Linking electric field directionality to treatment outcome in OCD: Insights from patient-specific tDCS modeling

Abstract

Transcranial direct current stimulation (tDCS) is a promising intervention for treatment-resistant obsessive-compulsive disorder (OCD), yet clinical outcomes remain inconsistent. To investigate the neural mechanisms underlying therapeutic variability, we conducted a patient-specific finite element (FE) modeling study of electric fields (EF) induced by tDCS in OCD patients. Forty-two patients from a double-blind, randomized clinical trial received active tDCS with the cathode over the pre-supplementary motor area and the anode over the right supraorbital region. Clinical response was assessed using the Yale-Brown Obsessive-Compulsive Scale (Y-BOCS), and responders were defined as those achieving ≥35% score reduction. Individual head models were created using SimNIBS, and current density directionality ($J_n$) and magnitude ($J_{\mathit{magn}}$) were analyzed. Voxel-wise comparisons revealed significantly greater depolarization ($J_n$> 0) in the left anterior prefrontal cortex (BA10) and right frontal eye field (BA 8) associated with a reduction of Y-BOCS. A link between hyperpolarization of right pars orbitalis (BA47) and improvement in symptoms was also found. Notably, no significant findings emerged using EF magnitude ($J_{\mathit{magn}}$), underscoring the relevance of current directionality in treatment response. To our knowledge, this is the first study to associate directional EF modeling with clinical outcomes in OCD. Our findings highlight the importance of considering both EF direction and anatomical variability when optimizing tDCS protocols. This approach may contribute to more personalized and effective neuromodulation strategies for psychiatric disorders.

Introduction

Transcranial direct current stimulation (tDCS) is a non-invasive technique that modulates cortical excitability by delivering a low-intensity direct current through the scalp. tDCS research has mainly focused on major depressive disorder (MDD) [1–4], with a recent FDA approval for moderate to severe major depressive disorder in patients who are not treatment-refractory to medication [5], based on the Empower randomized controlled trial (RCT) [6], despite effects similar to sham stimulation in larger patient cohorts [7]. Its application in other conditions, such as obsessive-compulsive disorder (OCD), remains an emerging area of research. The evidence supporting tDCS for OCD is not yet as robust, with optimal clinical indications and stimulation parameters still under investigation [8], especially for treatment-resistant OCD, which affects 30% to 40% of patients [9].

Several targets have been studied, such as the orbitofrontal cortex [10–18], the cerebellum and occipital cortex [10–12, 19–22], the supplementary motor area (SMA) and pre-SMA [10, 14–17, 22–24], or the DLPFC [18–20]. However, no consensus has been reached. To address this issue, our team recently conducted an open-label study and the largest randomized controlled trial in this field. In the RCT, we evaluated the therapeutic effect of 10 tDCS sessions in treatment-resistant OCD patients using a multisite, double-blind, sham-controlled design. Eighty treatment-resistant patients were randomized to receive either active or sham transcranial direct current stimulation. The cathode was positioned over the pre-supplementary motor area and the anode over the right supraorbital area. Patients were evaluated at baseline, at the end of treatment (day 14), at 1-month follow-up (day 45), and at 3-month follow-up (day 105) using the Yale-Brown Obsessive-Compulsive Scale (Y-BOCS). The results demonstrated a significant decrease in Y-BOCS scores over time, regardless of treatment. However, only five responders were identified at the 3-month follow-up among the 22 subjects included in the pilot study. To better understand this limited response, we explored the distribution of the electric field (EF) in patients who received active treatment and to compare this distribution between responders and non-responders. In this context, studying EFs may help increase the effects of tDCS in OCD and reduce variability across studies. Indeed, inter-individual differences in brain and head anatomy can affect current flow from one subject to another.

The EF, defined as the electric force per unit charge, is generated in the brain when direct current is applied to the scalp via electrodes. In the early 2010s, simulation of the EF during tDCS using realistic MRI-derived head models began [25]. There is considerable variation in the electric field produced by tDCS [26–28] due to individual structural and functional neuroanatomical differences [29–31]. Several numerical tools based on the finite element (FE) method have been developed to study variations in the EF, such as SimNIBS [32], ROAST [33] or Ansys PyBrainLib [34]. The goal of these workflows is to create patient-specific head models to analyze the EF during tDCS, primarily for prospective analyses and to better understand the response to neuromodulation. The ultimate aim is to predict the EF in advance and optimize electrode positioning for individualized targeting [35, 36]. The use of EF modeling to predict the effects of tDCS in psychiatric disorders has become a focal point of research, driven by the need to understand and enhance the clinical efficacy of tDCS. For example, one study reported low clinical efficacy of tDCS in a patient with schizophrenia who had a left frontal lesion, which was associated with a low simulated electric field [37]. Another study explored differences between responders and non-responders to frontotemporal tDCS using electric field modeling in 17 patients with schizophrenia [38]. That study indicated that the strength of the tDCS-induced electric field may significantly influence response to tDCS. Understanding the electric field distribution in individual patients could therefore be crucial for optimizing tDCS protocols, especially in psychiatric populations in which brain lesions or structural abnormalities are prevalent.

While EF magnitude has often been used to correlate with clinical response, no study has used current direction as a response marker, despite previous works showing that current direction is an important factor in tDCS efficacy [39, 40]. Current directionality can be used as a proxy for the injected current along the somatodendritic axis of neurons by projecting the EF vector onto the normal to the cortical surface. This assumption is based on the idea that neuronal dendrites are aligned with the cortical surface [41].

To date, no study has used individualized FE modeling to analyze tDCS response in a clinical trial of OCD. A few studies have explored patient-specific electric field modeling to explain differences in therapeutic response [38, 42, 43]. In OCD, only two studies have evaluated electric field distribution by comparing different electrode configurations [44, 45], but none has focused on patient-specific field distribution in relation to therapeutic response.

In this study, we present the first analysis of tDCS treatment in OCD patients using patient-specific finite element modeling.

Methods

Clinical procedures

MRI and patient data were collected from two clinical trials conducted at the Henri Laborit Psychiatric Hospital (Poitiers, France). Both trials were approved by institutional review boards and registered on ClinicalTrials.gov (NCT03304600 and NCT02384671). The studies followed similar methodologies in terms of stimulation parameters, inclusion/exclusion criteria, and duration. However, the second study was designed as a randomized, double-blind, controlled trial.

A total of 70 patients were retrospectively selected, but only 42 were retained after successful MRI segmentation, completion of all assessments, and inclusion in the active treatment group of the second study. Eligible patients (aged 18–70 years) met DSM-IV-TR criteria for obsessive-compulsive disorder (OCD), had a Yale-Brown Obsessive-Compulsive Scale (Y-BOCS) score ≥ 21, a disease duration of at least 2 years, and documented failure to respond to at least 12 weeks of serotonin reuptake inhibitors (SRIs) and/or cognitive-behavioral therapy (CBT). Exclusion criteria included severe psychiatric comorbidities, unstable medical conditions, or metallic implants.

Clinical assessments, including Y-BOCS, the Montgomery-Åsberg Depression Rating Scale (MADRS), and the Hospital Anxiety and Depression Scale (HADS), were conducted at baseline, day 14, and day 45 by trained psychiatrists. Medication regimens were maintained throughout the study. Responders were defined as patients showing a ≥ 35% reduction in Y-BOCS scores. Additional details about these studies have been published elsewhere [16, 17].

Structural MRI scans were acquired prior to the studies using a Philips Intera system with a 1.5 T magnetic field strength, an echo time of 4.2 ms, and a repetition time of 25 ms. The 2-week clinical trial used transcranial direct current stimulation (tDCS) delivered once daily with an HDCKit from NEWRONIKA S.r.l. (Via Dante, 4-20121 Milano, Italy, CE0068). Two electrodes, each with a surface area of 35 cm², were placed in saline-soaked sponges, and a current of 2 mA was delivered for 30 min. The anode was positioned above the right eyebrow over the orbitofrontal cortex, and the cathode was placed over the pre-supplementary motor area (pre-SMA). Patients were instructed to relax and remain awake.

Computational modelling

DICOM images were converted to NIfTI format with 1 mm isotropic spacing and segmented using the headreco procedure from SimNIBS software to reconstruct the white matter, gray matter, cerebrospinal fluid (CSF), skull, and scalp anatomy of each subject. Each segmentation was manually reviewed to identify and correct potential mismatched geometries. Electrode placement followed clinical practice guidelines and was performed using the SimNIBS MATLAB interface. The anode was positioned 1 cm above the Fp2 EEG landmark, based on the 10–10 system, using the Jurcak positioning method [46]. The cathode was placed at FCz according to the same reference system.

After generation of all models, electrode positioning was verified and adjusted if necessary, using the SimNIBS graphical user interface (GUI). Subsequently, we computed the current density vector ($\vec{J})$, and its magnitude ($J_{\mathit{magn}}$) using the SimNIBS solver, which relies on GetDP [47] to generate the FE model linear system and PETSc [48], to solve it with an iterative preconditioned conjugate gradient method [32]. The vector $\vec{J}$ was then projected onto the middle gray matter surface across the whole brain using a scalar product between $\vec{J}$ and the cortical surface vector ($n)$ obtained from the segmentation process:

$$J_n=J.\vec{n}$$ 1

The $J_n$ mappings represented the current injection into the somatodendritic axis of neurons [39]. Specifically, $J_n$ < 0 indicated hyperpolarization, while $J_n$ > 0 indicated depolarization of neurons.

The $J_n$ mappings were converted into a tetrahedral mesh using PyVista [49]. To mimic a 3D MRI, a hexahedral mesh with 1 mm isotropic spacing was created. The $J_n$ map was then interpolated onto this mesh using the PyVista “sample” method, resulting in a patient-specific MRI-like visualization of polarization levels for each patient (Fig. 1). For the magnitude $J_{\mathit{magn}}$, SimNIBS automatically converted results into an MRI-like image.

Fig. 1. Finite-element simulations of current density distribution across two subjects.

a Mapping of current direction relative to the somatodendritic orientation ($J_n$); b Interpolation of the $J_n$ mapping into a hexahedral mesh; c. MRI-like NIfTI file of $J_n$.

Statistical analysis

First, baseline demographic and clinical variables were compared between responders and non-responders. Sex, age, age of onset, age at first treatment, disease duration, baseline MADRS score, HADS total score, and baseline Y-BOCS score were compared using Student’s t-tests with Bayes factor computation. Next, a repeated-measures ANOVA was performed on Y-BOCS scores at baseline, day 14, and day 45.

Two predictive clinical analyses were performed. A linear regression and a binomial regression were used to assess whether clinical variables predicted clinical response. All demographic and clinical variables were included in the models to predict Y-BOCS evolution in the linear regression and responder status in the binomial regression. Maximum current density $J_{\max }$ and current focality, defined as the gray matter volume in which $J_i\ge 0.75J_{\max }$ were then added to both the linear and binomial models to determine whether models including FE metrics performed better.

Inspired by a previous study in schizophrenia [38], voxel-wise analyses were performed on both $J_n$ and $J_{\mathit{magn}}$ maps. First, the maps were smoothed in SPM12 [50] using a 3 mm full-width at half-maximum Gaussian kernel. Subsequently, displacement fields from T1-weighted MRI space to MNI space were obtained and applied to the smoothed maps to obtain normalized $J_n$ and $J_{\mathit{magn}}$ maps. A voxel-wise second-level analysis was then conducted to compare these maps between responders and non-responders and to perform a multiple regression analysis between Y-BOCS score evolution and polarization levels. Y-BOCS evolution was computed as:

$${\mathit{YBOCS}}_{\mathit{evolution}}=\frac{{\mathit{YBOCS}}_{D45}-{\mathit{YBOCS}}_{\mathit{BASELINE}}}{{\mathit{YBOCS}}_{\mathit{BASELINE}}}$$ 2

Because baseline MADRS scores differed between responders and non-responders, an additional voxel-wise multiple regression analysis was performed to test the association between baseline MADRS scores and polarization levels. Age, sex, and baseline MADRS were included as covariates in the Y-BOCS analyses, whereas age and sex were included as covariates in the MADRS analyses. Two non-responders had missing baseline MADRS scores; for voxel-wise analyses, these values were replaced by the mean baseline MADRS score of the non-responder group.

A mask corresponding to 50% of the mean maximum current density value observed across patients was applied. For statistical testing, a conservative approach based on the reference study [38] was used, with an uncorrected voxel-level threshold of p = 0.001 and an uncorrected cluster-level threshold of p = 0.05. Each cluster was anatomically localized according to its peak coordinates using Brodmann areas in BioImage Suite Web [51]. Post hoc tests were conducted for each cluster at peak level, and false discovery rate (FDR) values were also reported. The same analyses were performed on $J_{\mathit{magn}}$ mappings.

Results

Clinical results

As shown in Table 1, a significant difference in MADRS scores was found between responders and non-responders, with a Bayes factor of 3.6, suggesting anecdotal evidence for a significant result [52]. No other differences were found between responders and non-responders. Additionally, repeated-measures ANOVA of the main effect of time on the Y-BOCS evolution indicated a significant decrease in Y-BOCS scores between baseline and 14 days, between baseline and 45 days, and between 14 days and 45 days, suggesting that tDCS was effective in reducing OCD symptoms. Because MADRS scores were significantly different between responders and non-responders, it was included as a covariate in the following statistical analyses. Two non-responders had missing MADRS scores. For the voxel-wise analysis, their scores were replaced by the mean MADRS score of the non-responder group.

Table 1.

Statistical analyses of clinical variables.

	Responders	Non-Responders	Statistic	df	P	Bayes factor
Sex (female/male)	3/5	14/20	0.0363	1	1.00
Age (years)	38.1 ± 11.0	44.6 ± 11.5	1.450	40	0.155	0.781
Age of onset	15.9 ± 9.61	17.8 ± 10.5	0.460	36	0.648	0.374
Age of first treatment	24.0 ± 7.86	29.4 ± 12.5	1.156	35	0.256	0.597
Disease duration	20.4 ± 14.4	24.4 ± 15.3	0.672	40	0.506	0.396
MADRS	6.38 ± 2.83	11.7 ± 5.72	2.547	38	0.015	3.671
HADS-Total	18.4 ± 7.61	20.8 ± 6.78	0.895	37	0.377	0.489
Y-BOCS at baseline	28.0 ± 3.55	28.4 ± 4.17	0.276	40	0.784	0.374
RM-ANOVA – Main time effect
RM Factor		Mean Difference	SE	df	t	pbonferroni
D0	D14	3.57	0.550	41.0	6.50	<0.001
	D45	5.60	0.977	41.0	5.73	<0.001
D14	D45	2.02	0.789	41.0	2.57	0.042

Data are presented as mean ± SD. Y-BOCS Yale-Brown Obsessive Compulsive Disorder Scale, MADRS Montgomery-Asberg Depression Rating Scale, HADS hospital anxiety and depression scale. D0 corresponds to Baseline assesment.

Table 2 showed that none of the linear models were significant for predicting Y-BOCS evolution (p = 0.413 and p = 0.431), but R² increased from 0.248 to 0.306 in the model including FE metrics. For the binomial models, the FE model was significant (p = 0.005), while the model including only clinical metrics was not significant (p = 0.086). R² was also higher in the FE model (R² = 0.669 vs R² = 0.367).

Table 2.

General performance of linear and binomial predictive models.

Model Fit - Linear Regression
			Overall Model Test
Model	R	R²	F	df1	df2	p
1	0.498	0.248	1.07	8	26	0.413
Model Fit - Linear Regression (FE metrics)
			Overall Model Test
Model	R	R²	F	df1	df2	p
1	0.553	0.306	1.06	10	24	0.431
Model Fit Measures - Binomial Regression
				Overall Model Test
Model	Deviance	AIC	R²McF	χ²	df	p
1	23.8	41.8	0.367	13.8	8	0.086
Model Fit Measures - Binomial Regression (FE metrics)
				Overall Model Test
Model	Deviance	AIC	R²McF	χ²	df	p
1	12.5	34.5	0.669	25.2	10	0.005

Note: Models estimated using sample size of N = 35.

The post hoc test, performed on the FE binomial model, showed that none of the covariates were significant (Table 3).

Table 3.

Post hoc test for the Finite-Element Binomial Model.

Model Coefficients - Binomial Regression (FE metrics)
Predictor	Estimate	SE	Z	p
Intercept*	−32.28	36.52	−0.88	0.38
Sex:
F – M	−2.10	4.99	−0.42	0.67
Age	1.11	0.97	1.15	0.25
MADRS Baseline	−2.78	2.23	−1.25	0.21
HADS Total Baseline	0.37	0.37	1.00	0.32
Y-BOCS baseline	0.60	0.84	0.71	0.48
Age of onset	−0.55	0.53	−1.04	0.30
Age of first treatment	−1.45	1.27	−1.14	0.25
Disease duration	−0.21	0.21	−1.01	0.31
J_max (A/m)	236.41	215.36	1.10	0.27
Focality (mm3)	0.00	0.00	−0.98	0.33

Voxel-wise analysis

First, the differences in $J_n$ values between non-responders ($J_n^{\mathit{NR}}$) and responders ($J_n^R$) were analyzed and no clusters were identified. Next, the relationship between polarization levels and Y-BOCS score evolution was examined through multiple regression analysis. Two clusters were identified in which depolarization was associated with a reduction in Y-BOCS scores in the left anterior prefrontal cortex (BA 10, p = 0.025, k = 65, FDR at peak value = 0.151, T at peak value = 5.01) and the right frontal eye field (BA 8, p = 0.015, k = 79, FDR at peak value = 0.151, T at peak value = 4.89) (Fig. 2).

Fig. 2.

Regression model in each cluster where Y-BOCS reduction was associated with depolarization: x-axis is the Y-BOCS evolution and y-axis the polarization level.

Additionally, one cluster was found in the right pars orbitalis (BA 47, p = 0.032, k = 59, FDR at peak value = 0.264), where hyperpolarization was associated with Y-BOCS score reduction (Fig. 3). Once again, no correlations were found with $J_{\mathit{magn}}$ maps or MADRS.

Fig. 3.

Regression model in each cluster where Y-BOCS reduction was associated with hyperpolarization: x-axis is the Y-BOCS evolution and y-axis the polarization level.

Discussion

In this study, we present the first investigation of simulated tDCS using FE modeling of EF on treatment response in OCD. Although more studies have used individualized FE modeling with predefined regions of interest [1, 43, 53], or voxel-wise comparison [38, 54, 55] for other pathologies. None has been reported for OCD treatment and none analyzed depolarization levels.

First, we found that MADRS scores differed between responders and non-responders to tDCS treatments. However, further analysis suggested that MADRS was not significant in predicting a response to tDCS treatment or current distribution. In OCD, depression is a common comorbidity, with 41% specifically meeting criteria for major depressive disorder [56]. The presence of MDD is also linked to more severe OCD symptoms and higher suicide risk [57]. In the MDD literature, studies have found that tDCS was less effective in patients with more severe depressive symptoms [58]. Similar results were also found in other neuromodulation methods such as rTMS [59, 60], suggesting that a similar effect may also apply to OCD patients.

When comparing Y-BOCS evolution, the first two clusters were found in the left anterior prefrontal cortex (BA 10) and the right frontal eye field (BA 8), between the two electrodes. The third one was found in the right pars orbitalis (BA 47). In both BA 8 and BA 10 clusters, higher depolarization levels were associated with a reduction in Y-BOCS scores. In the pars orbitalis, a higher hyperpolarization was associated with a reduction in Y-BOCS scores. Notably, with our statistical thresholds, no correlation was found for $J_{\mathit{magn}}$ maps.

We consider several hypotheses to explain our results. The first is that tDCS succeeded in regulating the cortico-striato-thalamo-cortical (CSTC) circuits. The second is that electrode size and positioning can affect focality and EF diffusion.

For the first hypothesis, hyperpolarization of the right pars orbitalis was associated with a reduction in Y-BOCS scores. The pars orbitalis region as defined cytoarchitecturally is included in the broader orbitofrontal cortex (OFC) [61], which plays a role in OCD in the cortico-striato-thalamo-cortical (CSTC) model [62], particularly within the limbic circuit, which is involved in the generation and evaluation of emotional responses [63, 64] such as fear and anxiety (with the amygdala also playing an important role) [65]. In OCD, hyperactivation of the OFC has been consistently reported, often accompanied by hyperactivation of the ACC during symptom provocation [66, 67]. These activations are thought to reflect exaggerated error monitoring and increased emotional reactivity, leading to an abnormal emotional valence attributed to external stimuli. In tDCS, cathodal stimulation has been associated with inhibition of neuronal activity, demonstrated via a change in motor-evoked potential [68]. These findings suggest that long-term effects of tDCS may inhibit OFC activity, reduce OCD symptoms, and regulate CSTC circuits. This observation is also supported by fMRI studies in which tDCS improved inhibition task performance (Stroop) [69]. For BA 8 and BA 10, depolarization of these areas was associated with a decrease in Y-BOCS scores. These clusters could indicate more favorable current flow in the frontal cortex without necessarily affecting cortical excitability.

Concerning the second hypothesis, to date, only three studies have used finite element (FE) modeling to analyze the response to tDCS treatment in OCD. A meta-analysis of tDCS protocols, incorporating EF modeling in one patient to explore electric field variability across different montages was conducted [45]. Their analysis suggested that montages with the main electrode over the (pre)supplementary motor area (pre-SMA) and an extracephalic reference electrode tend to produce stronger electric fields in predefined ROIs. Similarly, another study tested 13 different montages, identifying two main types: focal montages, using two cephalic electrodes placed over the frontal cortex to produce strong, localized EF in prefrontal areas, and diffuse montages, with one electrode on the frontal cortex and an extracephalic reference, resulting in a broader EF spread [44]. Pinto, Cavendish, and da Silva (2022) further supported the use of a cathode over the pre-SMA with an extracephalic anode as an optimal diffuse configuration [45].

In our study, Y-BOCS reduction was associated with higher depolarization levels in the left anterior prefrontal cortex, which appears inconsistent with the cathodal stimulation of the pre-SMA. This suggests that cathodal stimulation of the pre-SMA may not only hyperpolarize the SMA but also depolarize frontal regions. This indicates that tDCS effects may involve mechanisms more complex than simple hyperpolarization or depolarization of targeted regions and may reflect the combined action of several neuronal mechanisms. Given that current flows from anode to cathode, our simulations also revealed hyperpolarization clusters in the right pars orbitalis, potentially reflecting that, despite the anode placed near the OFC, hyperpolarization could occur because of gyrus orientation [39]. As highlighted in prior meta-analyses [45, 70], the use of a pre-SMA/extracephalic montage enhances EF in target ROIs, though it comes at the cost of reduced focality [44]. This trade-off is evident in our own montage, which, while effective in stimulating prefrontal regions, might benefit from refinement for greater precision.

Beyond montage selection, the exact positioning of electrodes plays a critical role in determining EF distribution. In our study, EEG landmarks were used to position electrodes, with the cathode placed at FCz—a method shown to be accurate within <1.0 cm for M1 targeting in clinical settings [71]. The frontal anode was initially placed at Fp2 but adjusted ~1 cm above to avoid ocular overlaps, an approach also implemented in our EF modeling through manual placement in SimNIBS. However, inter-individual anatomical variability can lead to deviations exceeding this margin [72], thereby altering the resulting EF distribution [73]. Additionally, the use of relatively large electrodes (35 cm²) likely reduced focality, contributing to a broader EF spread [74]. To enhance focality in future studies, smaller electrodes or HD-tDCS setups may be considered [75]. It is also noted that modeling could lead to discrepancies between modeled and clinical electrode placement, which can alter the computed current flow [76]. However, deviation from planned montages is known to be higher in HD-tDCS setups than in standard tDCS protocols [76].

Although investigating the electric field has led us to certain conclusions, this study has several limitations. First, the study includes a limited number of patients, with an especially low proportion of responders compared to non-responders. This issue is, to some extent, inherent to the current clinical use of tDCS, which is often reserved for patients resistant to pharmacological treatments and therefore shows relatively low response rates. However, improving tDCS efficacy precisely requires the identification of robust biomarkers, which in turn demands more balanced samples with higher numbers of responders.

Additionally, since our study was retrospective, a prospective study would be more valuable in predicting the relationship between electric field diffusion and therapeutic response, given that these modeling pipelines have now reached a level of reliability suitable for clinical studies [77]. Finally, we did not account for the heterogeneity of OCD symptom dimensions, which are known to involve distinct neural circuits [78]. OCD is a highly heterogeneous disorder, and different symptom profiles (e.g., contamination/washing, checking, symmetry/ordering, etc.) have been associated with specific patterns of functional and structural brain alterations [78]. Therefore, a one-size-fits-all stimulation protocol may not optimally target the relevant dysfunctional circuits in all patients. Future studies should explore the potential of symptom-specific neuromodulation approaches by tailoring stimulation parameters and target regions to the individual’s clinical presentation.

A final limitation is that none of our results passed FDR correction, which may indicate a strong possibility of type I error. These errors are well known in imaging studies, especially in fMRI studies. However, our setup differs from fMRI studies. Indeed, FE modeling is based on matrix computation from known boundary conditions, such as electrode positioning and current intensity, and is not based on direct imaging to generate results except for the geometry generation of the model, based on T1 MRI. This approach removes some sources of error common in fMRI, such as the presence of noise or magnetic field inhomogeneity. In a similar study, using the same methodology, a non-conservative approach was also used [38]. Using a more conservative approach with higher p-value correction in this context could also lead to an increase in type II errors (i.e., missing true effects) [79].

In conclusion, studying the diffusion of the electric field in the management of OCD using tDCS is essential for optimizing treatment efficacy. Understanding how the electric field diffuses across different brain regions can help tailor electrode placement and potentially enhance therapeutic outcomes. Studies examining the impact of current direction across various tDCS montages have concluded that substantial inter-individual variability in current direction likely contributes to the variability in neuromodulation effects [39]. In our study, statistical results were stronger when using polarization computation instead of current density or EF magnitude. Additionally, we observed that two studies [80, 81] employed a framework to compute depolarization and hyperpolarization at the neuronal level. However, the feasibility of applying this computation in clinical settings and its potential contribution to our understanding of tDCS effects remain unclear. In our study, we noted that the montage used may be suboptimal, as evidenced by the limited number of responders in our cohort. Here, we showed through electric field modeling that close cephalic placement of two large electrodes may not be the optimal option. Indeed, better alternatives would include combined cephalic and extracephalic placement, as recommended by Pinto et al. [45], or the use of smaller electrode sizes in close-placement configurations, or HD-tDCS. OCD remains a challenging disorder to treat. This line of research holds promise for refining non-invasive neuromodulation techniques and offering more precise, effective interventions for patients.

Author contributions

JG: Conceptualization; data curation; formal analysis; investigation; methodology; software; validation; visualization; writing – original draft; Writing – review and editing. AG: Conceptualization; Methodology; Project administration, Resources supervision; Writing – review and editing. KEH: Project administration; Supervision; Writing – review and editing. MR: Resources; Supervision; Writing – review and editing. PB: Data curation; Supervision; Writing – review and editing. NL: Investigation; Methodology; Writing – review and editing. RG: Resources; Writing - review and editing. NJ: Supervision, Writing – review and editing. GHG: Formal analysis; Investigation; Project administration; Supervision; Validation; Wiriting – original draft; Writing – review and editing.

Funding

J.G. discloses support for the research of this work from a CIFRE PhD fellowship [grant number 2022/1763.]. The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The other authors declare no relevant funding.

Data availability

The data and materials supporting the findings of this study are not publicly available due to ethical and privacy restrictions related to clinical data. They are available from the corresponding author upon reasonable request, subject to approval by the relevant institutional and ethical authorities, where applicable.

Competing interests

Julie Gosez, Karim El Houari, Michel Rochette and Prasanth Bokam report a relationship with ANSYS Inc that includes employment. Other authors declared that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Ethics approval and consent to participate

All methods were performed in accordance with the relevant guidelines and regulations. The studies were registered on ClinicalTrials.gov under the registration numbers NCT03304600 and NCT02384671. Written informed consent was obtained from all participants prior to inclusion in the studies.