we included 28 eligible cross-sectional studies. The detailed selection procedure is shown in Figure 1. The reasons for exclusion are listed in Supplementary Table 3.

The characteristics of included studies are shown in Table 1. Among 28 included studies, 20 studies (17–19, 40–47, 49, 50, 52, 53, 55, 57–59, 61) were published in English and 8 (37–39, 48, 51, 54, 56, 60) were in Chinese. A total of 656 patients with depression and 680 healthy subjects were involved. The average age of participants ranged from 15 to 66 years old, and only 2 studies (40, 55) focused on youths with depression. Depressive patients in 20 studies (17–19, 40–47, 49, 50, 52–55, 57, 59, 61) were diagnosed according to the DSM, 1 (37) in the light of CCMD, and 2 (51, 58) based on ICD; 3 studies applied DSM-IV and CCMD-3 (38, 39, 60); 1 used both DSM-IV and ICD-10 (56); 1 adopted CCMD-3 and ICD-10 (48). Hamilton Depression Rating Scale, Beck Depression Inventory, Montgomery–Asberg Depression Scale, Center for Epidemiologic Studies Depression Scale, and Children's Depression Rating Scale-Revised were adopted to screen depression. Tasks used in included studies were classified into the following three categories: emotion categorization (37, 46, 51, 52, 54, 59), emotion matching task (40–42, 44, 47, 57, 61), and emotion perception (17–19, 38, 39, 43, 45, 48–50, 53, 55, 56, 58, 60) (including valence assessment, emotion judgment, and elicited emotion rating). 6 studies (17–19, 40, 50, 55) performed tasks with alternation of positive stimuli and negative stimuli, the results were the reflection of overall task performance; 20 studies (37–39, 41–49, 51, 54, 56–61) analyzed negative stimuli; and 11 studies (37–39, 43, 46, 49, 52, 53, 56–58) focused on positive stimuli. As for imaging tools, 21 studies (17, 18, 40–43, 46, 48–61) acquired images on a 3T scanner, and 6 studies (37–39, 44, 45, 47) on a 1.5T scanner. 21 studies (17, 18, 37–39, 42–47, 49–54, 56, 58–60) used SPM software for pre-processing and analysis, three studies (40, 48, 55) used AFNI software, and 2 studies (41, 61) used FSL software. The details of fMRI acquisition and analysis are shown in Supplementary Table 4.

The detailed information of each included study is shown in Supplementary Table 5. The selection part contained two items for included patients and healthy controls respectively. In the domain of case definition, the diagnosis of all studies was based on standard criteria, which was considered as “+”. As for the representativeness of the cases, there were 3 studies (38, 51, 54) that included only women or men, which was judged as “–”; and 1 study (58) was rated as “?”, because it did not provide information on the sex of included subjects; a defined procedure for inclusion was reported in the remaining studies, which were appraised as “+”. In the selection of controls, 13 studies reported that the healthy controls were recruited from the same community of individuals with depression; while 6 studies (17, 43, 48, 50, 52, 53) got “–” because the sources of healthy subjects were different from depressive patients; 9 studies (37, 39, 42, 45–47, 51, 58, 60) did not state the origin of the healthy subjects, and were determined as “?”. With reference to definition of controls, psychiatric assessment of healthy subjects were not conducted in 2 studies (43, 53), were judged as “-”, and 7 (19, 38, 41, 42, 45, 47, 58) did not specify the screening tool for healthy subjects, thus were considered as “?”.

Regarding the comparability of studies, it contained two aspects: age/gender and other variables. Among 26 studies, the age and gender were matched between depression and healthy controls, which were regarded as “+”; 2 studies (46, 59) did not mention the resemblance of age or sex between the two groups, then were considered as “?”. As for the comparability of other variables, 6 studies (41, 45–47, 53, 59) were rated as “?” because they did not provide the comparability of other demographic variables, the remaining studies were all reported clearly, and were rated as “+”.

For the exposure, all the included studies reported that the same experimental procedure was performed in depressive patients and healthy subjects, thus were considered as “+”. Eleven studies reported the rate and causes of drop-out, and were rated as “+”; 15 trials (17, 18, 38–40, 43, 46, 48, 50, 52, 55, 56, 58, 60, 61) did not report the drop-out rate and relevant reasons, thus were evaluated as “-”; and 2 studies (51, 53) reported drop-out rate without explanation, then were rated as “?”. All the studies used behavioral tasks which could cause a significant difference between depressive patients and healthy controls, which were judged as “+”.

In the statistical analysis part, p-value > 0.001 was reported in 7 studies (40, 41, 44, 47, 51, 53, 54), which were regarded as “–”; 13 studies (17–19, 37, 43, 45, 50, 52, 55–57, 60, 61) without the description of uncorrected thresholds were considered as “?”; the remaining studies were classified as “+”. 10 (37–39, 42, 45–47, 49, 51, 59) out of 28 studies were rated as “–” due to a lack of false positive correction; 2 studies (56, 57) performed false positive correction without detailed description, and were appraised as “?”.

The jackknife sensitivity analysis results of the decreased activity revealed that the STG was replicable in all combinations; the inferior parietal lobule and superior frontal gyrus remained significant in 20 out of 22 combinations; insula sustained significance in 19 out of 22 combinations. The sensitivity analysis of increased activity showed MTG, STG, and PHG were significant in all but two combinations; the cuneus presented significance in all but three combinations. Details are shown in Supplementary Table 6.

We calculated the times that each brain region was mentioned in the 28 studies: the inferior parietal cortex was the most mentioned brain region (15 times), followed by the cingulate cortex (14 times), MFG (12 times), MTG (12 times), precuneus (11 times), insula (11 times), STG (11 times), middle occipital cortex (10 times), fusiform gyrus (10 times), and precentral gyrus (9 times). Considering the abnormal brain activation areas in response to negative stimuli, the top 10 most mentioned brain regions were MFG (12 times), cingulate cortex (10 times), IPL (9 times), precuneus (8 times), STG (7 times), superior frontal cortex (7 times), insula (6 times), inferior temporal cortex (6 times), precentral gyrus (6 times), and fusiform gyrus (6 times). In the positive stimuli, MFG (6 times), inferior temporal cortex (6 times), MTG (6 times), IPL (6 times), STG (5 times), insula (4 times), middle occipital cortex (4 times), precentral gyrus (4 times), cingulate cortex (3 times), and amygdala (3 times) were mentioned most. The results of each dataset are shown in Table 1 and presented in Figure 3.

Psychoradiology is an interdisciplinary subject of radiology and psychosis, which uses imaging techniques to detect the structural or functional activity of the brain in healthy populations or in people with psychiatric disorders (62–65). It breaks through the traditional diagnostic methods (scale assessment and subjective judgment of clinicians) and performs quantitative analysis to observe the subtle changes in the brain which is able to optimize the diagnosis, treatment, and monitoring of psychiatric disorders (62). In the present study, to investigate the brain activation changes in response to explicit emotion processing tasks, we pooled the results from 28 neuroimaging studies and the results indicated that people with depression had abnormal brain activities in contrast to healthy subjects during the explicit emotion recognition tasks. The results showed that MTG (BA39), STG (BA39), PHG (BA36), and cuneus (BA7) displayed a greater likelihood of hyperactivity, while the IPL (BA 40), bilateral STG (BA 22), insula (BA 13), and SFG (BA 8) exhibited hypoactivity.

We mapped these brain regions to functional networks according to the Stanford Atlas of Functional ROI. As for activated

brain regions, default mode network [DMN, including PHG, STG (BA39), cuneus, and MTG] was involved. DMN is related to rumination and cognition functions, as well as the self-referential processes. The over-activation of the DMN reflects excessive self-referential thoughts, which are related with negative cognition. Chao et al. (66) proved that the dysregulation of the DMN played an important role in depression. Xu et al. (67) discovered an increased activation of DMN in depression during the resting state. As the network node of DMN, PHG is the center of the cognitive systems in emotion recognition (68). The activation of PHG is related with emotional memory encoding (69). Chen et al. (70) found a positive correlation between the activation of right PHG and depressive symptoms. The cuneus belongs to the visual areas; the activation of the sensory-related brain areas is a prerequisite to the generation of consciousness (8, 71). Yuan and Zhang et al. proved that the left cuneus exhibited a decreased activation in depressive patients (72, 73). However, task performance seems to be more in favor of the right cuneus in depression (74). The MTG is mainly responsible for facial recognition and explicit memory, which refers to a memory retrieval mode that requires the involvement of the consciousness (75). The abnormal activity of MTG may be responsible for negative emotion and cognitive impairment (76). The STG is a part of the facial-response region; the hyperactivation of the STG increased sensitivity to faces (77). Miller et al. (20) detected an overactivity centered at STG in depression during the affective processing tasks. Our findings showed that depression failed to inhibit the activation of DMN, which could be the reason for depressive tendency (78). With regards to deactivated brain areas, the abnormal network entailed salience network (SN, including IPL and insula) and executive control network (ECN, including SFG). SN takes part in the perception and integration of emotional information. The study showed that the resting-state FC alteration of SN decreased in treatment-resistant patients with depression (79). The insula is responsible for regulating attention, working memory, and some other higher cognitive functions and facilitating the processing of task-related information (80–82). In addition, the activation of the insula makes individuals prone to neuroticism, which increases sensitivity to negative stimuli (83). IPL is involved in socio-emotional processing and sensory integration (84). Lu et al. (85) found that the deactivation of the IPL was associated with anhedonia in depression. The SFG is responsible for self-consciousness and cognitive control (86). Zhao et al. (87) proved that the FC of SFG decreased in depressive patients during the resting state. Abnormal activity of the DMN, SN, and ECN together constitute a “triple network” model, which is involved in various psychiatric and neurological disorders including depression (88–90). During task performance, the ECN has a close relationship with SN; ECN usually has synergistic effects on SN (91). Meanwhile, SN can deactivate DMN (80). In the present study, we found triple-network abnormalities during emotion recognition tasks in depressive patients. Moreover, the activation of the DMN and deactivation of SN and ECN during the explicit tasks in depression are consistent with the findings of resting-state FC (67, 79, 87). In addition, when performing an explicit emotion recognition task, depressive patients are asked to classify emotional stimuli. During this process, subjects receive emotion-related information through multiple sensory input pathways, and high-level cognitive systems are engaged in processing the information and forming a perception. Haxby et al. (77) proposed three key face-selective regions when subjects perceived facial information: the occipital face area (OFA), the fusiform face area (FFA), and the superior temporal sulcus area (pSTS-FA) (92). The OFA and FFA are in charge of processing static facial information. The pSTS-FA plays a crucial role in processing dynamic features of eyes and mouth, and facilitating the identification of facial expressions (92–94). Our results showed the decreased neuro-activity of the pSTS-FA (insula and STG) in depressive patients (95), which were consistent with previous studies (20, 21). The aberrant activity of the pSTS-FA in depression indicated that depressive patients may perform poorly in processing dynamic emotional information.

To sum up, the MTG, STG, PHG, cuneus, IPL, insula, and SFG are the characteristic areas in processing emotional information consciously, and the abnormal activation of these areas could be related to poor emotional processing in patients with depression, which could be used as target brain regions for diagnosis or treatment of depression.

When identifying negative emotion stimuli, depression was associated with an increased activation on the precentral gyrus (BA 6), MFG (BA 6), precuneus (BA 7), and STG (BA 21). Among them, the precentral gyrus, MFG, and precuneus were attributed to DMN node. MFG is a part of the dorsolateral prefrontal cortex, which is involved in cognitive control and emotion processing (96). Zhang et al. (97) found that the MFG was responsible for aberrant negative emotion processing. The precentral gyrus is a crucial part of dorsal medial prefrontal cortex (dmPFC), which plays a critical role in the regulation of social cognition (98). Research discovered that precentral gyrus tended to process negative information, and increased activity of dmPFC was associated with impairment of processing negative emotion (17). The precuneus participates in a variety of cognitive functions including perception and memory cognition (68). Lin et al. (83) reported that the left STG was negatively correlated with neuroticism in the resting-state. Neuroticism is a personality and includes multiple dimensions. Lyon et al. (99) concluded that the dimensions of neuroticism such as anxiety, hostility and self-consciousness of neuroticism were positively related with depression. Ding et al. (100) proved that, compared to neutral stimuli, individuals exhibited a higher activation of the bilateral STG during sad stimuli. In summary, the aberrant activation of precentral gyrus, MFG, precuneus, and STG were responsible for the deficits in negative emotion processing.

As for positive emotion stimuli, the hyperactivity of the medial frontal gyrus (BA 10), MTG (BA 39), and insula (BA 13) were mainly responsible for the abnormal emotion processing, while no hypoactivation was observed. The DMN node (medial frontal gyrus), ECN node (MTG), and SN node (insula) were predominately involved. Watson and de Gelder (101) found an activation in insula, SFG, and medial frontal gyrus in explicit emotion tasks. In addition, the activation of the prefrontal cortices and insula may indicate a higher cognitive effort during identification of the positive emotion (102). Li and Wang (21) revealed a higher activation in the insula in the processing of emotion, which was consistent with our result. Medial frontal gyrus is the main region related to emotion perception (101, 103).

Compared with depressive adults, we discovered that the abnormal brain regions of depressive adolescents were more concentrated in the limbic system and striatum (caudate and lentiform nucleus). The limbic system and striatum are known as the emotional regulation system. Xia et al. (104) held that the limbic system was associated with emotion processes, and any abnormality in the limbic system could induce impairment of emotional control and render adolescents more impulsive. Tottenham et al. (105) pointed out that adverse experience could affect brain development in early life, especially the limbic structures such as the hippocampus were most sensitive to adverse experiences. In addition, accelerated or delayed limbic structure maturation might account for children's and adolescents' behavioral inadequacies which could be a risk factor for depression (106). Consistent with our results, Harlé et al. (107) found an increased activation of the striatum, particularly the caudate, in depressive adolescents. The striatum is one of the important structures in mood regulation and reward processing (108, 109). Reward is the stimuli that promotes organisms' efforts to obtain what is necessary for survival (110). The impairment in the reward system was inversely correlated with negative affection, irritability, and anhedonia in depression (107). The striatal response to reward tasks is the manifestation of depression (111, 112). Tseng et al. (113) found a positive correlation between irritability and striatal activation. The younger the age, the stronger the correlation between increased striatal activation and irritability. Anderson et al. (114) discovered that a higher functional connectivity of striatum may be responsible for anhedonia.

The decreased activity of the SN (IPL and postcentral gyrus) and ECN (SFG) were also detected in patients with severe depression. Studies showed that discrepant severity of depression manifested differential brain activation (115, 116). Lin et al. (116) constructed prediction models for depression (mild, moderate, and severe); they found that the fMRI signals of IPL, postcentral gyrus, and SFG were able to classify the severity of depression. Liu et al. (117) detected the decreased activation of the SFG, postcentral gyrus, and IPL in patients with severe depressive. In addition, we found that the amygdala was deactivated in severe depression. However, the activation of the amygdala in our study was inconsistent with the previous task-based meta-analyses, which concluded that the overactivity of amygdala had a bearing on the impaired emotional processing in depression (21, 118). The amygdala is a limbic structure located in the medial temporal lobe, which has the ability to identify emotion (119). Moreover, the amygdala is the “behavioral brake” that protects organisms from potential harm (120). Ernst et al. (121) pointed that a weak harm-avoidant system (amygdala) could contribute to depression. Meyer et al. (122) discovered that amygdala activation was negatively associated with depressive symptoms. Lu et al. (123) recorded the impaired connectivity of amygdala during facial emotional stimulus in depressive patients. Since studies exploring abnormal brain regions in different severities of depression are rarely reported, future studies are needed to address this issue.

Our results showed abnormal brain regions of depression included the DMN, SN, and ECN. The coupling of between/within “triple network” during explicit tasks in depression needs further investigation. The correlation of the abnormal brain areas and depressive symptoms was unknown. Fisher's Z-values could be extracted for correlation analysis, or meta-regression could be performed. In the present study, we focused on the brain activity during explicit emotion tasks; further research could investigate the differential activity changes of brain regions between explicit emotion tasks and implicit emotion tasks in depression. In the future, researchers could utilize a multimodal combination with electroencephalogram, functional near-infrared spectroscopy, or single photon emission computed tomography to thoroughly investigate the neurophysiological mechanisms of depression.

There were several strengths of this study. First, this is the first neuroimaging meta-analysis that explored abnormal brain activation during the performance of explicit tasks in depression. Second, we used ALE to synthesize the coordinates from included studies, which increase the credibility of the results to some extent. Third, sensitivity analyses were conducted to evaluate the robustness of the findings.

However, some limitations existed in the present study, so the results should be interpreted with caution. First, we used uncorrected p-value for the spatial consistency of brain activation; the uncorrected threshold may cause the possibility of false positives. Second, there was no restriction on the handedness of included patients, which could be an influencing factor to the results. Third, due to limited included studies, we failed to investigate the brain activity of different severities of depression, which needs further researches.