Volume 2025, Issue 1 7645625

Research Article

Open Access

Identification of Subthreshold Depression Based on fNIRS–VFT Functional Connectivity: A Machine Learning Approach

Lin Li

orcid.org/0000-0002-8002-6563

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Jingxuan Liu,

Jingxuan Liu

orcid.org/0009-0007-2489-3594

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Yifan Zheng,

Yifan Zheng

orcid.org/0009-0003-0682-7068

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Chengchao Shi,

Chengchao Shi

orcid.org/0009-0007-6945-4237

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Wenting Bai,

Corresponding Author

Wenting Bai

[email protected]

orcid.org/0009-0002-9938-8967

School of Physical Education and Health , Shanghai University of International Business and Economics , Shanghai , 201620 , China , suibe.edu.cn

Search for more papers by this author

Lin Li,

Lin Li

orcid.org/0000-0002-8002-6563

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Jingxuan Liu,

Jingxuan Liu

orcid.org/0009-0007-2489-3594

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Yifan Zheng,

Yifan Zheng

orcid.org/0009-0003-0682-7068

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Chengchao Shi,

Chengchao Shi

orcid.org/0009-0007-6945-4237

Key Laboratory of Adolescent Health Assessment and Exercise Intervention of Ministry of Education , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

College of Physical Education and Health , East China Normal University , Shanghai , 200241 , China , ecnu.edu.cn

Search for more papers by this author

Wenting Bai,

Corresponding Author

Wenting Bai

[email protected]

orcid.org/0009-0002-9938-8967

School of Physical Education and Health , Shanghai University of International Business and Economics , Shanghai , 201620 , China , suibe.edu.cn

Search for more papers by this author

First published: 31 January 2025

https://doi.org/10.1155/da/7645625

Academic Editor: Drozdstoy Stoyanov

Share a link

Email
Wechat
Bluesky

Abstract

Background: Subthreshold depression (SD) is regarded as a prodromal stage and a substantial risk factor for major depressive disorder (MDD). The timely identification of SD is of critical clinical significance. This study aimed to develop a machine learning (ML) classification model for the identification of individuals with SD using functional near-infrared spectroscopic imaging (fNIRS) and the verbal fluency task (VFT).

Methods: This study recruited a total of 70 participants with SD and matched 73 healthy controls (HCs) to differentiate between the two groups based on functional connectivity (FC) features during fNIRS–VFT, using an interpretable random forest (RF) classification model.

Results: The RF model demonstrated an area under the curve (AUC) of 0.77, an accuracy (ACC) of 75.86%, a sensitivity of 75.00%, a specificity of 76.00% and an F1 score of 0.75 for identifying participants with SD. The highest-ranked FC features, in terms of importance, were identified between Channel (CH) 26 (the right frontal eye fields (FEFs)) and CH 30 (the right FEF), CH 3 (the left premotor and supplementary motor cortex (PMC-and-SMA)) and CH 42 (the right PMC-and-SMA), as well as CH 26 (the right FEF) and CH 32 (the right primary somatosensory cortex (PSC)).

Conclusion: The RF model has the capacity to effectively classify individuals with SD efficacy based on the abnormal FC features of fNIRS–VFT, particularly in the right FEF, bilateral PSC and right PMC-and-SMA. The findings of this study have provided a foundation for large-scale screening of SD populations, offering promising opportunities for the early diagnosis and prevention of MDD.

1. Introduction

The prevalence of major depressive disorder (MDD) continues to increase, representing one of the most significant challenges to public health worldwide [1–3]. Subthreshold depression (SD) is defined as a subclinical depressive state where individuals experience some depressive symptoms for at least 2 weeks, but do not meet the clinical diagnostic criteria for MDD, regarded as the prodromal stage and a significant risk indicator for MDD [4]. A prevalence rate of 32% for SD was reported among university students in China [5]. Moreover, two large-sample studies (n = 30,034) found that the closer the level of depression was to a threshold, the higher the associated risk of dysfunction and suicide [6, 7]. Therefore, it is imperative to enhance the identification, diagnosis and monitoring of SD in the college student population. This is not only a crucial step in kerbing the undesirable developmental trend of MDD but also a highly urgent practical necessity.

Currently, the principal methodology for diagnosing and identifying SD is based on clinical investigations and interviews ([8], p. 202). However, this subjective method is influenced by the heterogeneity of individuals’ functional and behavioural performance, which presents challenges for the assessment and continuous monitoring of large populations [9]. Consequently, there is a pressing need to develop an objective, cost-effective and efficient technological tool that can facilitate large-scale identification and monitoring of SD. From the perspective of neuroimaging, functional near-infrared spectroscopic imaging (fNIRS) technology has the potential to offer considerable advances. As a promising tool, fNIRS has the advantage of providing continuous, real-time monitoring of the cerebral haemodynamic response and changes in blood oxygen levels, thus, enabling a deeper understanding of functional changes in the brain. Previous studies have demonstrated the considerable value of fNIRS in the early diagnosis, assessment and monitoring of depression [10]. The verbal fluency task (VFT) is an operationally simple cognitive task that, in conjunction with fNIRS, has been widely used to probe the activity of the cerebral cortex in depressed individuals, thereby, reflecting their cognitive neural function [9]. The evidence from fNIRS–VFT studies has indicated that the VFT is able to activate the brain regions of healthy control (HC), such as frontal, parietal and temporal cortex [10], which is more broadly than the classical N-back and Stroop tasks that are typically used to assess executive functioning [11, 12] and that it is involved in a variety of complex human cognitive processes such as executive function and working memory [13]. Previous studies have revealed that individuals with SD exhibited significant deactivation in specific brain regions, including the bilateral dorsolateral prefrontal cortex (DLPFC), the bilateral dorsal frontal pole cortex and the bilateral superior temporal cortex [14]. Hence, it can be concluded that fNIRS–VFT may be more suitable for brain function studies in individuals with SD, especially when the study involves cross-brain region and brain network level.

The advent of sophisticated data analysis tools has prompted an increasing number of researchers to propose that the onset and progression of depression may be more likely to manifest as abnormal connectivity within brain networks [15,–18]. As an effective indicator of the correlation of neural activities between different brain regions, functional connectivity (FC) is of great significance for the in-depth understanding of the brain’s information processing and the synergistic work between different brain regions [19]. In fact, abnormal FC has been recognised as a reliable neurobiological marker for the clinical diagnosis of severe psychiatric disorders such as MDD and schizophrenia [20]. Nevertheless, only two studies have employed fNIRS–VFT to investigate FC features in the brains of individuals with SD [8, 21]. These studies have found that FC strength is diminished in specific brain regions, such as between the orbitofrontal and the dorsolateral prefrontal region, as well as the dorsolateral prefrontal and the globus region.

An increasing number of studies have demonstrated the existence of pathophysiological mechanisms in SD that are similar to those of MDD [22–24], especially the abnormal FC [10, 25–27]. These studies subsequently pointed to an association between the abnormal FC in the frontotemporal lobe and depressive symptoms in SD [28]. However, these studies are based on conventional statistical analysis methods, such as tests for differences between groups and correlation analysis. Although these methods have ensured a certain degree of reproducibility of results, the resulting models frequently reflect only characteristics of the current sample, which may limit the generalisability of the findings [29, 30].

As a novel statistical technique, machine learning (ML) has developed rapidly in recent years and has been recognised as a valuable tool for clinical diagnosis [31]. ML is capable of learning from data and automatically iterating to predict and identify unknown data, thereby, enabling the performance of tasks such as classification and regression [32]. The increasing integration of ML and neuroimaging techniques has enabled the development of individual-level classification and prediction models based on the functional level of the brain [33–35]. This is crucial for the clinical diagnosis of MDD, schizophrenia and other psychiatric disorders. For example, Chen et al. [28] combined the random forests (RFs) algorithm with resting-state FC features and successfully achieved an accuracy (ACC) of 73.1% ± 2.8% in a four-way classification task that identifying the HC group, nonpsychotic MDD, psychotic MDD, and schizophrenia. In a review of relevant studies from the last decade, Eken, Nassehi, and Erogul [36] demonstrated that combining ML with fNIRS may prove more beneficial than functional magnetic resonance imaging (fMRI) in identifying potential biomarkers in specific cortical regions associated with disorders. Nevertheless, to the best of our knowledge, no study has yet combined ML with fNIRS signals to explore the potential of FC features of the cortex in identifying individuals with SD when completing cognitive tasks.

In summary, the objective of this study was to construct a classification model using the ML method to identify individuals with SD based on fNIRS–VFT. Specifically, we aimed to explore the distinctive FC during the VFT, which could effectively distinguish individuals with SD, thus, facilitating future large-scale screening and diagnosis of SD.

2. Participants and Methods

2.1. Participants

The study was approved by the Human Subjects Protection Committee of East China Normal University (HR-427-2022). Participants were recruited from East China Normal University in Shanghai, China, between August 2022 and June 2023. After receiving a complete study description, all participants provided written informed consent.

We recruited 77 participants with SD (7 males and 70 females) for the study. Due to the significant gender imbalance, only female participants were included in the SD group to eliminate potential confounders factors and ensure the reliability of the results. Seventy-three HC participants were selected and matched accordingly. Exclusion criteria for participants were as follows: (1) left-handedness; (2) colour vision deficiency or impaired colour perception; (3) history of depression or other mental illness; (4) history of traumatic brain injury or other organic brain disease; (5) substance-related or addiction disorders; (6) history of major physical illness; (7) pregnancy or lactation; (8) non-native Chinese speakers; (9) inability to complete the VFT.

2.2. Depression Scale

Following the previous study [37], we defined ‘SD’ as the presence of two to five depressive symptoms for at least 2 weeks. Depressive symptoms were assessed using the Beck Depression Inventory-II (BDI-II) [38]. The BDI-II consists of 21 items, each rated on a four-point Likert scale ranging from 0 (asymptomatic) to 3 (highly symptomatic). It has demonstrated better internal consistency in Chinese college students (Cronbach’s α = 0.85) [39] and exhibits high stability across gender, race and ethnicity [40].

In this study, all participants were assessed for depressive symptoms using the BDI-II at two time points, with an interval of 14 days between assessments. Those with BDI-II scores of 14–28 with both assessments were identified as SD, while those with scores below 14 were included in the HC group.

2.3. Activation Task

We conducted a Chinese version of the VFT similar to previous studies [41, 42]. The VFT consisted of three distinct parts: a 30-s prerest period, a 60-s task period, and a 60-s postrest period. During the pre- and postrest periods, participants were required to recite aloud the numerical sequence ‘1, 2, 3, 4 and 5’ on each occasion. During the 60-s task period, participants were asked to construct as many words as possible, starting with four common Chinese characters, such as ‘上’, ‘大’, ‘天’ and ‘家’. The given Chinese characters were changed every 15 s. The number of unique words uttered during the task period was used to measure task performance. To control for the independent variable, all participants were presented with the same character cue and in the same order (Figure 1).

Details are in the caption following the image — **Figure 1**
Open in figure viewer PowerPoint

Schematic diagram of the verbal fluency task (VFT).

2.4. fNIRS Data

2.4.1. Data Acquisition

During the VFT, a 44-channel (CH) fNIRS system (ETG-7100, Hitachi Medical Corporation, Tokyo, Japan) was used to continuously monitor and record the fNIRS signals of participants. The system had wavelengths of 695 and 830 nm, a sampling rate of 10 Hz and employed two 3 × 5 probe boards with a spacing of 3 cm between the probes to cover the left and right frontal, temporal and parietal regions, respectively. The region between each pair of emitters and detectors is called a CH, and each probe consists of 22 CHs of eight emitters and seven detectors, for a total of 44 CHs of fNIRS signals monitored.

To ensure that the spatial positions of the probes were accurately recorded, we used a three-dimensional localiser to calibrate four key reference points (Nz, Cz, AL and RL). All CHs were mapped to the Montreal Neurological Institute (MNI) space [43] by virtual alignment [44] and using the NIRS–SPM [45] and projected onto the Brodmann cortical partitioning template [46]. We used the maximum overlap probability method to obtain a more accurate analysis of cortical activity in the monitored brain regions. These CHs were classified into the following regions: the supramarginal gyrus (SMG) part of Wernicke’s area, the primary somatosensory cortex (PSC), the premotor and supplementary motor cortex (PMC-and-SMA), the DLPFC and the frontal eye fields (FEFs). More details regarding the precise locations of the CHs can be found in Figure 2.

2.4.2. Data Preprocessing

The raw fNIRS data were analysed using the HOMER 2 toolbox [47]. The steps include converting the raw light intensity data into optical density values, detecting motion artefacts in each CH [48] and correcting them using spline interpolation [49]. The data were then filtered in the range of 0.05–0.1 Hz using a bandpass filter [50]. The modified Beer–Lambert law was then used to convert the filtered optical data into oxyhaemoglobin (HbO), deoxyhaemoglobin (HbR) and total haemoglobin (HbT) concentrations [51]. As changes in HbO concentration have been shown to reflect better individual cortical activation related to cognitive tasks [52], we focused only on changes in HbO concentration.

2.5. Feature Extraction and Simplification

Figure 3 shows the flowchart of SD recognition based on fNIRS spatial features. After preprocessing, the HbO concentration change curves of all individuals in each CH at 60 s during the task execution phase were extracted. Then, a 44 × 44 FC matrix of all SD and HC participants was constructed using the Pearson correlation coefficient. Finally, a classifier was trained using the RFs algorithm based on the FC features of all participants.

Referring to the previous study [53], this study was conducted to maximise the data-driven nature of the study with the completeness of the FC data for a more accurate identification of the SD. At the same time, the advantages of the RF algorithm itself were taken into account, such as the excellent and flexible ability to cope with the noise data as well as the ability to identify the features of the high-dimensional data. Therefore, this study did not apply additional simplification methods, such as feature filtering or dimensionality reduction, for FC features. More detailed information about the RF algorithm can be found in Section 2.6.

2.6. ML Model

2.6.1. Model Selection

In this study, the RF algorithm is implemented using the Scikit-learn library in Python.

RF is an advanced ensembles algorithm with a flexible hyperparameter tuning strategy [54], which is able to effectively optimise the impact of features on the modelling process, thus, improving the performance and stability of the model. The algorithm is capable of generating and processing multiple decision trees in parallel and automatically performs feature selection based on the amount of information in the features in each decision tree [55]. This built-in feature selection mechanism helps the RF algorithm to focus on the most informative features when dealing with high-dimensional feature data such as FC, in order to reduce the interference of redundant features in the model [54]. In addition, considering that the RF algorithm possesses advantages over other methods such as strong robustness against noise, and the ability to capture complex nonlinear relationships among data based on the decision tree structure [56], it can effectively deal with fNIRS anomalous FC data generated due to noise or interference (e.g., motion artefacts, signal loss, etc.) during the experiment.

Specifically, this study divides the dataset into a training set (80%) and a test set (20%). The RF algorithm generates numerous decision trees based on the training set and constructs a predictive model through cross-validation and hyperparameter grid search. The final predictive model is then evaluated using performance metrics such as ACC on the test set.

2.6.2. Modelling Process

During the modelling process, this study employed sixfold cross-validation combined with grid search to mitigate overfitting. The optimal hyperparameter configuration was determined within predefined ranges, resulting in the following settings: ‘max_depth’: none, ‘min_samples_leaf’: 2, ‘min_samples_split’: 2 and ‘n_estimators’: 100. These hyperparameters are defined and explained bellow:

Setting ‘max_depth’ to none allows the decision tree to expand fully until all leaves are pure or the number of samples in a leaf falls below the value specified by ‘min_samples_split’. While this approach enables the model to capture complex patterns in the data, it also increases the risk of overfitting.

‘min_samples_leaf’ specifies the minimum number of samples required for a leaf node. In this study, the value was set to 2, ensuring that each leaf node contains at least two samples. This constraint helps to reduce the likelihood of overfitting by preventing overly specific splits.

‘min_samples_split’ defines the minimum number of samples required to split an internal node. A value of 2 was chosen, meaning that each internal node must contain at least two samples to be considered for splitting. Lowering this value may improve the model’s ability to capture finer details in the training data, although it may increase the risk of overfitting.

‘n_estimators’ indicates the number of decision trees to be constructed in the forest. A value of 100 was used, meaning that the model builds 100 decision trees. The final classification predictions are made using soft voting, aggregating the output of all trees.

2.6.3. Model Evaluation

The most commonly used and accepted metrics to evaluate the performance of the RF classifier model are ACC, sensitivity (true positive rate (TPR)), specificity (true negative rate (TNR)), F1 score and area under the curve (AUC). The following definitions apply to the above metrics:

The most fundamental metric for evaluating the performance of a classification model is the ACC, which represents the ratio of correctly predicted samples to the total number of samples, calculated from true positive (TP), true negative (TN), false positive (FP) and false negative (FN) values. In this study, the term ‘ACC’ is used to denote the proportion of correct identifications made by the classification model for all participants with and without SD.

The TPR represents the ratio of the number of samples from SD participants predicted by the classification model to the number of samples from the test set that suffered from SD. A higher TPR indicates a higher correct detection rate of the model for SD participants.

The TNR represents the ratio of the number of samples from HC participants predicted by the classification model to the number of healthy samples in the test set. A higher TNR indicates a higher correct detection rate for HC participants in the model.

The F1 score represents a balance between precision and recall. It is used to assess the performance of a modal in balancing these two factors.

The AUC represents the area under the receiver operating characteristic (ROC) curve and reflects the TPR and FP rate (FPR) of the model. A value of 1 indicates absolute ACC. In addition, the TP, TN, FP and FN metrics in the confusion matrix can be used to evaluate the performance of the model.

To explore the mechanism of the RF in predicting participants with SD, this study used the mean decrease in impurity (MDI) method to evaluate the importance of each feature in the model’s classification ACC [57]. The MDI method quantifies the importance of each feature by calculating the average decrease in impurity (such as Gini impurity or entropy) achieved when a particular feature is used for splitting during tree-building. Features with higher MDI values contribute more to the model’s predictive power.

3. Result

3.1. Demographics, BDI-II Scores and VFT Behavioural Performance

The statistical analysis was performed using the SPSS version 23 (IBM, Armonk, NY, USA).

The results of the independent samples t-test indicated that there are no statistically significant differences between the SD and HC groups with regard to age, education level and VFT behavioural performance. The BDI-II scores of the SD group were significantly higher than the HC group (Table 1).

Table 1. Two groups’ demographics, BDI-II scores and VFT behavioural performance (mean ± standard deviation).

Variant	HC group (n = 73)	SD group (n = 70)	t	p
Age (years)	20.55 ± 2.34	20.30 ± 2.00	0.683	0.496
Education level (years)	14.55 ± 2.04	14.46 ± 1.79	0.261	0.795
BDI-II scores	4.62 ± 3.89	21.10 ± 4.13	−24.593	<0.001 ^∗∗∗
VFT performance^a	16.23 ± 4.18	17.54 ± 4.19	−1.872	0.063

Abbreviations: BDI-II, Beck Depression Inventory-II; HC, healthy control; SD, subthreshold depression; VFT, verbal fluency task.
^∗∗∗p < 0.001.
^aTotal number of correct and unrepeated words during VFT.

3.2. Functional Connectivity Analysis

An average FC chord plot was constructed for the SD and HC groups to provide a visual representation of the differences between groups regarding FC features (Figure 4). The SD group exhibited a notable reduction in FC strength compared to the HC group.

3.3. Classification Results

The FC of each CH was analysed between the two groups, revealing the FC features derived from the fNIRS–VFT. The findings provided a valuable foundation for predicting the presence of SD. Table 2 presents the performance metrics of the RF model for identifying participants with SD, achieving an ACC of 75.86%. The result indicated that the model correctly identified individuals with SD in 75.86% of cases. Additionally, the classification model demonstrated a TNR of 76%, indicating that it correctly classified 76% of HC individuals. The TPR of the model was 75%, reflecting its ability to accurately identify 75% of participants with SD within the actual SD group. Figure 5 shows the confusion matrix and the ROC curve.

Table 2. Performance of RF model in identifying participants with SD.

Model	AUC	ACC (%)	TPR (%)	TNR (%)	F1 score
RF	0.77	75.86	75	76	0.75

Note: F1 score represents a balance between precision and recall.
Abbreviations: ACC, accuracy; AUC, area under the curve; RF, random forest; SD, subthreshold depression; TNR, true negative rate; TPR, true positive rate.

3.4. Feature Importance Ranking

Figure 6 shows the ranking of FC feature importance as determined by the RF model, illustrating the top 10 CH pairs that contribute to the model’s performance. The most significant FC features were CH26 (the right FEF) and CH30 (the right FEF). The next highest FC features were CH3 (the left PMC-and-SMA) and CH42 (the right PMC-and-SMA). The third highest FC features were CH26 (the right FEF) and CH32 (the right PSC).

4. Discussion

In this study, a novel RF model was constructed using FC features based on fNIRS–VFT, and its ACC and robustness in identifying individuals with SD were verified. The results of the RF model demonstrated that the model achieved an AUC of 0.77, an ACC of 75.86%, a sensitivity of 75%, a specificity of 76% and an F1 score of 0.75. A further analysis of the feature’s importance revealed that specific FC features made a significant contribution to the model’s predictive performance, including CH26 (the right FEF) and CH30 (the right FEF), CH3 (the left PMC-and-SMA) and CH42 (the right PMC-and-SMA) and including CH26 (the right FEF) and CH32 (the right PSC).

The performance of the RF model established in this study for diagnosing psychiatric disorders, particularly in hyperparameter optimisation, was comparable to that reported in previous studies combining fNIRS–VFT and ML techniques. Li et al. [58] achieved comparable classification performance for MDD using an SVM model. The approach employed features such as the integral value of changes in HbO concentration during fNIRS–VFT and the centre of mass value. It is noteworthy that several studies have reported superior performance in the MDD classifications compared to our study. A previous study distinguishing MDD from HC demonstrated that an SVM classifier achieved a mean AUC of 0.82 (95% CI: 0.75–0.90), a mean ACC of 82.02% ± 10.04%, a sensitivity of 70.00% ± 18.6% and a specificity of 94% ± 7.32% [59]. Indeed, accurately identifying SD through the construction of RF models presents a challenge due to the clinical characteristics of SD and MDD [60], as well as the fact that the structure and function of the brain exhibit continuity [61, 62]. In summary, our study findings indicated that the RF algorithm model based on the FC features of fNIRS–VFT has the potential to serve as a clinical diagnostic tool for SD and to facilitate the early screening of MDD.

Based on the feature importance rankings calculated from the MDI values, this study identified that the top three FC features were primarily concentrated in the PMC-and-SMA, FEF and PSC brain regions. Previous studies have demonstrated that patients with MDD exhibited decreased resting-state FC of the FEF with the middle occipital gyrus compared to HC [63]. Additionally, several studies have reported functional abnormalities in the PMC-and-SMA during the completion of various cognitive tasks [64–66]. These results are consistent with our findings of frontal lobe abnormalities in the FEF and the SMA in SD, indicating that the frontal lobes are essential in the pathogenesis of depression. Moreover, both the PMC-and-SMA and FEF are fundamental to cognitive processes. The PMC-and-SMA play an essential role in executive functions and in integrating emotional, behavioural and cognitive functions [67]. The FEF is closely associated with higher-order cognitive functions and behaviours, including language, self-awareness and emotion [68–72]. These observations indicated that abnormal FC in the FEF and PMC-and-SMA regions in individuals with SD may serve as a crucial pathological indicator of cognitive impairment [21].

A recent fNIRS study has observed significantly reduced activation in the prefrontal cortex and inferior parietal regions in individuals with MDD compared to HC during VFT performance [10]. The FEF is located at the prefrontal–parietal junction [73] and is a critical component of the frontoparietal network. This network is associated with executive functions, cognitive flexibility and attentional modulation [74, 75]. It is hypothesised that the FEF mediates FC within the frontoparietal network, regulating the activities of the PSC in a top-down manner, particularly in contexts involving language processing [76]. In particular, the VFT process, which pertains to language processing, revealed that patients with depression exhibited impairments in language processing [77]. The FEF appeared to guide language-related selective attentional information, including visual and auditory inputs, before transmitting this information to the PSC, which facilitates the fluency of language production during tasks like the VFT [76, 78]. Impairments in this mechanism may provide an explanation for the language processing deficits observed in individuals with depression during the VFT. We hypothesised that individuals with SD, similar to those with MDD, may demonstrate aberrant FC between the FEF and PSC, which may contribute to impaired language processing abilities. This result not only validates the abnormal FC with continuity in the development of HC–SD–MDD, but also further emphasises the importance of timely identification and diagnosis of SD for the early diagnosis and prevention of MDD at the cognitive neuroscience level.

4.1. Limitations

There were several limitations in the present study. First, all participants were female, which may limit the generalizability of the findings. Given the influence of gender on brain structure and function, future studies should aim to include a balanced representation of male and female participants and build a small-sample external validation test set to enhance the applicability of the results. Second, the considerable volume and high dimensionality of the FC data presented significant analytical challenges. Future research could incorporate a feature screening step to simplify the complex and large FC data or apply deep learning algorithms to capture deeper, potentially more informative features. These approaches may further improve the model’s recognition performance.

5. Conclusion

This study is the first to focus on fNIRS–VFT FC features combined with ML methods to develop an effective model. The fNIRS–VFT FC features can accurately identify individuals with SD, especially in the right FEF, bilateral PSC and right PMC-and-SMA. The findings of this study have provided a foundation for large-scale screening of SD populations, offering promising opportunities for the early diagnosis and prevention of MDD.

Conflicts of Interest

The authors declare no conflicts of interest.

Funding

This work was supported by the National Social Science Fund of China (21BTY094).

Open Research

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

References

1 Moitra M., Santomauro D., and Collins P. Y., et al.The Global Gap in Treatment Coverage for Major Depressive Disorder in 84 Countries From 2000-2019: A Systematic Review and Bayesian Meta-Regression Analysis, PLOS Medicine. (2022) 19, no. 2, https://doi.org/10.1371/journal.pmed.1003901, e1003901.
10.1371/journal.pmed.1003901
PubMed Web of Science® Google Scholar
2 Perahia D. G. S., Quail D., Desaiah D., Montejo A. L., and Schatzberg A. F., Switching to Duloxetine in Selective Serotonin Reuptake Inhibitor Non- and Partial-Responders: Effects on Painful Physical Symptoms of Depression, Journal of Psychiatric Research. (2009) 43, no. 5, 512–518, https://doi.org/10.1016/j.jpsychires.2008.07.001, 2-s2.0-59849094689.
10.1016/j.jpsychires.2008.07.001
PubMed Web of Science® Google Scholar
3 Willner P., The Neurobiology of Depression and Antidepressant Action, Neuroscience & Biobehavioral Reviews. (2013) 37, no. 10, 2331–2371, https://doi.org/10.1016/j.neubiorev.2012.12.007, 2-s2.0-84890224196.
10.1016/j.neubiorev.2012.12.007
CAS PubMed Web of Science® Google Scholar
4 Cuijpers P. and Smit F., Subthreshold Depression as a Risk Indicator for Major Depressive Disorder: A Systematic Review of Prospective Studies, Acta Psychiatrica Scandinavica. (2004) 109, no. 5, 325–331, https://doi.org/10.1111/j.1600-0447.2004.00301.x, 2-s2.0-1842855342.
10.1111/j.1600-0447.2004.00301.x
CAS PubMed Web of Science® Google Scholar
5 Jiang L., Wang Y., and Zhang Y., et al.The Reliability and Validity of the Center for Epidemiologic Studies Depression Scale (CES-D) for Chinese University Students, Frontiers in Psychiatry. (2019) 10, https://doi.org/10.3389/fpsyt.2019.00315, 2-s2.0-85068154479.
10.3389/fpsyt.2019.00315
Web of Science® Google Scholar
6 An J. H., Subthreshold Lifetime Depression and Anxiety Are Associated With Increased Lifetime Suicide Attempts: A Korean Nationwide Study, Journal of Affective Disorders. (2022) 302, 170–176, https://doi.org/10.1016/j.jad.2022.01.046.
10.1016/j.jad.2022.01.046
PubMed Web of Science® Google Scholar
7 Balázs J., Miklósi M., and Keresztény Á., et al.Adolescent Subthreshold-Depression and Anxiety: Psychopathology, Functional Impairment and Increased Suicide Risk, Journal of Child Psychology and Psychiatry. (2013) 54, no. 6, 670–677, https://doi.org/10.1111/jcpp.12016, 2-s2.0-84878105828.
10.1111/jcpp.12016
PubMed Web of Science® Google Scholar
8 Da H., Xiang N., Qiu M., Abbas S., Xiao Q., and Zhang Y., Characteristics of Oxyhemoglobin During the Verbal Fluency Task in Subthreshold Depression: A Multi-Channel Near-Infrared Spectroscopy Study, Journal of Affective Disorders. (2024) 356, 88–96, https://doi.org/10.1016/j.jad.2024.04.005.
10.1016/j.jad.2024.04.005
CAS PubMed Google Scholar
9 Ho C. S. H., Wang J., and Tay G. W. N., et al.Interpretable Deep Learning Model for Major Depressive Disorder Assessment Based on Functional Near-Infrared Spectroscopy, Asian Journal of Psychiatry. (2024) 92, https://doi.org/10.1016/j.ajp.2023.103901, 103901.
10.1016/j.ajp.2023.103901
PubMed Google Scholar
10 Husain S. F., Tang T.-B., and Yu R., et al.Cortical Haemodynamic Response Measured by Functional Near Infrared Spectroscopy During a Verbal Fluency Task in Patients With Major Depression and Borderline Personality Disorder, EBioMedicine. (2020) 51, https://doi.org/10.1016/j.ebiom.2019.11.047, 102586.
10.1016/j.ebiom.2019.11.047
PubMed Web of Science® Google Scholar
11 Matsubara T., Matsuo K., and Nakashima M., et al.Prefrontal Activation in Response to Emotional Words in Patients With Bipolar Disorder and Major Depressive Disorder, NeuroImage. (2014) 85, 489–497, https://doi.org/10.1016/j.neuroimage.2013.04.098, 2-s2.0-84889646243.
10.1016/j.neuroimage.2013.04.098
PubMed Web of Science® Google Scholar
12 Yeung M. K., Lee T. L., and Chan A. S., Negative Mood Is Associated With Decreased Prefrontal Cortex Functioning During Working Memory in Young Adults, Psychophysiology. (2021) 58, no. 6, https://doi.org/10.1111/psyp.13802, e13802.
10.1111/psyp.13802
PubMed Web of Science® Google Scholar
13 Takizawa R., Kasai K., and Kawakubo Y., et al.Reduced Frontopolar Activation During Verbal Fluency Task in Schizophrenia: A Multi-Channel Near-Infrared Spectroscopy Study, Schizophrenia Research. (2008) 99, no. 1–3, 250–262, https://doi.org/10.1016/j.schres.2007.10.025, 2-s2.0-39149117438.
10.1016/j.schres.2007.10.025
PubMed Web of Science® Google Scholar
14 Yeung M. K. and Lin J., Probing Depression, Schizophrenia, and Other Psychiatric Disorders Using fNIRS and the Verbal Fluency Test: A Systematic Review and Meta-Analysis, Journal of Psychiatric Research. (2021) 140, 416–435, https://doi.org/10.1016/j.jpsychires.2021.06.015.
10.1016/j.jpsychires.2021.06.015
PubMed Web of Science® Google Scholar
15 Lui S., Wu Q., and Qiu L., et al.Resting-State Functional Connectivity in Treatment-Resistant Depression, American Journal of Psychiatry. (2011) 168, no. 6, 642–648, https://doi.org/10.1176/appi.ajp.2010.10101419, 2-s2.0-79960549391.
10.1176/appi.ajp.2010.10101419
PubMed Web of Science® Google Scholar
16 Rolls E. T., Cheng W., and Du J., et al.Functional Connectivity of the Right Inferior Frontal Gyrus and Orbitofrontal Cortex in Depression, Social Cognitive and Affective Neuroscience. (2020) 15, no. 1, 75–86, https://doi.org/10.1093/scan/nsaa014.
10.1093/scan/nsaa014
PubMed Web of Science® Google Scholar
17 Zhang A., Yang C., and Li G., et al.Functional Connectivity of the Prefrontal Cortex and Amygdala Is Related to Depression Status in Major Depressive Disorder, Journal of Affective Disorders. (2020) 274, 897–902, https://doi.org/10.1016/j.jad.2020.05.053.
10.1016/j.jad.2020.05.053
CAS PubMed Web of Science® Google Scholar
18 Zhu Z., Wang S., Lee T. M. C., and Zhang R., Habenula Functional Connectivity Variability Increases With Disease Severity in Individuals With Major Depression, Journal of Affective Disorders. (2023) 333, 216–224, https://doi.org/10.1016/j.jad.2023.04.082.
10.1016/j.jad.2023.04.082
PubMed Web of Science® Google Scholar
19 Mennes M., Kelly C., and Zuo X.-N., et al.Inter-Individual Differences in Resting-State Functional Connectivity Predict Task-Induced BOLD Activity, NeuroImage. (2010) 50, no. 4, 1690–1701, https://doi.org/10.1016/j.neuroimage.2010.01.002, 2-s2.0-77950595917.
10.1016/j.neuroimage.2010.01.002
PubMed Web of Science® Google Scholar
20 Baker J. T., Dillon D. G., and Patrick L. M., et al.Functional Connectomics of Affective and Psychotic Pathology, Proceedings of the National Academy of Sciences of the United States of America. (2019) 116, no. 18, 9050–9059, https://doi.org/10.1073/pnas.1820780116, 2-s2.0-85065510631.
10.1073/pnas.1820780116
CAS PubMed Web of Science® Google Scholar
21 Liu X., Cheng F., and Hu S., et al.Cortical Activation and Functional Connectivity during the Verbal Fluency Task for Adolescent-Onset Depression: A Multi-Channel NIRS Study, Journal of Psychiatric Research. (2022) 147, 254–261, https://doi.org/10.1016/j.jpsychires.2022.01.040.
10.1016/j.jpsychires.2022.01.040
PubMed Web of Science® Google Scholar
22 Li H., Wei D., Sun J., Chen Q., Zhang Q., and Qiu J., Brain Structural Alterations Associated With Young Women With Subthreshold Depression, Scientific Reports. (2015) 5, no. 1, https://doi.org/10.1038/srep09707, 2-s2.0-84929603536, 9707.
10.1038/srep09707
CAS PubMed Web of Science® Google Scholar
23 Li J., Wang Z., and Hwang J., et al.Anatomical Brain Difference of Subthreshold Depression in Young and Middle-Aged Individuals, NeuroImage: Clinical. (2017) 14, 546–551, https://doi.org/10.1016/j.nicl.2017.02.022, 2-s2.0-85014715447.
10.1016/j.nicl.2017.02.022
PubMed Web of Science® Google Scholar
24 Ma H., Zhang D., Sun D., Wang H., and Yang J., Gray and White Matter Structural Examination for Diagnosis of Major Depressive Disorder and Subthreshold Depression in Adolescents and Young Adults: A Preliminary Radiomics Analysis, BMC Medical Imaging. (2022) 22, no. 1, https://doi.org/10.1186/s12880-022-00892-5.
10.1186/s12880-022-00892-5
Google Scholar
25 Liao D., Liang L.-S., and Wang D., et al.Altered Static and Dynamic Functional Network Connectivity in Individuals With Subthreshold Depression: A Large-Scale Resting-State fMRI Study, European Archives of Psychiatry and Clinical Neuroscience. (2024) https://doi.org/10.1007/s00406-024-01871-3.
10.1007/s00406-024-01871-3
Google Scholar
26 Xu M., Li X., and Teng T., et al.Reconfiguration of Structural and Functional Connectivity Coupling in Patient Subgroups With Adolescent Depression, JAMA Network Open. (2024) 7, no. 3, https://doi.org/10.1001/jamanetworkopen.2024.1933, e241933.
10.1001/jamanetworkopen.2024.1933
PubMed Web of Science® Google Scholar
27 Zhang T., Zhao B., and Shi C., et al.Subthreshold Depression May Exist on a Spectrum With Major Depressive Disorder: Evidence From Gray Matter Volume and Morphological Brain Network, Journal of Affective Disorders. (2020) 266, 243–251, https://doi.org/10.1016/j.jad.2020.01.135.
10.1016/j.jad.2020.01.135
PubMed Web of Science® Google Scholar
28 Chen G., Chen P., and Yang Z., et al.Increased Functional Connectivity Between the Midbrain and Frontal Cortex Following Bright Light Therapy in Subthreshold Depression: A Randomized Clinical Trial, American Psychologist. (2024) 79, no. 3, 437–450, https://doi.org/10.1037/amp0001218.
10.1037/amp0001218
PubMed Google Scholar
29 Goodman S. N., Fanelli D., and Ioannidis J. P. A., What Does Research Reproducibility Mean?, Science Translational Medicine. (2016) 8, no. 341, https://doi.org/10.1126/scitranslmed.aaf5027, 2-s2.0-85063362367.
10.1126/scitranslmed.aaf5027
PubMed Web of Science® Google Scholar
30 Ioannidis J. P. A., Why Most Clinical Research Is Not Useful, PLOS Medicine. (2016) 13, no. 6, https://doi.org/10.1371/journal.pmed.1002049, 2-s2.0-84978105459, e1002049.
10.1371/journal.pmed.1002049
PubMed Web of Science® Google Scholar
31 Davatzikos C., Machine Learning in Neuroimaging: Progress and Challenges, NeuroImage. (2019) 197, 652–656, https://doi.org/10.1016/j.neuroimage.2018.10.003, 2-s2.0-85054430582.
10.1016/j.neuroimage.2018.10.003
PubMed Web of Science® Google Scholar
32 Dwyer D. B., Falkai P., and Koutsouleris N., Machine Learning Approaches for Clinical Psychology and Psychiatry, Annual Review of Clinical Psychology. (2018) 14, no. 1, 91–118, https://doi.org/10.1146/annurev-clinpsy-032816-045037, 2-s2.0-85046779663.
10.1146/annurev-clinpsy-032816-045037
PubMed Web of Science® Google Scholar
33 Cao J., Garro E. M., and Zhao Y., EEG/fNIRS Based Workload Classification Using Functional Brain Connectivity and Machine Learning, Sensors. (2022) 22, no. 19, https://doi.org/10.3390/s22197623, 7623.
10.3390/s22197623
Google Scholar
34 Chen S., Zhang X., and Lin S., et al.Suicide Risk Stratification Among Major Depressed Patients Based on a Machine Learning Approach and Whole-Brain Functional Connectivity, Journal of Affective Disorders. (2023) 322, 173–179, https://doi.org/10.1016/j.jad.2022.11.022.
10.1016/j.jad.2022.11.022
CAS PubMed Google Scholar
35 Zhu Y., Maikusa N., and Radua J., et al.Using Brain Structural Neuroimaging Measures to Predict Psychosis Onset for Individuals at Clinical High-Risk, Molecular Psychiatry. (2024) 29, no. 5, 1465–1477, https://doi.org/10.1038/s41380-024-02426-7.
10.1038/s41380-024-02426-7
PubMed Google Scholar
36 Eken A., Nassehi F., and Eroğul O., Diagnostic Machine Learning Applications on Clinical Populations Using Functional Near Infrared Spectroscopy: A Review, Reviews in the Neurosciences. (2024) 35, no. 4, 421–449, https://doi.org/10.1515/revneuro-2023-0117.
10.1515/revneuro-2023-0117
CAS PubMed Google Scholar
37 Rodríguez M. R., Nuevo R., Chatterji S., and Ayuso-Mateos J. L., Definitions and Factors Associated With Subthreshold Depressive Conditions: A Systematic Review, 2012.
Google Scholar
38 Beck A. T., Steer R. A., Ball R., and Ranieri W. F., Comparison of Beck Depression Inventories-IA and-II in Psychiatric Outpatients, Journal of Personality Assessment. (1996) 67, no. 3, 588–597, https://doi.org/10.1207/s15327752jpa6703_13, 2-s2.0-0029803149.
10.1207/s15327752jpa6703_13
CAS PubMed Web of Science® Google Scholar
39 Yang W., Wu D., and Peng F., Application of Chinese Version of Beck Depression Inventory-II to Chinese First-Year College Students, Chinese Journal of Clinical Psychology. (2012) 20, no. 6, 762–764, https://doi.org/10.16128/j.cnki.1005-3611.2012.06.020.
10.16128/j.cnki.1005-3611.2012.06.020
Google Scholar
40 Whisman M. A., Judd C. M., Whiteford N. T., and Gelhorn H. L., Measurement Invariance of the Beck Depression Inventory-Second Edition (BDI-II) Across Gender, Race, and Ethnicity in College Students, Assessment. (2013) 20, no. 4, 419–428, https://doi.org/10.1177/1073191112460273, 2-s2.0-84880632307.
10.1177/1073191112460273
PubMed Web of Science® Google Scholar
41 Huang J., Zhang J., Zhang T., Wang P., and Zheng Z., Increased Prefrontal Activation During Verbal Fluency Task After Repetitive Transcranial Magnetic Stimulation Treatment in Depression: A Functional Near-Infrared Spectroscopy Study, Frontiers in Psychiatry. (2022) 13, https://doi.org/10.3389/fpsyt.2022.876136, 876136.
10.3389/fpsyt.2022.876136
PubMed Web of Science® Google Scholar
42 Wei Y., Chen Q., and Curtin A., et al.Functional Near-Infrared Spectroscopy (fNIRS) as a Tool to Assist the Diagnosis of Major Psychiatric Disorders in a Chinese Population, European Archives of Psychiatry and Clinical Neuroscience. (2021) 271, no. 4, 745–757, https://doi.org/10.1007/s00406-020-01125-y.
10.1007/s00406-020-01125-y
PubMed Web of Science® Google Scholar
43 Singh A. K., Okamoto M., Dan H., Jurcak V., and Dan I., Spatial Registration of Multichannel Multi-Subject fNIRS Data to MNI Space without MRI, 2005.
Google Scholar
44 Tsuzuki D., Jurcak V., Singh A. K., Okamoto M., Watanabe E., and Dan I., Virtual Spatial Registration of Stand-Alone fNIRS Data to MNI Space, NeuroImage. (2007) 34, no. 4, 1506–1518, https://doi.org/10.1016/j.neuroimage.2006.10.043, 2-s2.0-33846581425.
10.1016/j.neuroimage.2006.10.043
PubMed Web of Science® Google Scholar
45 Ye J., Tak S., Jang K., Jung J., and Jang J., NIRS-SPM: Statistical Parametric Mapping for Near-Infrared Spectroscopy, NeuroImage. (2009) 44, no. 2, 428–447, https://doi.org/10.1016/j.neuroimage.2008.08.036, 2-s2.0-56349151815.
10.1016/j.neuroimage.2008.08.036
PubMed Web of Science® Google Scholar
46 Lancaster J. L., Woldorff M. G., and Parsons L. M., et al.Automated Talairach Atlas Labels for Functional Brain Mapping, Human Brain Mapping. (2000) 10, no. 3, 120–131.
10.1002/1097-0193(200007)10:3<120::AID-HBM30>3.0.CO;2-8
CAS PubMed Web of Science® Google Scholar
47 Huppert T. J., Diamond S. G., Franceschini M. A., and Boas D. A., HomER: A Review of Time-Series Analysis Methods for Near-Infrared Spectroscopy of the Brain, Applied Optics. (2009) 48, no. 10, https://doi.org/10.1364/AO.48.00D280, 2-s2.0-65249169372, D280.
10.1364/AO.48.00D280
PubMed Web of Science® Google Scholar
48 Cooper R. J., Selb J., and Gagnon L., et al.A Systematic Comparison of Motion Artifact Correction Techniques for Functional Near-Infrared Spectroscopy, Frontiers in Neuroscience. (2012) 6, https://doi.org/10.3389/fnins.2012.00147, 2-s2.0-84870944860.
10.3389/fnins.2012.00147
PubMed Google Scholar
49 Scholkmann F., Spichtig S., Muehlemann T., and Wolf M., How to Detect and Reduce Movement Artifacts in Near-Infrared Imaging Using Moving Standard Deviation and Spline Interpolation, Physiological Measurement. (2010) 31, no. 5, 649–662, https://doi.org/10.1088/0967-3334/31/5/004, 2-s2.0-77951098827.
10.1088/0967-3334/31/5/004
CAS PubMed Web of Science® Google Scholar
50 Zhu Y., Rodriguez-Paras C., Rhee J., and Mehta R. K., Methodological Approaches and Recommendations for Functional Near-Infrared Spectroscopy Applications in HF/E Research, Human Factors: The Journal of the Human Factors and Ergonomics Society. (2020) 62, no. 4, 613–642, https://doi.org/10.1177/0018720819845275, 2-s2.0-85066823460.
10.1177/0018720819845275
Google Scholar
51 Sakatani K., Yamashita D., and Yamanaka T., et al.Changes of Cerebral Blood Oxygenation and Optical Pathlength During Activation and Deactivation in the Prefrontal Cortex Measured by Time-Resolved Near Infrared Spectroscopy, Life Sciences. (2006) 78, no. 23, 2734–2741, https://doi.org/10.1016/j.lfs.2005.10.045, 2-s2.0-33646055793.
10.1016/j.lfs.2005.10.045
CAS PubMed Web of Science® Google Scholar
52 Strangman G., Boas D. A., and Sutton J. P., Non-Invasive Neuroimaging Using Near-Infrared Light, Biological Psychiatry. (2002) 52, no. 7, 679–693, https://doi.org/10.1016/S0006-3223(02)01550-0, 2-s2.0-0036790009.
10.1016/S0006-3223(02)01550-0
PubMed Web of Science® Google Scholar
53 Chen H., Lei Y., and Li R., et al.Resting-State EEG Dynamic Functional Connectivity Distinguishes Non-Psychotic Major Depression, Psychotic Major Depression and Schizophrenia, Molecular Psychiatry. (2024) 29, no. 4, 1088–1098, https://doi.org/10.1038/s41380-023-02395-3.
10.1038/s41380-023-02395-3
PubMed Google Scholar
54 Sarica A., Cerasa A., and Quattrone A., Random Forest Algorithm for the Classification of Neuroimaging Data in Alzheimer’s Disease: A Systematic Review, Frontiers in Aging Neuroscience. (2017) 9, https://doi.org/10.3389/fnagi.2017.00329, 2-s2.0-85031098588, 329.
10.3389/fnagi.2017.00329
PubMed Web of Science® Google Scholar
55 Hu J. and Szymczak S., A Review on Longitudinal Data Analysis With Random Forest, Briefings in Bioinformatics. (2023) 24, no. 2, https://doi.org/10.1093/bib/bbad002, bbad002.
10.1093/bib/bbad002
PubMed Web of Science® Google Scholar
56 Caruana R. and Niculescu-Mizil A., An Empirical Comparison of Supervised Learning Algorithms Using Different Performance Metrics, 2006, 161–168.
Google Scholar
57 Kumar V., Steinbach M., and Tan P., Introduction to Data Mining, 2006, 1st edition, Pearson Addison Wesley.
Google Scholar
58 Li Z., McIntyre R. S., and Husain S. F., et al.Identifying Neuroimaging Biomarkers of Major Depressive Disorder From Cortical Hemodynamic Responses Using Machine Learning Approaches, EBioMedicine. (2022) 79, https://doi.org/10.1016/j.ebiom.2022.104027, 104027.
10.1016/j.ebiom.2022.104027
CAS PubMed Web of Science® Google Scholar
59 Ho C. Sh, Chan Y. L., Tan T. Wk, Tay G. Wn, and Tang T. B., Improving the Diagnostic Accuracy for Major Depressive Disorder Using Machine Learning Algorithms Integrating Clinical and Near-Infrared Spectroscopy Data, Journal of Psychiatric Research. (2022) 147, 194–202, https://doi.org/10.1016/j.jpsychires.2022.01.026.
10.1016/j.jpsychires.2022.01.026
PubMed Google Scholar
60 Wesselhoeft R., Sørensen M. J., Heiervang E. R., and Bilenberg N., Subthreshold Depression in Children and Adolescents – A Systematic Review, Journal of Affective Disorders. (2013) 151, no. 1, 7–22, https://doi.org/10.1016/j.jad.2013.06.010, 2-s2.0-84888641796.
10.1016/j.jad.2013.06.010
PubMed Web of Science® Google Scholar
61 Szymkowicz S. M., Woods A. J., and Dotson V. M., et al.Associations Between Subclinical Depressive Symptoms and Reduced Brain Volume in Middle-Aged to Older Adults, Aging & Mental Health. (2018) 23, no. 7, 819–830, https://doi.org/10.1080/13607863.2018.1432030, 2-s2.0-85041231278.
10.1080/13607863.2018.1432030
PubMed Google Scholar
62 Vulser H., Paillère Martinot M.-L., and Artiges E., et al.Early Variations in White Matter Microstructure and Depression Outcome in Adolescents With Subthreshold Depression, American Journal of Psychiatry. (2018) 175, no. 12, 1255–1264, https://doi.org/10.1176/appi.ajp.2018.17070825, 2-s2.0-85057593778.
10.1176/appi.ajp.2018.17070825
PubMed Web of Science® Google Scholar
63 Guo Y., Lu R., and Ou Y., et al.A Study on the Association Between Prefrontal Functional Connectivity and Non-Suicidal Self-Injury in Adolescents With Depression, Frontiers in Neurology. (2024) 15, https://doi.org/10.3389/fneur.2024.1382136, 1382136.
10.3389/fneur.2024.1382136
PubMed Google Scholar
64 Backes H., Dietsche B., and Nagels A., et al.Increased Neural Activity during Overt and Continuous Semantic Verbal Fluency in Major Depression: Mainly a Failure to Deactivate, European Archives of Psychiatry and Clinical Neuroscience. (2014) .
10.1007/s00406-014-0491-y
Google Scholar
65 Miskowiak K. W., Glerup L., and Vestbo C., et al.Different Neural and Cognitive Response to Emotional Faces in Healthy Monozygotic Twins at Risk of Depression, Psychological Medicine. (2015) 45, no. 7, 1447–1458, https://doi.org/10.1017/S0033291714002542, 2-s2.0-84925847179.
10.1017/S0033291714002542
CAS PubMed Web of Science® Google Scholar
66 Sahib A. K., Modulation of Inhibitory Control Networks Relate to Clinical Response following Ketamine Therapy in Major Depression, 2020.
Google Scholar
67 Nachev P., Kennard C., and Husain M., Functional Role of the Supplementary and Pre-Supplementary Motor Areas, Nature Reviews Neuroscience. (2008) 9, no. 11, 856–869, https://doi.org/10.1038/nrn2478, 2-s2.0-56749186799.
10.1038/nrn2478
CAS PubMed Web of Science® Google Scholar
68 Miller E. K., The Prefrontal Cortex and Cognitive Control, 2000.
Google Scholar
69 Muggleton N. G., Juan C.-H., Cowey A., Walsh V., and O’Breathnach U., Human Frontal Eye Fields and Target Switching, 2010.
Google Scholar
70 Murd C., Moisa M., Grueschow M., Polania R., and Ruff C. C., Causal Contributions of Human Frontal Eye Fields to Distinct Aspects of Decision Formation, Scientific Reports. (2020) 10, no. 1, https://doi.org/10.1038/s41598-020-64064-7, 7317.
10.1038/s41598-020-64064-7
CAS PubMed Google Scholar
71 Offen S., Gardner J., Schluppeck D., and Heeger D., Differential Roles for Frontal Eye Fields (FEF) and Intraparietal Sulcus (IPS) in Visual Working Memory and Visual Attention, 2010.
Google Scholar
72 Yuan P. and Raz N., Prefrontal Cortex and Executive Functions in Healthy Adults: A Meta-Analysis of Structural Neuroimaging Studies, 2015.
Google Scholar
73 Jiang C. and Mei Y., Encyclopedia of China: Education, 1994.
Google Scholar
74 Chen M., He Y., and Hao L., et al.Default Mode Network Scaffolds Immature Frontoparietal Network in Cognitive Development, Cerebral Cortex. (2023) 33, no. 9, 5251–5263, https://doi.org/10.1093/cercor/bhac414.
10.1093/cercor/bhac414
PubMed Web of Science® Google Scholar
75 Ester E. F., Sutterer D. W., Serences J. T., and Awh E., Feature-Selective Attentional Modulations in Human Frontoparietal Cortex, Journal of Neuroscience. (2016) 36, no. 31, 8188–8199, https://doi.org/10.1523/JNEUROSCI.3935-15.2016, 2-s2.0-84982947641.
10.1523/JNEUROSCI.3935-15.2016
CAS PubMed Google Scholar
76 Kaiser R. H., Andrews-Hanna J. R., Wager T. D., and Pizzagalli D. A., Large-Scale Network Dysfunction in Major Depressive Disorder: A Meta-Analysis of Resting-State Functional Connectivity, JAMA Psychiatry. (2015) 72, no. 6, https://doi.org/10.1001/jamapsychiatry.2015.0071, 2-s2.0-84931081503, 603.
10.1001/jamapsychiatry.2015.0071
PubMed Web of Science® Google Scholar
77 Dionne J. K., Meehan S. K., Legon W., and Staines W. R., Crossmodal Influences in Somatosensory Cortex: Interaction of Vision and Touch, Human Brain Mapping. (2010) 31, no. 1, 14–25, https://doi.org/10.1002/hbm.20841, 2-s2.0-74049117603.
10.1002/hbm.20841
PubMed Web of Science® Google Scholar
78 Mesulam M., Large-Scale Neurocognitive Networks and Distributed Processing for Attention, Language, and Memory, Annals of Neurology. (1990) 28, no. 5, 597–613, https://doi.org/10.1002/ana.410280502, 2-s2.0-0025029346.
10.1002/ana.410280502
CAS PubMed Web of Science® Google Scholar

All articles

Identification of Subthreshold Depression Based on fNIRS–VFT Functional Connectivity: A Machine Learning Approach

Abstract

1. Introduction