Resting-state functional magnetic resonance imaging and support vector machines for the diagnosis of major depressive disorder in adolescents

Abstract

BACKGROUND

Research has found that the amygdala plays a significant role in underlying pathology of major depressive disorder (MDD). However, few studies have explored machine learning-assisted diagnostic biomarkers based on amygdala functional connectivity (FC).

AIM

To investigate the analysis of neuroimaging biomarkers as a streamlined approach for the diagnosis of MDD in adolescents.

METHODS

Forty-four adolescents diagnosed with MDD and 43 healthy controls were enrolled in the study. Using resting-state functional magnetic resonance imaging, the FC was compared between the adolescents with MDD and the healthy controls, with the bilateral amygdala serving as the seed point, followed by statistical analysis of the results. The support vector machine (SVM) method was then applied to classify functional connections in various brain regions and to evaluate the neurophysiological characteristics associated with MDD.

RESULTS

Compared to the controls and using the bilateral amygdala as the region of interest, patients with MDD showed significantly lower FC values in the left inferior temporal gyrus, bilateral calcarine, right lingual gyrus, and left superior occipital gyrus. However, there was an increase in the FC value in Vermis-10. The SVM analysis revealed that the reduction in the FC value in the right lingual gyrus could effectively differentiate patients with MDD from healthy controls, achieving a diagnostic accuracy of 83.91%, sensitivity of 79.55%, specificity of 88.37%, and an area under the curve of 67.65%.

CONCLUSION

The results showed that an abnormal FC value in the right lingual gyrus was effective as a neuroimaging biomarker to distinguish patients with MDD from healthy controls.

Key Words: Major depressive disorder; Adolescent; Support vector machine; Machine learning; Resting-state functional magnetic resonance imaging; Neuroimaging; Biomarker

Core Tip: We want to explore the potential neuroimaging biomarkers of adolescents with major depressive disorder using resting-state functional magnetic resonance imaging and support vector machines. The results showed that using the abnormal functional connectivity value of the right linguistic gyrus as a biomarker to distinguish patients and healthy controls had certain advantages, which was of great significance for the early diagnosis and treatment of adolescent major depressive disorder.

INTRODUCTION

Major depressive disorder (MDD) in adolescents is a relatively common and severe mental health condition that has a significant impact on both individuals and society[1,2]. Early diagnosis and treatment are crucial for improving the prognosis of adolescents with MDD[3]. However, traditional diagnostic methods rely primarily on clinical symptoms and the scores of depression scales[4]. Unfortunately, as adolescents are not yet cognitively or psychologically mature, the self-reporting process may be subject to bias, and the lack of objective biological markers often leads to misdiagnosis and missed diagnosis[5], reducing the accuracy and repeatability of the diagnosis. Therefore, the search for reliable biomarkers[6] is essential for early diagnosis and treatment of adolescent MDD.

The development of neuroimaging technology has provided a new way to study the biological basis of adolescent MDD[7]. In particular, analysis of functional connectivity (FC) using resting-state functional magnetic resonance imaging (rs-fMRI) has provided valuable insights into the intrinsic functional structure of the brain and how it is altered in psychiatric disorders[8,9]. Rs-fMRI is a non-invasive neuroimaging technique that can be used to study the FC of the brain in non-task-specific states. In terms of the evaluation of adolescent MDD, rs-fMRI has become an important tool for identifying disease-related changes in brain networks[10,11]. FC refers to statistical dependencies between different brain regions, and is a reflection of the way in which information is transmitted and processed within the brain. Studies have shown that adolescents with disease symptoms and impairments in cognitive function. Therefore, biomarkers based on FC may help identify adolescents with depression[12]. The amygdala is a major center for emotional processing in the brain, and has been found to play a key role in the pathophysiology of depression[13]. Studies have shown that patients with MDD often show abnormal activity in the amygdala, and functional connections between the amygdala and other brain regions are also strongly associated with depressive symptoms[14,15]. Therefore, the present study used the amygdala as a seed point to analyze changes in FC between various brain regions in adolescents with MDD[16,17].

Currently, the application of machine learning in clinical diagnosis has become more and more extensive[18]. Support vector machines (SVM) are powerful machine learning algorithms that can identify the features of data that best distinguish different categories by finding the optimal hyperplane[19]. In neuroimaging studies, SVM can be used to establish models predicting the value of biomarkers based on FC[20], thus providing automatic identification and classification of adolescents with depression. It has been reported that predictive models based on FC and SVM have high levels of accuracy and repeatability in identifying adolescents with MDD[21].

In this study, rs-fMRI technology was used to collect brain FC data in adolescents with MDD and healthy controls (HCs). These data are typically derived from resting brain activity and reflect the level of synchronized activity between different brain regions. Preprocessing and feature extraction from these FC data allow the identification of network features of the brain associated with depression. The SVM algorithm was then used for pattern analysis and classification of these features[22]. SVM is able to identify an optimal hyperplane that distinguishes FC patterns in MDD patients from those of HCs. The performance of the classification model can be evaluated through cross-validation and statistical analysis[23], including information on its accuracy, sensitivity, specificity, reliability, and other indicators[24].

Based on these methodological advances, the purpose of this study was to explore the use of rs-fMRI and SVM in the diagnosis of MDD in adolescents. The objective was to identify neuroimaging biomarkers associated with adolescent MDD by conducting pattern analysis and machine learning modeling on FC data from adolescents with MDD and HCs[25], and evaluate their application in disease diagnosis. This would help clinicians identify adolescents with MDD more accurately[26], enabling early intervention and treatment. In addition, the identification of the features of brain networks associated with MDD[27], provides insight into the neurobiological mechanisms of adolescent MDD, as well as a theoretical basis for future treatment strategies.

In this study, rs-fMRI combined with the SVM algorithm was used to identify a new biomarker-based method for the diagnosis of adolescent MDD. It is hoped that this method can be applied in clinical practice and provide new ideas and methods for early identification and treatment of adolescent MDD.

MATERIALS AND METHODS

Patients

The adolescents with MDD consisted of 44 individuals, comprising 10 males and 34 females, with an average age of 15 ± 2 years and an average of 10 ± 3 years of schooling. The patients were recruited at the psychiatric outpatient department of the First Affiliated Hospital of Chongqing Medical University, China, between August 2021 and July 2022. Two experienced attending psychiatrists conducted the Mini international neuropsychiatric interview for children and adolescents (MINI-kid)[28] for each patient. The 17-item Hamilton depression rating scale (HAMD-17)[29] was also used for diagnosis. All participants met the diagnostic and statistical manual of mental disorders (DSM)-IV criteria for a “depressive episode” as defined in the diagnostic and statistical manual of mental disorders, 4^th edition[30].

The inclusion criteria for the enrolled adolescents with MDD were as follows: (1) The participants should have at least a background of primary school education, and be aged between 9 and 18 years old. This age group was chosen because it covers most of adolescence, a critical period of rapid mental and physical change for individuals and a time when the symptoms of depression appear and intensify; (2) The patients should be fully evaluated by at least two experienced psychiatrists, using the MINI-kid interview scale as the primary diagnostic tool. In addition, their symptoms were required to meet the diagnostic criteria of a “depressive episode” in the DSM-IV; (3) Participants were required to have a HAMD-17 score of 17 or over; (4) Only patients who had been first diagnosed with adolescent MDD and who had not taken any psychiatric medication or received any form of treatment in the previous four weeks were included; and (5) To control for the potential influence of genetic backgrounds and physiological differences on the research results, only participants who were of Han nationality and were right-handed were enrolled.

The exclusion criteria for adolescents with depression were as follows: (1) Individuals with neurological disorders (such as epilepsy or multiple sclerosis) or other major physical health problems (such as heart disease or cancer) where the symptoms associated with these disorders may have interacted with depressive symptoms or have a significant impact on the subject's overall health were excluded; (2) Subjects with other mental disorders or borderline personality disorder that could influence the clinical presentation and treatment response of depression and thus interfere with the accuracy of the study results, were excluded; (3) Placeholder patients, as these may affect the diagnosis and clinical manifestations of depression; (4) Patients with a family history of mental illness or self-injurious behavior were excluded to reduce the influence of genetic factors on the study results; (5) Patients with a history of severe craniocerebral trauma associated with loss of consciousness, which may have had long-term effects on brain structure and function, were excluded; (6) Patients with a history of drug or alcohol abuse or dependence were excluded, as these factors could alter the chemical balance of the brain and affect mood and behavior, thereby interfering with understanding the nature of depression; (7) Patients with contraindications to MRI scanning or metal foreign bodies that could affect image quality (such as pacemakers, certain metal implants, and dental orthotics) and who could not complete MRI scanning were also excluded, as these individuals are not suitable for MRI scanning or experience mental or physical discomfort during the scanning process; and (8) The young mania rating scale score was utilized for assessment, excluding depressive episodes of bipolar disorder and thereby avoiding influencing the diagnosis of depression.

HCs: Forty-three normal volunteers were recruited in the community. The inclusion criteria for the normal control group were that they matched the adolescents with MDD in terms of sex, age, education, Han nationality, and being right-handed. This matching was designed to ensure consistency in the baseline demographic characteristics between the two groups so that the results are an accurate reflection of the specific effects of depression rather than differences due to baseline variables. In addition, the controls had HAMD-17 scores below 7 to ensure that they did not have significant depressive symptoms. This criterion was set to explicitly ensure that the members of the control group had good mental health, thus providing a clear baseline for the study. The exclusion criteria of the normal control group were the same as those for the adolescents with MDD.

fMRI data acquisition

The Signa HDx 3.0T MRI (GE, United States) utilizes a standard 8-channel head coil for scanning. The sequences employed included T2 weight imaging (T2WI), fluid attenuated inversion recovery (FLAIR), high-resolution T1 imaging, and blood oxygen level dependent (BOLD) imaging. The detailed parameters included the use of a fast spin echo sequence with a repetition time (TR) of 8400 millisecond and an echo time (TE) of 120 millisecond by the T2WI FLAIR. The field of view (FOV) was 240 mm × 240 mm, with one excitation time. The reconstruction matrix was 288 × 210, the slice thickness and spacing were 5.0 mm and 1.5 mm respectively, resulting in 22 image layers, and the scanning duration was 1 minute and 50 seconds. For high-resolution T1 imaging, an inversion recovery and spoiled gradient recalled echo sequence were used, with TR of 8.3 millisecond, TE of 3.3 millisecond, and a flip angle of 12 degrees. The FOV was 24 cm × 24 cm, with one excitation time, a reconstruction matrix of 256 × 256, and a slice thickness and slice spacing of 1.0 mm and 0 mm, respectively. The duration of this sequence was 3 minutes and 20 seconds. The BOLD imaging sequence employed a gradient echo-planar imaging sequence, with a TR of 2000 millisecond, TE of 30 millisecond, and a flip angle of 90 degrees. The FOV was 24 cm × 24 cm, with one excitation time, a reconstruction matrix of 64 × 64, a slice thickness of 3.5 mm, a slice spacing of 0 mm, and included 240 time points. The duration of this scan sequence was 8 minutes.

Data preprocessing

The rs-fMRI data processing and analysis were conducted using the matrix laboratory (MATLAB) platform with RESTplus v1.27 software. The procedure involved several specific steps, namely, the initial conversion of the image data format and the discarding of the first 10 time points, after which time slice and head motion corrections were performed. Spatial standardization was achieved to a resolution of 3 mm × 3 mm × 3 mm, followed by smoothing with a full width at half maximum value of 4 mm × 4 mm × 4 mm. Linear drift was removed, and covariates such as Friston 24, white matter, and cerebrospinal fluid signals were excluded. Finally, low-frequency filtering was applied within the range of 0.01 to 0.08 Hz.

Functional connection analysis

Using RESTplus v1.27 software, the seed points in the left and right amygdalas were extracted using the analytical automatic labeling template, followed by an analysis of FC with whole brain voxels. Fisher’s test was used to obtain the z values, ensuring that they conformed to a normal distribution.

Statistical analysis

Demographic and clinical data were analyzed using statistical product and service solutions (IBM Corp., Armonk, NY, United States). Two sample t-tests were used to determine and analyze differences in age and education between the MDD and HCs groups, and χ² tests were used to assess differences in sex between the groups. P values < 0.05 were considered statistically significant.

Using statistical parametric mapping software on the MATLAB platform, two independent sample t-tests were conducted on the FC of the left and right amygdalas between the MDD and HC groups. Sex, age, and years of education were used as covariates to identify brain regions with differences (P < 0.01, voxel number 18). False discovery rate corrections were also performed, with a main threshold of P < 0.01.

Classification and receiver operating characteristics analysis

Using the SVM library software package in MATLAB, SVM was used to classify the FCs in different brain regions and identify neurophysiological features associated with depression. Receiver operating characteristic (ROC) curves were then drawn to assess the accuracy of the criteria for distinguishing between patients with MDD and HCs in the specific brain regions.

RESULTS

Demographic and clinical characteristics

Two adolescents with MDD were excluded for the following reasons: One due to an intracranial mass and the other due to a pituitary hormone disorder. One HC was excluded due to the presence of orthodontic appliances, which precluded MRI scanning. Ultimately, a total of 44 adolescent patients with MDD and 43 HC were included in the analysis (Figure 1).

Figure 1 Flow chart of the selection process for the major depressive disorder patients and the healthy controls. MDD: Major depressive disorder; HCs: Healthy controls; MINI-kid: The mini international neuropsychiatric interview for children and adolescents; HAMD-17: The 17-item Hamilton depression rating scale; DSM-IV: The diagnostic and statistical manual of mental disorders, 4^th edition; rs-fMRI: Resting-state functional magnetic resonance imaging.

There were no significant differences in sex, age, and educational level between patients with MDD and HCs (P > 0.05). However, there was a significant difference (P < 0.05) in Hamilton anxiety scale scores between the two groups (Table 1).

Table 1 Demographic and clinical characteristics comparison of the two groups, mean ± SD.

Demographic data	Patients (n = 44)	HCs (n = 43)	t value (orx²)	P value
Gender (male/female)	44 (10/34)	43 (10/33)	0.003	0.9531
Age (years)	15.02 ± 1.389	15.49 ± 1.831	-1.338	0.1842
Education (years)	9.43 ± 1.662	9.98 ± 2.006	-1.381	0.1712
HAMD score	27.09 ± 6.069	1.00 ± 1.746	27.114	< 0.0012

¹P for gender distribution was obtained by the χ² test.

²P were obtained by two sample I-tests.

HCs: Healthy controls; HAMD: Hamilton depression scale.

fMRI results

Using a two-sample t-test to compare the differences in FC values between patients with MDD patients and HCs, it was found that the FC values of patients with MDD were significantly reduced in the left inferior temporal gyrus, bilateral calcarine, right lingual gyrus, and left superior orbital gyrus with the left amygdala compared to HCs, while the FC values in the left amygdala Permis-10 were increased (Figure 2 and Table 2).

Figure 2 Differences in functional connectivity values between patients with major depressive disorder patients and the healthy controls. The color bar represents the t-values in the group analysis.

Table 2 Abnormal functional connectivity of brain regions in patients with major depressive disorder.

Cluster location	Peak (MNI)			Cluster size (voxels)	t value
	X	Y	Z
Patients < HCs
Temporal_Inf_L	-45	3	-36	24	-6.08
Vermis_10	0	-45	-30	25	6.09
Calcarine_R	15	-93	-3	21	-5.19
Calcarine_L	-9	-78	3	38	-4.87
Lingual_R	24	-48	-3	21	-5.43
Occipital_Sup_L	-15	-84	21	41	-5.33

MNI: Montreal neurological institute.

SVM results

The SVM method was used for the classification of patients with MDD and HCs. The results showed that the decrease in FC value in the right lingual gyrus could easily distinguish between the MDD and HC groups with good accuracy (83.91%), sensitivity (79.55%), and specificity (88.37%) (Figure 3 and Figure 4).

Figure 3 Visualization of the support vector machine classification based on reduced functional connectivity values in the right lingual gyrus for the differentiation of patients with major depressive disorder from healthy controls. A: Three-dimensional visualization of support vector machine with the most optimal parameters; B: Classification map of functional connectivity values for the right lingual gyrus.

Figure 4 Assessment of the accuracy of the use of abnormal functional connectivity values in different regions of the brain for distinguishing between patients with major depressive disorder patients and the healthy controls. FC: Functional connection.

ROC results

The ROC curves used to analyze the accuracy of the right lingual gyrus revealed that the FC value in this region was effective for distinguishing between patients with MDD and HCs, achieving both good sensitivity and specificity. These results indicated that abnormal FC values in the right lingual gyrus are effective for use as biomarkers for distinguishing between patients with MDD and HCs (Figure 5 and Figure 6 and Table 3).

Figure 5 The functional connectivity value of the right lingual gyrus is effective for distinguishing between patients with major depressive disorder patients and the healthy controls, based on the receiver operating characteristic curve. AUC: Area under the curve.

Figure 6  Radar map showing the accuracy, sensitivity, and specificity of the use of the functional connectivity value in the right lingual gyrus for distinguishing between patients with major depressive disorder patients and the healthy controls, together with the corresponding area under the curve values.

Table 3 Results of the receiver operating characteristic analysis for functional connection values of different brain regions in distinguishing between the two groups.

Brain regions	Accuracy	Sensitivity	Specificity
Temporal_Inf_L	68.97%	79.55%	58.14%
Vermis_10	70.11%	84.09%	55.81%
Calcarine_R	70.11%	61.36%	79.07%
Calcarine_L	66.67%	75.00%	58.14%
Lingual_R	83.91%	79.55%	88.37%
Occipital_Sup_L	65.52%	68.19%	62.79%

DISCUSSION

The present study used rs-fMRI technology[31] to collect data on brain FC from adolescents with MDD and HCs. Following the preprocessing of the data together with feature extraction, several brain network features related to adolescent MDD were identified. These features may reflect abnormal changes in brain FC in patients with depression[32], providing important clues for the understanding of the underlying neurobiological mechanisms of depression. The SVM algorithm was then used to perform pattern analysis and classification on these features[33,34]. The results indicated that the SVM model based on FC had both high accuracy and repeatability in identifying adolescents with MDD[35]. This indicates that the SVM algorithm is an effective machine-learning method[36] that can assist in the discovery and utilization of biomarkers, and identified a novel biomarker-based approach for the early recognition and treatment of adolescent MDD[37].

Studies have shown that abnormal functioning of the amygdala, which plays a key role in the processing of emotional information in the brain, is a major factor contributing to depression[15], and changes in the amygdala are one of the biomarkers reflecting depression[38]. The present study used rs-fMRI and FC analysis to investigate changes in FC in various brain regions between adolescents with MDD and HCs, with the left and right amygdalas as regions of interest[39,40]. The results indicated the presence of significant changes in the FC between the left amygdala and certain brain regions in adolescents with MDD. Compared to HCs, patients with MDD showed significantly reduced FC values relative to HCs in the left inferior temporal gyrus[41,42], bilateral calcarine[43], right lingual gyrus[44], and left superior occipital gyrus[45], while increased values were observed in Vermis-10[46]. Subsequent analysis using SVM revealed that a reduced FC value in the right lingual gyrus was effective in distinguishing patients with MDD from HCs, demonstrating a diagnostic accuracy of 83.91%, sensitivity of 79.55%, and specificity of 88.37%. Further analysis using ROC curves showed that the area under the curve[47] was 67.65%. These results indicate that an abnormal FC value associated with the left amygdala-right lingual gyrus has potential as a neuroimaging biomarker for distinguishing patients with MDD from HCs, and has high predictive efficacy for diagnosing adolescent MDD.

The right lingual gyrus is located in the temporal lobe of the brain and is usually involved in sensory processing and the understanding of language[48]. However, in recent years, more and more studies have shown that the right lingual gyrus is also involved in both emotional regulation and cognitive control[49]. The findings of the present study suggested that an abnormal functional connection between the left amygdala and the right lingual gyrus may be related to the impairment of emotional regulation and cognitive control seen in adolescent patients with depression, providing a new clue for the further understanding of the underlying neurobiological mechanism of depression. This suggests that assessment of the right lingual gyrus could be used in the diagnosis of adolescent MDD[50]. The accuracy and repeatability of the diagnosis of depression can be enhanced by combining this tool with other clinical evaluation indicators, such as the patient’s symptoms and results of psychological evaluations[51]. In addition, the application of biomarkers in the right lingual gyrus may also be helpful in monitoring the effects of treatment and evaluation of prognosis[52].

The results of this study provide new ideas and methods for the early diagnosis and treatment of adolescent MDD. Diagnostic tools based on rs-fMRI and SVM can assist clinicians in the accurate identification of patients with depression[53], allowing early intervention and treatment. In addition, the identification of brain network features associated with depression enhances understanding of the neurobiological mechanisms underlying adolescent MDD, and provides a theoretical basis for future treatment strategies.

Although this study has achieved some results, there are still some limitations. Firstly, the sample size was relatively small, which may limit the universality and generalizability of the research results; Future studies should consider expanding the sample size and conducting multicenter studies to verify the reliability of the present results. Secondly, the study only considered the characteristics of FC. Future studies could try to combine other types of neuroimaging data, such as structural MRI[54] and diffusion tensor imaging[55] to further improve the accuracy and reliability of diagnosis. In addition, other machine learning methods and technologies[56,57] can be explored to optimize the performance of the model.

CONCLUSION

In conclusion, this study explored the diagnosis of adolescent MDD by using rs-fMRI and SVM[58,59], leading to the identification of disease-related biomarkers and providing a new perspective for understanding the neural mechanisms underlying adolescent MDD, as well as evaluating its potential value in disease diagnosis. It was found that the right lingual gyrus may be an effective biomarker for adolescent MDD. This finding provides new ideas and methods for the early diagnosis and treatment of adolescent MDD. Although it was found that the right lingual gyrus may be a major biomarker, further studies are needed to verify this finding. In addition, the neurobiological basis of abnormal FC of the right lingual gyrus and its role in the pathogenesis of depression requires further investigation. It is hoped hope that future research can further verify and use this biomarker to bring revolutionary changes to the diagnosis and treatment of adolescent MDD.

ACKNOWLEDGEMENTS

The authors would like to thank all the reviewers who participated in the review process, as well as to the patients and their families for their support of the study.