Investigation of quantitative synthetic MRI in the evaluation of attention-deficit/hyperactivity disorder

Introduction

Attention-deficit/hyperactivity disorder (ADHD) is one of the most common childhood-onset psychiatric disorders and is clinically characterized by inattention, hyperactivity and impulsivity (1). Growing studies pointed out that up to 3.4% of children were affected with a ratio of 3-4:1 for boys and girls (2). Further, approximately 64% of children with ADHD continue to have symptoms into adulthood, and these symptoms would increase the high risk of substance abuse and worsen the poor quality of social and academic life (1-3). Currently, the diagnosis of ADHD relies primarily on clinical signs of children and questionnaires from parents and teachers on observations of behavioral symptoms. While the accurate clinical diagnosis is crucial for individuals with ADHD, a deeper exploration of the fundamental brain pathology is even more essential for elucidating the underlying mechanisms.

Over the past years, there has been increasing interest in the application of machine learning to neuroimaging data in various psychiatric disorders including ADHD (4). The combination of machine learning and neuroimaging data in ADHD allows for both individual-level classification and the effective capture of subtle and spatially distributed information due to taking the intercorrelation between regions into consideration (5). To date, studies of structural magnetic resonance imaging (sMRI) (6), diffusion tensor imaging (DTI) (7), task-based functional magnetic resonance imaging (fMRI) (8) and resting-state fMRI (rs-fMRI) (9) combined with machine learning have shown widespread and distributed micro- and macrostructural changes (i.e., in the frontal lobe, temporal lobe and cerebellum) in ADHD with classification model accuracy varying from 60% to 90%. Nevertheless, the heterogeneous results from diverse neuroimaging modalities highlighted the need for further exploration to develop reliable and consistent biomarker for clinical diagnostic translation and uncover a unified neurobiological mechanism underlying ADHD.

More recently, the newly developed synthetic magnetic resonance imaging (SyMRI), which is performed by multi-dynamic multi-echo (MDME) sequence to quantify tissue relaxometry, has been proven to be a useful and objective neuroimaging approach to detect disease-related tissue pathology (10). The quantitative MRI technique provides direct measurement of actual tissue parameters (i.e., T1 and T2 relaxometry values), as well as the measurement of myelin content with high utility and reproducibility within a few minutes of scanning (11). Generally, the relaxometry values depend on the biophysical and biochemical environment of the tissue, and its alterations are influenced by the content of microstructural components (i.e., iron, myelin, gliosis, and water). Thus, multiparametric quantifications are considered to provide early detection and comprehensive characterization of the microscopic processes related to tissue remodeling. Studies using SyMRI-derived quantitative measurements have shown considerable potential in the investigation of the brain in aging (12) and neurologic diseases such as autism (13), multiple sclerosis (14), and epilepsy (15). However, to date, no studies have focused on quantitative SyMRI combined with machine learning to explore ADHD-related brain pathology.

In the present study, we employed the multi-parameter and multi-feature classification approach to distinguish ADHD from healthy controls (HCs) based on whole-brain gray matter and white matter quantitative values obtained from SyMRI data. Additionally, we evaluated and compared the classification performance based on every single feature and combined features, as well as the classification performance across different models. We expected that the combined gray and white matter quantitative features could achieve robust performance across different classification models, and these discriminative brain regions would help elucidate the ADHD-related intrinsic microstructural changes. We present this article in accordance with the TRIPOD+AI reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1266/rc).

Methods

Subjects

The ADHD cohort was enrolled from the pediatrics department of The First Affiliated Hospital of Sun Yat-sen University according to the following criteria: (I) diagnosis was determined in accordance with the Diagnostic and Statistical Manual of Mental Disorders 5th edition (DSM-V) criteria (16), as well as supported by the parent and teacher reports on Conners Symptom Questionnaire (17), with ADHD positively defined as indices ≥75^th percentile in both evaluations; (II) age between 6–14 years and right-handedness; (III) without psychoactive drug use (namely the drug-naïve participants); (IV) no contraindications for MRI scanning; (V) MRI data underwent image quality control, with exclusions for apparent artifacts or excessive motion; (VI) absence of head trauma, neurological disorders or psychiatric comorbidities. The HCs recruited via public advertisement campaign were screened using the same questionnaire to rule out ADHD diagnosis and underwent systematic clinical interviews conducted by an experienced pediatrician to verify the absence of structural brain lesions or neuropsychiatric disorders. Additionally, the ADHD participants completed the digital cancellation test (DCT) (18-20) to assess attention function. Performance was quantified using the cancellation test score and concentration index, with lower scores indicating attention deficits. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committee of The First Affiliated Hospital of Sun Yat-sen University (No. [2019]328). Informed consent was obtained from the legal guardians of all participants.

MRI acquisition

All participants underwent MRI scanning on a 3.0 T scanner (SIGNA Pioneer GE Healthcare, WI, USA) equipped with a 32-channel head coil array. High-resolution 3D structural T1-weighted data were acquired via the fast spoiled gradient recalled echo (FSPGR) sequence [repetition time (TR) =8.6 ms, echo time (TE) =3.3 ms, flip angle =12°, slice thickness/gap =1.0 mm/0 gap, slice number =192, field of view (FOV) =256 mm × 256 mm, and matrix =256×256]. Quantitative SyMRI data were obtained using 2D MDME sequence with 2 echo times and 4 delay times (TR =10,205.0 ms, TE_1/2 =11.3 ms/90.3 ms, thickness/gap =2 mm/0 gap, slice number =192, FOV =256 mm × 256 mm, and matrix =128×128, scanning time =5.5 min). T2-weighted FLAIR images were also acquired to rule out gross cranial lesions. To mitigate the head motion and scanning noise, the foam padding and earplugs were used for each participant during scanning.

SyMRI data processing

Post-processing of SyMRI data was conducted using the SyMRI software (version 11.22; SyntheticMR, Linköping, Sweden) to obtain quantitative maps including T1, T2 and proton density (PD) maps. Meanwhile, the myelin volume fraction (MVF) can be estimated based on four compartments model (i.e., myelin volume, cellular volume, free water volume, and excess parenchymal water volume fraction) which assumes that compartment-specific T1, T2 and PD values of each compartment contribute to the effective T1, T2 and PD values measured in each voxel (21). Besides, the processing procedures for gray matter quantification mainly involved the following steps: the co-registration of individual T1W images with quantitative T1 maps; spatial normalization of co-registered T1W images and the resulting transformation information were then applied to quantitative maps to transfer them into standardized space. For white matter quantification, we followed the established procedures as described in our previous study (22), which involved spatial normalization procedures for white matter structures. In particular, synthetic PD map served as registration intermediary for spatial normalization of white matter structures. This strategy was implemented based on its spatial coherence with other quantitative maps and tissue contrast properties analogous to T1W images. The approach could improve the normalization accuracy for white matter structures especially for the MVF map. To enable regional analysis, the Automated Anatomical Labeling 116 (AAL116) atlas (23) for gray matter and the International Consortium for Brain Mapping (ICBM) atlas (24) for white matter were used in our study. From the normalized maps, mean T1, T2 values were extracted for 116 gray matter regions, and mean T1, T2 values and MVF values were extracted for 50 white matter regions. Finally, we obtained 116×2=232 gray matter quantitative values and 50×3=150 white matter quantitative values per participant for subsequent analytical procedures.

Classification models with machine learning

The extracted quantitative values from gray and white matter served as initial features for classification models, and three machine learning algorithms, namely support vector machine (SVM), random forest (RF) and linear regression (LR), were adopted in this study. For model fitting and evaluation, nested cross-validation scheme was used in the current study. In the outer loop, a leave-one-out cross-validation (LOOCV) scheme was employed, iteratively designating each subject as the testing set while the remaining subjects for model training. Within each training set, an inner 5-fold cross-validation process with grid search was performed for hyperparameter optimization. To avoid overfitting risks, feature selection was implemented to obtain optimal subset of features, thereby enhancing the robustness and generalizability of the classification model (25). Hence, we performed a two-sample t-test on the Z-score standardized features, and kept the features with P<0.05. Then the least absolute shrinkage and selection operator (LASSO), a sparse learning algorithm based on L1 regularization, was adopted with 5-fold cross-validation to identify the optimal alpha and further retain features with non-zero coefficients. Notably, to mitigate information leakage, we implemented the feature selection strategy within LOOCV framework. Briefly, in each iteration of LOOCV, feature selection was performed exclusively on the training subset, and the selected features were then applied to the training and test data for model fitting and evaluation. Thus, the selected feature set slightly differed from iteration to iteration due to the different training datasets in each iteration of LOOCV. To better illustrate the features with high discriminative power, we presented the features with selection frequency more than 70 times as previous studies described (26). Model performance was assessed using sensitivity, specificity, accuracy and area under the curve (AUC). The significance of the model performance was evaluated through 1,000 times permutation test. The analytical pipeline described above was carried out in Python 3.8, primarily utilizing the scikit-learn package for the machine learning workflow (27).

Statistical analysis

The Shapiro-Wilk (S-W) test was conducted to test the normal distribution of continuous variables. We conducted two-sample t-test for quantitative variables and chi-squared test for qualitative variables. To estimate the significance for the machine learning models, we performed a nonparametric permutation test to calculate the P value for accuracy. This involved repeating the classification process 1,000 times with randomly reassigned group labels. The number of times achieving higher accuracy than that obtained from true labels was counted and divided by 1,000 permutation times to get the P value for classification accuracy. Finally, to investigate whether the selected features with high discriminative power were related to the symptom severity, Pearson’s partial correlation analysis was performed between the quantitative values of regions and clinical symptom severity in the ADHD group with age and gender as covariates.

Results

Demographic and clinical characteristics

In our study, 50 ADHD and 50 HCs were included. There were no significant differences in gender (43 male, 7 female in both groups) or age (8.97±2.03 vs. 8.92±2.10). Demographic and clinical information is shown in Table 1.

Classification performance

The classification performance is summarized in Table 2 and Figure 1. The SVM classifier based on combined gray and white matter quantitative features achieved better performance than every single feature, with an AUC of 0.847, sensitivity of 78.0%, specificity of 78.0% and accuracy of 78.0% (P<0.001). Moreover, based on combined features, different classifiers showed similar performance. The RF classifier achieved an AUC of 0.832, sensitivity of 78.0%, specificity of 76.0% and accuracy of 77.0% (P<0.001). The LR classifier achieved an AUC of 0.826, sensitivity of 78.0%, specificity of 78.0% and accuracy of 78.0% (P<0.001).

Figure 1 The classification performance of the machine learning models. (A) The ROC curve of SVM model based on single feature and combined features. (B) The ROC curve of different classifiers based on combined features. AUC, area under the curve; FPR, false positive rate; GM, gray matter; LR, linear regression; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machine; TPR, true positive rate; WM, white matter.

Discriminative features

The classification model combining quantitative parameters of both gray and white matter yielded a total of 14 discriminative features (9 gray matter and 5 white matter) that were selected ≥70 times across 100 iterations (Figures 2,3). Specifically, these discriminative features included: T1 values of the left medial superior frontal gyrus, left hippocampus, left fusiform gyrus, left angular gyrus, left paracentral lobule, left cerebellar VIIb, and cerebellar vermis III; T2 values of the right Heschl’s gyrus, cerebellar vermis IX, right corticospinal tract, and left cingulum; and MVF values of the genu of corpus callosum, right anterior corona radiata, and right external capsule. The detailed quantitative values of the discriminative features are extracted for both groups and reported in Table 3. Additionally, based on each single feature (gray matter or white matter alone), the discriminative features with selection frequency more than 70 times were also listed in Tables S1,S2.

Figure 2 The features with selection frequency more than 70 times in 100 iterations of LOOCV. LOOCV, leave-one-out cross-validation; MVF, myelin volume fraction.

Figure 3 The significant discriminative gray matter and white matter in the classification of ADHD vs. HCs. ADHD, attention deficit hyperactivity disorder; ANG, angular gyrus; CRBL, cerebellum; FFG, fusiform gyrus; HCs, healthy controls; HES, Heschl’s gyrus; HIP, hippocampus; L, left; PCL, paracentral lobule; R, right; SFGmed, superior frontal gyrus, medial part.

Relationship between discriminative features and symptom severity

The detailed results of the partial correlation analysis are shown in Figure 4. The T2 value of the left cingulum was positively correlated with the concentration index (r=0.285, P=0.050, uncorrected). Nevertheless, the correlation results did not reach statistical significance after false discovery rate (FDR) correction.

Figure 4 Correlation analysis results between the significant discriminative feature and clinical symptom severity in ADHD. ****, P<0.0001. ADHD, attention-deficit/hyperactivity disorder; HCs, healthy controls.

Discussion

This study combined SyMRI-derived quantitative parameters with machine learning algorithms to identify ADHD at the individual level, and to further explore ADHD-related microstructural processes. Three main findings were obtained: (I) the multiparametric quantitative values of gray and white matter derived from SyMRI demonstrated good classification performance across different machine learning algorithms; (II) the gray matter regions with high discriminative power were mainly distributed in the frontal, temporal lobes and cerebellum, and white matter tracts with high discriminative power were primarily involved the corpus callosum, frontostriatal tracts, corticospinal tract and cingulum; (III) a nominally significant correlation was found between T2 value of the left cingulum and the concentration index in ADHD, though it did not survive correction for multiple correction.

In the past few years, the application of machine learning algorithms has made substantial progress in aiding the diagnosis of individuals with ADHD (4). Studies combined clinical behavioral information, such as family and medical histories (28), Conner’s rating scale (29), and neuropsychological task performance (30) with machine learning to identify the core characteristics of ADHD. Further, neuroimaging data, as the more objective, stable and direct biomarker, were increasingly used to classify ADHD as well as investigate the brain network mechanisms underlying ADHD. Various feature patterns such as cortical morphometry of sMRI (6), diffusion properties of DTI (7), regional homogeneity, power spectra and functional connectivity of rs-fMRI (31) have been explored to construct classification model and achieved good classification of individuals of ADHD. In addition to the aforementioned efforts, the newly developed SyMRI technique seems to provide distinct perspective for such investigations, as it not only enables efficient postprocessing to obtain the quantification of actual tissue parameters which can be directly compared across subjects and even time points and imaging centers, but also has the advantage to detect early subtle impairment of the microstructural tissue integrity (10,11). Particularly, our study showed that SyMRI-derived quantitative measurements of gray and white matter combined with machine learning models were able to differentiate ADHD from HCs with good accuracy, and the performance of combined features outperformed that of single feature of gray matter or white matter. Although incorporating more variables is considered to enhance classification performance, this also indicates that multiple quantitative parameters detecting biologically distinct microstructural processes could more comprehensively characterize ADHD-related tissue changes. Thus, our findings further suggested that multi-parametric quantitative features obtained from SyMRI data combined with machine learning not only hold considerable promise in identifying diagnostic biomarkers of ADHD, but more importantly, also provide a novel way to reveal the microscopic brain changes associated with ADHD.

Further, the availability and utility of a set of quantitative parameters obtained from SyMRI data provide a novel way to investigate the pathological mechanisms underlying disease. Growing evidence have shown that these quantitative parameters could be used as an objective and reliable tool to reveal age-related changes in tissue properties in children and adults (12,32), as well as disease-related changes in tissue properties in many neurological diseases (13-15,33). Consistent with these studies, our findings also revealed significant alterations in T1, T2 relaxation time and MVF for gray and white matter in ADHD. The alterations of relaxation time were informative of a series of pathological processes. Our observation of increased T1 and T2 relaxation time in gray matter can be interpreted as the results of iron deficiency in ADHD. This is consistent with the established consensus that low iron levels can lead to an increase in tissue T1 and T2 values (34), providing a plausible neurobiological explanation for our findings. Furthermore, previous quantitative susceptibility mapping studies have consistently demonstrated reduced brain iron levels in children with ADHD (35,36), lending support to our hypothesis that increased relaxation time of gray matter was linked to brain iron dyshomeostasis in ADHD. In addition to the changes in gray matter, our study also found MVF and T2 values changes in white matter. Although increased MVF in white matter was commonly considered as an increase in myelin content, considering increased T2 values of white matter in our study, these results may indicate more complex microstructural alterations in ADHD. Microstructural changes such as axonal damage and demyelination can lead to prolongation of white matter T2 values by increasing the proportion of extracellular water (34), while an abnormal increase of MVF can be explained by the compensatory myelination. Thus, the heterogenous change patterns in different regions of white matter further indicated that the microstructural changes were inhomogeneously distributed across the brain.

Besides, growing evidence from macrostructural studies has posited delayed brain development theory in ADHD, with the most consistent findings being decreased gray matter volume in the frontal and temporal lobe structures and cerebellum (37). Similarly, the most discriminative gray matter regions with relaxation time alterations in our study involved the medial superior frontal gyrus, hippocampus, fusiform gyrus, Heschl’s gyrus, cerebellar VIIb, cerebellar vermis III, cerebellar vermis IX. Considering these brain regions with microscopic changes were spatially consistent with those previously reported with macrostructural volume changes, it is plausible to hypothesize that the microstructural process reflected by relaxometry values may underlie the tissue volume reduction. Moreover, the discriminative gray matter regions obtained in this study constituted different brain systems. Among them, the medial superior frontal gyrus, hippocampus, and angular gyrus were closely linked to the default mode network (DMN), with a large overlap to the medial prefrontal cortex, temporal cortex and inferior parietal lobe of the DMN (38). The proposed delayed DMN neuromaturation were thought to lead to lapses in attention, distractibility, and increased variability in task performance in ADHD (39). As part of the visual cortex, the fusiform gyrus was involved in high-level visual function, including the perception of emotions in facial stimuli, objective recognition and reading (40). The abnormality of this region may account for the deficits in visual-spatial memory and reading ability associated with ADHD. Besides, our study also extended to the cerebellum which engaged not only in the motor domain but also in cognitive and emotional processes (41). The importance of the cerebellum in ADHD was supported by the fact that the abnormalities of the cerebellum could be normalized by psychostimulant methylphenidate treatments in children with ADHD (42). In line with this, our study reinforced the role of disrupted cerebellar development in the ADHD pathophysiology.

With regard to microstructural abnormalities, our study presented disturbed white matter integrity in several white matter tracts, and these tracts can be grouped into commissural pathways (the corpus callosum) which transfer information between the hemispheres of the brain; projection pathways (frontostriatal tracts and corticospinal tract) which transmit dopaminergic and sensory-motor information; and association pathways (the cingulum) which integrate function of regions within the same hemisphere. The identified white matter changes in our study were largely consistent with current meta-analysis which showed the most prominent findings in the frontostriatal tracts, corpus callosum, superior longitudinal fasciculus, cingulum bundle, thalamic radiations, internal capsule and corona radiata (43). Furthermore, our findings of the prolonged T2 value of the cingulum in ADHD were correlated with inattention severity. The cingulum is a collection of white matter tracts that anatomically connects the components of the DMN, such as the frontal lobe with the precuneus, posterior cingulate cortex, hippocampus and parahippocampus (44). As one of the most prominent networks, DMN is a system of brain regions that is activated at rest but deactivated while performing attention-demanding tasks (45). Notably, ADHD was increasingly considered as DMN disorder in which attention deficits resulted from the disturbed activation of DMN as well as between DMN and other task-related networks (39). Thus, the impaired white matter microstructure in the cingulum might interrupt the DMN activation, and this might be a pathological component of the attention dysfunction in ADHD.

Limitations

Some limitations in this study need to be noted. First, the individuals with ADHD were recruited form the same site with a relatively small sample size. Though we evaluated the model’s performance by using nested cross-validation strategy, the generalization ability of the classification model needs to be replicated on a larger cohort from multiple sites and validated with independent test data. Second, it was worth noting that the results of correlation analysis did not survive FDR correction; thus future studies with a larger sample size may be more able to increase the statistical power. Third, the male-to-female ratio (nearly 6:1) in our study was imbalanced, although the reported prevalence of ADHD was higher in boys; further study with more matching ratio can reduce the potential effect of confounding factors. Fourth, model fitting based on both SyMRI and conventional MRI from the same participants could offer more comprehensive and complementary information related to ADHD pathology. A rigorous and systematic investigation in the future can be conducted to explore their comparative and combined efficacy. Finally, we only performed binary classification between ADHD and HCs; additional classification between subtypes of ADHD will be helpful to further elucidate the unique mechanism underlying different clinical phenotypes.

Conclusions

In this study, we built the machine learning model to uncover the potential diagnostic biomarker of ADHD based on SyMRI data and achieved good accuracy in distinguishing ADHD from HCs. The distributed patterns of gray and white matter changes not only pose significant discriminative features but also provide novel and comprehensive information to investigate the neural mechanisms underlying ADHD.

Acknowledgments

We gratefully thank the participants and their families, as well as the staff of the MRI center of The First Affiliated Hospital of Sun Yat-sen University, for their help and support.

Footnote

Reporting Checklist: The authors have completed the TRIPOD+AI reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1266/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1266/dss

Funding: This research was supported by the Natural Science Fund Project of Guangdong Province (Nos. 2024A1515012561 and 2022A1515011910) and the National Natural Science Foundation of China (No. 82001439).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1266/coif). B.X. is an employee of MR Research, GE Healthcare. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the ethics committee of The First Affiliated Hospital of Sun Yat-sen University (No. [2019]328). Informed consent was obtained from the legal guardians of all participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Rajaprakash M, Leppert ML. Attention-Deficit/Hyperactivity Disorder. Pediatr Rev 2022;43:135-47. [Crossref] [PubMed]
2. Thapar A, Cooper M. Attention deficit hyperactivity disorder. Lancet 2016;387:1240-50. [Crossref] [PubMed]
3. Firouzabadi FD, Ramezanpour S, Firouzabadi MD, Yousem IJ, Puts NAJ, Yousem DM. Neuroimaging in Attention-Deficit/Hyperactivity Disorder: Recent Advances. AJR Am J Roentgenol 2022;218:321-32. [Crossref] [PubMed]
4. Cao M, Martin E, Li X. Machine learning in attention-deficit/hyperactivity disorder: new approaches toward understanding the neural mechanisms. Transl Psychiatry 2023;13:236. [Crossref] [PubMed]
5. Nielsen AN, Barch DM, Petersen SE, Schlaggar BL, Greene DJ. Machine Learning With Neuroimaging: Evaluating Its Applications in Psychiatry. Biol Psychiatry Cogn Neurosci Neuroimaging 2020;5:791-8. [Crossref] [PubMed]
6. Sun H, Chen Y, Huang Q, Lui S, Huang X, Shi Y, Xu X, Sweeney JA, Gong Q. Psychoradiologic Utility of MR Imaging for Diagnosis of Attention Deficit Hyperactivity Disorder: A Radiomics Analysis. Radiology 2018;287:620-30. [Crossref] [PubMed]
7. Zhou X, Lin Q, Gui Y, Wang Z, Liu M, Lu H, Multimodal MR. Images-Based Diagnosis of Early Adolescent Attention-Deficit/Hyperactivity Disorder Using Multiple Kernel Learning. Front Neurosci 2021;15:710133. [Crossref] [PubMed]
8. Iannaccone R, Hauser TU, Ball J, Brandeis D, Walitza S, Brem S. Classifying adolescent attention-deficit/hyperactivity disorder (ADHD) based on functional and structural imaging. Eur Child Adolesc Psychiatry 2015;24:1279-89. [Crossref] [PubMed]
9. Sun Y, Zhao L, Lan Z, Jia XZ, Xue SW. Differentiating Boys with ADHD from Those with Typical Development Based on Whole-Brain Functional Connections Using a Machine Learning Approach. Neuropsychiatr Dis Treat 2020;16:691-702. [Crossref] [PubMed]
10. Hagiwara A, Warntjes M, Hori M, Andica C, Nakazawa M, Kumamaru KK, Abe O, Aoki S. SyMRI of the Brain: Rapid Quantification of Relaxation Rates and Proton Density, With Synthetic MRI, Automatic Brain Segmentation, and Myelin Measurement. Invest Radiol 2017;52:647-57. [Crossref] [PubMed]
11. Andica C, Hagiwara A, Hori M, Kamagata K, Koshino S, Maekawa T, Suzuki M, Fujiwara H, Ikeno M, Shimizu T, Suzuki H, Sugano H, Arai H, Aoki S. Review of synthetic MRI in pediatric brains: Basic principle of MR quantification, its features, clinical applications, and limitations. J Neuroradiol 2019;46:268-75. [Crossref] [PubMed]
12. Lee SM, Choi YH, You SK, Lee WK, Kim WH, Kim HJ, Lee SY, Cheon H. Age-Related Changes in Tissue Value Properties in Children: Simultaneous Quantification of Relaxation Times and Proton Density Using Synthetic Magnetic Resonance Imaging. Invest Radiol 2018;53:236-45. [Crossref] [PubMed]
13. Deoni SC, Zinkstok JR, Daly E, Ecker CMRC AIMS Consortium. Williams SC, Murphy DG. White-matter relaxation time and myelin water fraction differences in young adults with autism. Psychol Med 2015;45:795-805. [Crossref] [PubMed]
14. Ladopoulos T, Matusche B, Bellenberg B, Heuser F, Gold R, Lukas C, Schneider R. Relaxometry and brain myelin quantification with synthetic MRI in MS subtypes and their associations with spinal cord atrophy. Neuroimage Clin 2022;36:103166. [Crossref] [PubMed]
15. Parlak S, Coban G, Gumeler E, Karakaya J, Soylemezoglu F, Tezer I, Bilginer B, Saygi S, Oguz KK. Reduced myelin in patients with isolated hippocampal sclerosis as assessed by SyMRI. Neuroradiology 2022;64:99-107. [Crossref] [PubMed]
16. American Psychiatric Association. Diagnostic and statistical manual of mental disorders: DSM-5. 5th ed. Washington, D.C.: American Psychiatric Publishing, 2013.
17. Conners CK, Pitkanen J, Rzepa SR. Conners 3rd Edition (Conners 3; Conners 2008). In: Kreutzer JS, DeLuca J, Caplan B, editors. Encyclopedia of Clinical Neuropsychology. New York, NY: Springer New York; 2011:675-8.
18. Xie Y, Wang H, Chen Y, Liu F, Yao M, Zhang L, Liu P, Hong Q, Chi X, Yu D. Establishing Normative Data for the Number Cancelation Test Among Children in Kindergartens and Primary Schools in China. Front Psychiatry 2022;13:788825. [Crossref] [PubMed]
19. De la Torre GG, Barroso JM, León-Carrión J, Mestre JM, Bozal RG. Reaction Time and Attention: Toward a New Standard in the Assessment of ADHD? A Pilot Study. J Atten Disord 2015;19:1074-82. [Crossref] [PubMed]
20. Liu L, Cheng J, Su Y, Ji N, Gao Q, Li H, Yang L, Sun L, Qian Q, Wang Y. Deficiency of Sustained Attention in ADHD and Its Potential Genetic Contributor MAOA. J Atten Disord 2018;22:878-85. [Crossref] [PubMed]
21. Warntjes M, Engström M, Tisell A, Lundberg P. Modeling the Presence of Myelin and Edema in the Brain Based on Multi-Parametric Quantitative MRI. Front Neurol 2016;7:16. [Crossref] [PubMed]
22. Lin L, Cheng Y, Qiu H, Yan Z, Hou W, Huang S, Cui W, Cheung GL, Yang Z, Chen Q, Qian L, Su S. Evaluation of gray-matter and white-matter microstructural abnormalities in children with growth hormone deficiency: a comprehensive assessment with synthetic magnetic resonance imaging. Quant Imaging Med Surg 2025;15:314-25. [Crossref] [PubMed]
23. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, Mazoyer B, Joliot M. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 2002;15:273-89. [Crossref] [PubMed]
24. Mori S, Oishi K, Jiang H, Jiang L, Li X, Akhter K, Hua K, Faria AV, Mahmood A, Woods R, Toga AW, Pike GB, Neto PR, Evans A, Zhang J, Huang H, Miller MI, van Zijl P, Mazziotta J. Stereotaxic white matter atlas based on diffusion tensor imaging in an ICBM template. Neuroimage 2008;40:570-82. [Crossref] [PubMed]
25. Zhu L, Yang J, Song JN, Chou KC, Shen HB. Improving the accuracy of predicting disulfide connectivity by feature selection. J Comput Chem 2010;31:1478-85. [Crossref] [PubMed]
26. Zhang X, Chen H, Zhang X, Wang H, Tao L, He W, Li Q, Cheng O, Luo J, Man Y, Xiao Z, Fang W. Identification of essential tremor based on resting-state functional connectivity. Hum Brain Mapp 2023;44:1407-16. [Crossref] [PubMed]
27. Abraham A, Pedregosa F, Eickenberg M, Gervais P, Mueller A, Kossaifi J, Gramfort A, Thirion B, Varoquaux G. Machine learning for neuroimaging with scikit-learn. Front Neuroinform 2014;8:14. [Crossref] [PubMed]
28. Garcia-Argibay M, Zhang-James Y, Cortese S, Lichtenstein P, Larsson H, Faraone SV. Predicting childhood and adolescent attention-deficit/hyperactivity disorder onset: a nationwide deep learning approach. Mol Psychiatry 2023;28:1232-9. [Crossref] [PubMed]
29. Cheng CY, Tseng WL, Chang CF, Chang CH, Gau SS. A Deep Learning Approach for Missing Data Imputation of Rating Scales Assessing Attention-Deficit Hyperactivity Disorder. Front Psychiatry 2020;11:673. [Crossref] [PubMed]
30. Fair DA, Bathula D, Nikolas MA, Nigg JT. Distinct neuropsychological subgroups in typically developing youth inform heterogeneity in children with ADHD. Proc Natl Acad Sci U S A 2012;109:6769-74. [Crossref] [PubMed]
31. Colby JB, Rudie JD, Brown JA, Douglas PK, Cohen MS, Shehzad Z. Insights into multimodal imaging classification of ADHD. Front Syst Neurosci 2012;6:59. [Crossref] [PubMed]
32. Hagiwara A, Fujimoto K, Kamagata K, Murata S, Irie R, Kaga H, Someya Y, Andica C, Fujita S, Kato S, Fukunaga I, Wada A, Hori M, Tamura Y, Kawamori R, Watada H, Aoki S. Age-Related Changes in Relaxation Times, Proton Density, Myelin, and Tissue Volumes in Adult Brain Analyzed by 2-Dimensional Quantitative Synthetic Magnetic Resonance Imaging. Invest Radiol 2021;56:163-72. [Crossref] [PubMed]
33. van Wijnen A, Petrov F, Maiworm M, Frisch S, Foerch C, Hattingen E, Steinmetz H, Klein JC, Deichmann R, Wagner M, Gracien RM. Cortical quantitative MRI parameters are related to the cognitive status in patients with relapsing-remitting multiple sclerosis. Eur Radiol 2020;30:1045-53. [Crossref] [PubMed]
34. Seiler A, Nöth U, Hok P, Reiländer A, Maiworm M, Baudrexel S, Meuth S, Rosenow F, Steinmetz H, Wagner M, Hattingen E, Deichmann R, Gracien RM. Multiparametric Quantitative MRI in Neurological Diseases. Front Neurol 2021;12:640239. [Crossref] [PubMed]
35. Chen Y, Su S, Dai Y, Zou M, Lin L, Qian L, Zhou Q, Zhang H, Liu M, Zhao J, Yang Z. Quantitative susceptibility mapping reveals brain iron deficiency in children with attention-deficit/hyperactivity disorder: a whole-brain analysis. Eur Radiol 2022;32:3726-33. [Crossref] [PubMed]
36. Morandini HAE, Watson PA, Barbaro P, Rao P. Brain iron concentration in childhood ADHD: A systematic review of neuroimaging studies. J Psychiatr Res 2024;173:200-9. [Crossref] [PubMed]
37. Kasparek T, Theiner P, Filova A. Neurobiology of ADHD From Childhood to Adulthood: Findings of Imaging Methods. J Atten Disord 2015;19:931-43. [Crossref] [PubMed]
38. Buckner RL, Andrews-Hanna JR, Schacter DL. The brain's default network: anatomy, function, and relevance to disease. Ann N Y Acad Sci 2008;1124:1-38. [Crossref] [PubMed]
39. Sripada CS, Kessler D, Angstadt M. Lag in maturation of the brain's intrinsic functional architecture in attention-deficit/hyperactivity disorder. Proc Natl Acad Sci U S A 2014;111:14259-64. [Crossref] [PubMed]
40. Weiner KS, Zilles K. The anatomical and functional specialization of the fusiform gyrus. Neuropsychologia 2016;83:48-62. [Crossref] [PubMed]
41. Stoodley CJ, Schmahmann JD. Evidence for topographic organization in the cerebellum of motor control versus cognitive and affective processing. Cortex 2010;46:831-44. [Crossref] [PubMed]
42. Rubia K, Halari R, Cubillo A, Mohammad AM, Brammer M, Taylor E. Methylphenidate normalises activation and functional connectivity deficits in attention and motivation networks in medication-naïve children with ADHD during a rewarded continuous performance task. Neuropharmacology 2009;57:640-52. [Crossref] [PubMed]
43. Connaughton M, Whelan R, O'Hanlon E, McGrath J. White matter microstructure in children and adolescents with ADHD. Neuroimage Clin 2022;33:102957. [Crossref] [PubMed]
44. van den Heuvel M, Mandl R, Luigjes J, Hulshoff Pol H. Microstructural organization of the cingulum tract and the level of default mode functional connectivity. J Neurosci 2008;28:10844-51. [Crossref] [PubMed]
45. Raichle ME. The brain's default mode network. Annu Rev Neurosci 2015;38:433-47. [Crossref] [PubMed]