This study aimed to examine the structural brain differences between children with ADHD and children with TD by addressing the measurement bias in MRI data from different scanners using the TS method. This study demonstrates that the TS harmonization method can improve the reliability of structural brain data across multiple MRI scanners by applying the TS method to an independent dataset. This study also examined the differences between ADHD and TD to elucidate the brain structural characteristics of ADHD by comparing different harmonization methods and raw data. The results of TS-corrected data showed volumetric differences between the ADHD and TD groups in the cortical regions, including the bilateral middle temporal and bilateral orbitofrontal cortex, right inferior frontal gyrus, right middle frontal gyrus, left inferior temporal gyrus, left precuneus cortex, and bilateral insular cortex. In contrast, comparisons between ADHD and TD using ComBat-corrected and raw data showed different findings than when using TS-corrected data.
Before harmonization, the current study provided evidence that half of the brain regions showed significant differences among the same participants, demonstrating the necessity of harmonization for measurement bias in MRI scanners. The TS method, which estimates measurement bias more accurately [17], significantly reduces measurement bias compared with the raw data. This method involves considerable logistical effort, including TS recruitment and scan scheduling within a short duration; however, it corrects for measurement bias. The TS method addresses the variations introduced by different scanners and scanning protocols [17]. This is crucial for multi-site studies, wherein such variations can compromise the credibility of research findings. Although a previous study proved the validity of TS in structural brain data [18], few studies have applied TS to correct for measurement bias in an independent dataset. One of the contributions of this study is the validation of the TS method in structural brain data from an independent dataset, providing evidence that this method can be effectively used to harmonize multi-site MRI bias in structural brain data.
This study underscores the importance of harmonization methods in multi-site neuroimaging studies. Site differences consist of two types of bias: measurement and sampling. Measurement bias includes differences in the properties of MRI scanners, such as imaging variables, field strength, MRI manufacturers, and scanner models, whereas sampling bias refers to differences in participant groups between sites. The measurement bias can influence the accuracy of results in multi-site studies. Sampling bias can be considered as the biological part of site difference because of sampling from different subpopulations [17], which makes it possible to include the biological part of disorders. therefore it is inevitable that the data will contain sampling bias. Since the sampling bias included the biological part of the site difference, the reduction of sampling bias may influence the analysis of the difference in brain structures between ADHD and TD groups. The current study calculated the sampling bias is consistent to the previous study, which proved the existence of sampling bias in our dataset, and ComBat method could reduce it as measurement bias. Although ComBat can provide more power to reduce the measurement bias, the overcorrection of sampling bias may influence the accuracy of findings.
The results indicated that TS-corrected data exhibited a smaller measurement bias compared with the raw data, but a larger measurement bias than ComBat-corrected data. This suggests that while the TS method is effective in reducing some measurement biases inherent in multi-site data collection, the ComBat method achieves a greater reduction in measurement bias. This result is consistent with that of a previous study that proved the effectiveness of robust normalization methods for MRI data [16]. Our analysis also showed that ComBat-corrected data had a smaller sampling bias than the raw and TS-corrected data.
These results emphasize the importance of selecting appropriate correction methods for multi-site studies. While TS methods offer improvements over raw data and do not influence the sampling bias of different sites, the ComBat method’s rigorous statistical framework provides superior performance in reducing measurement bias while also reducing sampling biases such as measurement bias. Considering that reducing the sampling bias in patients influences the findings of the neural basis of a disorder [17] and that sampling bias does exist in raw data, harmonization using the TS method may be more appropriate for the current analysis.
Analysis using TS-corrected data revealed significant volumetric differences between the ADHD and TD groups in several brain regions. Notably, patients with ADHD exhibited smaller volumes in the bilateral middle temporal gyrus, bilateral orbitofrontal cortex, right inferior frontal gyrus, right middle frontal gyrus, left inferior temporal gyrus, left precuneus, and bilateral insular cortex. Reductions in the bilateral middle temporal gyrus and orbitofrontal cortex have been consistently reported. These regions are crucial for cognitive functions such as information processing, attention regulation, and emotional control, which are often impaired in ADHD[22, 23]. Even when comparing the ADHD group to a matched TD group, differences in the middle temporal gyrus remained significant. Moreover, the middle temporal gyrus was smaller in children with ADHD, even when using ComBat-corrected and raw data. This highlights the robustness of this finding across different correction methods. Castellanos and Aoki’s [24] suggestion that ADHD is a disorder of the default mode network (DMN) and Cubillo et al.’s [25] findings of frontostriatal dysfunction align with the current study’s results, particularly the consistent finding of abnormalities in the middle temporal gyrus [12]. The middle temporal gyrus is a critical node within the DMN, a network of brain regions that is typically more active at rest and less active during goal-directed tasks [26]. The middle temporal gyrus’s connectivity with these regions supports various DMN functions. The right middle temporal gyrus showed significant differences in all methods, emphasizing its potential role as a biomarker for ADHD. However, significant differences in brain volume were found in several brain regions among the TS-corrected, ComBat-corrected, and raw data. The TS method can reduce the measurement bias derived from multiple MRI scanners, and the results of the raw data may be influenced by this bias. Additionally, as the comparison of TS-corrected data with ComBat-corrected data revealed some differences in results, we considered that the excessive reduction of sampling bias when using ComBat may influence the accuracy of findings. Previous studies that did not use the TS method could not accurately correct these biases, which may have led to the discrepancies in the results.
In addition to the brain regions mentioned above, the right inferior frontal gyrus and the right middle frontal gyrus also show reduced volumes in patients with ADHD. These frontal regions are associated with executive functions, including inhibitory control and working memory, and their reduced volume may contribute to the characteristic impulsivity and inattention observed in ADHD [27]. Furthermore, studies have reported volume reductions in the left inferior temporal gyrus and left precuneus cortex in individuals with ADHD. The inferior temporal gyrus is involved in visual processing and object recognition, whereas the precuneus is involved in self-referential thinking and visuospatial processing. These structural differences may underlie some cognitive and perceptual challenges faced by patients with ADHD [28]. Finally, the bilateral insular cortex, which is involved in interoceptive awareness and emotional regulation, has also been shown to have a reduced volume in individuals with ADHD. The role of the insula in integrating emotional and cognitive processes suggests that its structural abnormalities could be linked to the emotional dysregulation frequently observed in ADHD[10]. Overall, these findings highlight a widespread pattern of cortical volume reduction in ADHD, which encompasses the regions involved in attention, executive function, emotional regulation, and cognitive processing. This broad network of structural abnormalities underscores the complexity of ADHD and the need for a comprehensive approach to its study and treatment. The observed structural differences in these brain regions align with previous neuroimaging studies, implicating alterations in the neural circuitry associated with attention, executive function, emotional regulation, and cognitive control in individuals with ADHD. However, these regions showed no significant differences between the ADHD and TD groups after comparing the ADHD group with a matched TD group, implying that these regions may be influenced by age, sex, and handedness.
Despite these promising results, this study had some limitations. The study sample may not fully represent the broader population of children with ADHD. The participants were drawn from specific geographical regions and clinical settings, which could limit the generalizability of the findings to other populations. Additionally, this study only examined the brain structure characteristics in children with ADHD elucidated using harmonization. For functional brain, Castellanos et al. (2016) [24]reported that children with ADHD have abnormalities in the default mode network and network of brain regions involved in the reward system, while a recent meta-analysis study of resting-state functional MRI reported that no findings specific to ADHD could be obtained [29]. Future studies should consider examining the functional brain characteristics in children with ADHD by using the TS method.
In conclusion, this study demonstrated the effectiveness of the TS method in correcting measurement bias in multi-site MRI studies involving children with ADHD. These findings highlight significant structural differences in the brains of patients with ADHD, particularly in the middle temporal gyrus, and underscore the importance of using robust harmonization techniques to improve the reproducibility and accuracy of neuroimaging research.