Skip to main content
PMC home page
Bioelectronic Medicine logoLink to Bioelectronic Medicine
. 2024 Oct 30;10:25. doi: 10.1186/s42234-024-00157-2

Machine learning-optimized non-invasive brain stimulation and treatment response classification for major depression

Alejandro Albizu 1,2, Aprinda Indahlastari 1,5, Paulo Suen 6, Ziqian Huang 4, Jori L Waner 1,5, Skylar E Stolte 3, Ruogu Fang 1,3,4, Andre R Brunoni 6, Adam J Woods 1,2,5,✉
PMCID: PMC11524011  PMID: 39473014

Abstract

Background/Objectives

Transcranial direct current stimulation (tDCS) is a non-invasive brain stimulation intervention that shows promise as a potential treatment for depression. However, the clinical efficacy of tDCS varies, possibly due to individual differences in head anatomy affecting tDCS dosage. While functional changes in brain activity are more commonly reported in major depressive disorder (MDD), some studies suggest that subtle macroscopic structural differences, such as cortical thickness or brain volume reductions, may occur in MDD and could influence tDCS electric field (E-field) distributions. Therefore, accounting for individual anatomical differences may provide a pathway to optimize functional gains in MDD by formulating personalized tDCS dosage.

Methods

To address the dosing variability of tDCS, we examined a subsample of sixteen active-tDCS participants’ data from the larger ELECT clinical trial (NCT01894815). With this dataset, individualized neuroimaging-derived computational models of tDCS current were generated for (1) classifying treatment response, (2) elucidating essential stimulation features associated with treatment response, and (3) computing a personalized dose of tDCS to maximize the likelihood of treatment response in MDD.

Results

In the ELECT trial, tDCS was superior to placebo (3.2 points [95% CI, 0.7 to 5.5; P = 0.01]). Our algorithm achieved over 90% overall accuracy in classifying treatment responders from the active-tDCS group (AUC = 0.90, F1 = 0.92, MCC = 0.79). Computed precision doses also achieved an average response likelihood of 99.981% and decreased dosing variability by 91.9%.

Conclusion

These findings support our previously developed precision-dosing method for a new application in psychiatry by optimizing the statistical likelihood of tDCS treatment response in MDD.

Supplementary Information

The online version contains supplementary material available at 10.1186/s42234-024-00157-2.

Introduction

Major Depressive Disorder (MDD) is a complex and multifaceted mental health condition that affects hundreds of millions of individuals worldwide (Otte et al. 2016). MDD has a substantial impact on an individual’s daily functioning and quality of life, leading to difficulties in personal relationships, work productivity, and social activities (Bromet et al. 2011; James et al. 2018). Depressive disorders are the third largest cause of disability worldwide (James et al. 2018), and the economic burden associated with this condition is substantial (James et al. 2018; Greenberg et al. 2021). In 2018, the annual cost of MDD in the United States was estimated to be $326.2 billion, which is a nearly 40% increase from 2010 (Greenberg et al. 2021).

Despite the availability of different treatment options, such as medication and psychotherapy, nearly 31% of patients do not achieve remission (Gibson-Smith et al. 2015), and the long-term outcomes are often unsatisfactory (Nemeroff 2007). Moreover, many of these treatments have non-trivial side effects, including sexual dysfunction, insomnia, weight gain, etc. (Trivedi et al. 2006; Nemeroff 2007). In some cases, these side effects can lead patients to turn away from pursuing or continuing treatment. Due to the high prevalence and substantial burden of MDD, there is a growing need for alternative or complementary interventions that can provide effective and long-lasting relief for individuals with treatment-resistant depression (Trivedi et al. 2006; Nemeroff 2007; Warden et al. 2007).

One such treatment is transcranial direct current stimulation (tDCS), a non-invasive brain stimulation technique that involves the application of a weak electrical current over the scalp to modulate cortical excitability (Albizu et al. 2019). tDCS alters the sub-threshold membrane potential of neurons (Albizu et al. 2019) and increases regional blood flow while modulating local neurotransmitter concentrations during stimulation (Radman et al. 2009a; Fritsch et al. 2010; Reato et al. 2013; Podda et al. 2016; Kronberg et al. 2017; Antonenko et al. 2019; Alvarez-Alvarado et al. 2021). In vivo and in vitro studies have shown that the weak electric field induced by tDCS can modulate cortical excitability (Esmaeilpour et al. 2018), as well as alter synaptic plasticity (Podda et al. 2016; Kronberg et al. 2017, 2020). tDCS has shown promising results for treating various neuropsychiatric conditions (Szymkowicz et al. 2016; Clancy et al. 2018; Indahlastari et al. 2021; Kim et al. 2022), including MDD (Knotkova et al. 2012; Brunoni et al. 2013, 2017). Several randomized controlled trials for MDD have achieved varying results by targeting the hypoactive dorsolateral prefrontal cortex (DLPFC) with tDCS (Fregni et al. 2006; Bares et al. 2008; Boggio et al. 2008; Murphy et al. 2009; Nitsche et al. 2009; DMMsF et al. 2012; Kalu et al. 2012; Palm et al. 2012; Brunoni et al. 2013, 2017). Some studies have reported significant improvements in depressive symptoms compared to sham stimulation (Fregni et al. 2006; Bares et al. 2008; Boggio et al. 2008; Murphy et al. 2009; Nitsche et al. 2009; Kalu et al. 2012; Brunoni et al. 2013, 2017), while others have failed to demonstrate significant effects (DMMsF et al. 2012; Palm et al. 2012).

This may be due to the “one-size-fits-all” or fixed dosing approach adopted by most tDCS trials. Specifically, all participants are assigned to receive the same stimulation parameters (e.g., F3/F4 at 2 mA) without regard to each individual’s unique anatomy. Thus, the level of variability in trial outcomes could be (at least partially) attributed to individual differences in head anatomy. For example, individual MRI-derived computational models of tDCS-induced current are linearly affected by anatomical attributes such as cortical atrophy (Indahlastari et al. 2020a), leukoaraiosis (Indahlastari et al. 2020b), skull thickness (Opitz et al. 2015), adiposity (Truong et al. 2013), etc. While these anatomical factors represent just one dimension of inter-individual variability, anatomical factors have the greatest effect on the distribution and targeting of tDCS, affecting individual response to tDCS (Bulubas et al. 2019; Albizu et al. 2020; Suen et al. 2020; Nandi et al. 2022). Although every brain possesses unique characteristics, variations in brain structure are also particularly evident in the presence of pathology, like depression (Kanner 2004; Palazidou 2012). However, the majority of research has investigated the impact of anatomical differences on tDCS response in healthy populations. Thus, there is a pressing need to investigate the predictive value of pathological differences in anatomy on treatment outcomes in tDCS. A method to account for individual anatomical differences in an MDD population may provide a unique pathway toward a precision medicine approach of tDCS for maximal therapeutic outcomes.

Considering inherent dosing variability across individuals, algorithms of machine learning (ML) can be applied to patient-specific, neuroimaging-derived computational models of electric current to address the heterogeneity of response to tDCS (Albizu et al. 2020, 2023; Kambeitz et al. 2020; Kim et al. 2022). While other studies have utilized machine learning to classify treatment response (Albizu et al. 2020, 2023; Kambeitz et al. 2020; Kim et al. 2022), only our previous results have leveraged this approach to optimize tDCS dosing (Albizu et al. 2020, 2023). Thus, the current study replicates our previous results in a new, clinical and affective domain by applying a machine learning approach to 1) classify individual response to tDCS application, 2) investigate the features of the electric field that are the best classifiers of treatment response, and 3) optimize tDCS parameters to maximize the likelihood of treatment response. Potential findings of this study may show promise in refining stimulation dosing strategy by accounting for individual anatomical differences. Precise targeting of brain regions in need of stimulation may potentially improve outcomes for patients with MDD.

Methods

Structural imaging and behavioral data were sourced from a phase-III, randomized, non-inferiority, triple-arm, placebo-controlled study (NCT01894815; Fig. 1). In the trial, 245 MDD patients were recruited to one of three groups: active-tDCS with placebo pill, sham-tDCS with escitalopram, and sham-tDCS with placebo pill (Brunoni et al. 2015).

Fig. 1.

Fig. 1

Open in a new tab

A schematic of the ELECT clinical trial design. The current study performed secondary analysis of the active-tDCS + placebo group with baseline MRIs (n = 16; green square). Created with Biorender.com

Participants

16 patients with Major Depressive Disorder (MDD) receiving active-tDCS stimulation were selected (in accordance with a previous study (Suen et al. 2020)) for further analysis by the current study [mean (sd) age = 42.8 (10.9), mean HDRS-17 = 21.6 (3.9), 9 F:7 M]. Eligible patients were diagnosed with MDD according to DSM-5 criteria and confirmed by psychiatrists using the Mini-International Neuropsychiatric Interview (MINI) (Amorim 2000). Eligible patients were between 18–75 years old, had a score of 17 or higher on the 17-item Hamilton Depression Rating Scale (HDRS-17), and had a low suicide risk as determined by the MINI. Patients with bipolar disorder, substance abuse or dependence, dementia, a personality disorder, brain injury, a current pregnancy, specific contraindications to tDCS, current or previous use of escitalopram, or previous or concurrent participation in other tDCS trials were excluded. Coexisting anxiety disorders (generalized anxiety disorder, specific phobia, panic disorder, or social anxiety disorder) did not result in exclusion. Before the trial, patients were free of antidepressants or completed a washout. Benzodiazepines were limited to a stable dose of 20 mg/day of diazepam equivalent. The trial was conducted at the University Hospital and Department and Institute of Psychiatry, University of São Paulo.

tDCS protocol and application

Using the Omni-Lateral-Electrode system (Seibt et al. 2015), the anode and cathode electrodes (5 × 5 cm2 pads) were positioned over the left and right dorsolateral prefrontal cortices, respectively (165° rostral and 10 cm apart near F5/F6 of the 10–20 system) (Brunoni et al. 2015). Each session involved a 30-minute application of 2 mA of tDCS (Soterix Medical, tDCS-CT for clinical trials) for 10 weeks. The first 15 sessions occurred daily, except for weekends, while the remaining 7 sessions were held once per week for a total of 22 sessions conducted. Patients in both the active and sham tDCS groups were subjected to the same protocol. The devices were pre-programmed with a randomized code to deliver either active or sham stimulation, However, in the sham group, the current was automatically turned off after 30 seconds.

Treatment outcomes

Trained psychiatrists and psychologists, who were unaware of the trial-group assignments, conducted all assessments. All assessors were trained using the structured interview guide for the HDRS-17 and only started participating in the study if the reliability was > 90% compared to the gold standard. Efficacy and safety were evaluated at screening, baseline, and at the end of weeks 3, 6, 8, and 10. Positive class labels (n = 10) were defined as > 50% reduction between the baseline and 10-week timepoint on the HDRS-17, the Montgomery–Åsberg Depression Rating Scale (MADRS), and Positive and Negative Affect Scale (PANAS), similar to a previous study (Suen et al. 2020). No significant safety concerns were identified during the study. To assess the integrity of trial-group blinding, patients were asked to guess which intervention they had received and to rate the confidence in their prediction. Participants were unable to guess tDCS assignment (Inline graphic).

Imaging sequences and parameters

Structural T1-weighted MRI scans were obtained on a 3T MR system (Achieva, Philips Healthcare, Netherlands). The 3D Fast Field Echo sequence parameters included: repetition time (TR) = 7 ms; echo time (TE) = 3.2 ms; flip angle = 8°; field of view (FOV) = 240 × 240 mm; resolution = 1 × 1 mm; slice thickness = 1 mm; and 180 sagittal slices. MR acquisitions were conducted between 4 and 8 days before baseline and took place at the Department of Radiology (Hospital das Clínicas da Universidade de São Paulo, São Paulo) during the weekends.

Computational model construction

The individual T1-weighted images were converted into a 256 mm3 isometric field of view, 1 mm3 voxel size, and RAS orientation with the FreeSurfer v7.1.1 image analysis suite (http://surfer.nmr.mgh.harvard.edu/). Individual head volumes were segmented into six tissue types: white matter, gray matter, cerebrospinal fluid, bone, skin, and intracranial air with HEADRECO (Nielsen et al. 2018) – provided by SimNIBS v3.2.1 (https://simnibs.github.io/simnibs/build/html/index.html). Individual tissue types were assigned default isotropic conductivity values using the Realistic vOlumetric-Approach to Simulate Transcranial Electric Stimulation (ROAST; https://www.parralab.org/roast/) toolbox (Huang et al. 2019). A custom MATLAB 2022a (https://www.mathworks.com/) script was used to simulate electrodes according to the Omni-lateral Electrode System (Seibt et al. 2015). At these custom locations, 5 × 5 cm2 electrodes with + 2 mA/-2 mA boundary conditions were assigned as anode and cathode, respectively. A finite element solver, getDP, was used to compute voltage solutions to the Laplace equation. Additional MATLAB routines were used to compute current density from electric field and tissue conductivity, in accordance with Ohm’s law (Inline graphic; Fig. 2). Further, to assess the importance of the current density magnitude, the current density direction, and the curren density vector (magnitude and direction) in classifying treatment response; we compare the classification performance of each data type (see supplemental materials for more information).

Fig. 2.

Fig. 2

Open in a new tab

The spatial distribution of current density for all 16 participants

Supervised machine learning

A Support Vector Machine (SVM) algorithm was used to classify the responders from non-responders, like our previous approach (Albizu et al. 2020). SVM searches for the optimal hyperplane that can separate two classes with maximal margin, under the assumption of independently and identically distributed (iid) data (Steinwart and Christmann 2008), which is met in this study. For the detailed machine learning methodology, please refer to the supplemental materials. To replicate our prior results, three data types: current density magnitude, direction, and vector were extracted to be utilized for classification (Albizu et al. 2020). For feature weight generation and deployment, a final model was trained on all 16 current density maps to derive overall classification weights. The feature weights at each voxel were separated by positive and negative weights that classify responders and non-responders, respectively. To identify the specific features that classify tDCS-related depressive symptom improvements, positive weights were normalized (i.e., Inline graphic) with the following equation to determine their percent contribution toward response classification:

graphic file with name M4.gif 1

Regions of interest (ROIs) were defined using the Harvard-Oxford atlas [52], and regions were ranked based on the average weight of each voxel in that region, Inline graphic: Inline graphic, where Inline graphic represents the set of voxels within a specified region of the atlas.

Dose optimization

Learned weights of the SVM model were used in a modified, weighted Gaussian mixture model (GMM) to generate a precision model that accounts for inter-personal variation based on the current distribution of responders. The empirical responder mean, empirical responder variance, and SVM feature weights were used to estimate each Gaussian model. The likelihood (ℓ) of a new subject belonging to the responders’ current distribution (i.e., response likelihood) was calculated using:

graphic file with name M8.gif 2

where Inline graphic is the SVM response classification weight (s.t., Inline graphic), Inline graphic is the empirical responder mean, and Inline graphicis the empirical responder variance for the Inline graphic of Inline graphic features. The likelihood estimate was used as the objective function to optimize tDCS parameters (i.e., electrode placement and injected current intensity; see Fig. 3) for each individual’s unique head anatomy. In total, electrode positions were optimized from 71 locations from the 10–10 system (71 × 70 = 4,970 potential electrode pairs). Further, the injected current intensity was simulated in 0.1 mA increments up to 4 mA for a total of 40 possible input current levels. Thus, the overall tDCS optimization space included 198,800 potential tDCS doses per person. Normalized mutual information, feature-wise regression, and feature-wise dot product were used as metrics to evaluate the performance of optimization by comparing the similarity of the optimization results with the average treatment responder. For visualization of optimization performance, principal component analysis (PCA) was also utilized to project the expansive feature space and Gaussian model into two dimensions (i.e., the first two principal components). As an additional metric of optimization performance, current density volumes of optimized doses for non-responders were passed back through the original SVM model to assess reclassification of non-responders as responders following optimization.

Fig. 3.

Fig. 3

Open in a new tab

A representative optimization space for a single participant. A 5th-degree polynomial fit was applied to demonstrate the gradient of treatment response across multiple parameters. The x-axis represents the applied current intensity in milliamps. The y-axis represents the electrode montage (i.e., the bipolar electrode configuration) by distance from the originally applied montage (e.g., F5/F6). The z-axis represents the statistical likelihood of treatment response for a single participant given the specific combination of parameters

Statistical analyses

The Statistics and Machine Learning Toolbox in MATLAB 2022a was used to carry out statistical analyses. Hedges’ g was computed to define effect sizes of mean differences, corrected for small sample bias. Since all sixteen participants in our study were individuals with no familial relationship, and each participant’s data was collected under the same condition, these data points met the statistical assumptions of iid data.

Results

The parent ELECT clinical trial of the current study primarily found a significant decrease in the HDRS-17 depression scores in both active-tDCS groups compared to sham-tDCS with placebo pill (tDCS with escitalopram: 5.5 points [95% CI, 3.1 to 7.8; P < 0.001; Response rate: 47%], tDCS with placebo pill: 3.2 points [95% CI, 0.7 to 5.5; P = 0.01; Response rate: 41%]) in the intent-to-treat analysis.

Machine learning classification of tDCS intervention efficacy

In this study, we used a Support Vector Machine (SVM) learning algorithm to distinguish tDCS responders from non-responders based on modeled current density magnitudes, directions, and vectors (a combination of magnitudes and directions). Here, the SVM model accurately distinguished tDCS responders from non-responders using current density magnitudes, achieving an overall accuracy of 91.25% [95% CI 88.54% – 93.96%] with an area under the curve (AUC) of 0.90, an F1 score of 0.93, and a Matthews correlation coefficient (MCC) of 0.82 (see Fig. 4). We also found that current density magnitudes outperformed current density vectors and current density directions in classifying treatment response (Inline graphic). Specifically, the accuracy of the model based on current density vector was 85%, with an AUC of 0.89, an F1 score of 0.88, and an MCC of 0.69. The accuracy of the model based on current density direction was 81%, with an AUC of 0.85, an F1 score of 0.86, and an MCC of 0.59 (see Table 1 for full results). These results indicate that current density magnitudes are a reliable classifier of tDCS treatment response in MDD.

Fig. 4.

Fig. 4

Open in a new tab

(A) A schematic diagram of the treatment response classification workflow. (B) The mean raw and fitted ROC curve of each data type across ten iterations. (C) The AUC of the ROC curve for each of the data types across ten iterations. (D) an aggregate confusion matrix across ten iterations for the best performing model. (E) Discrimination maps of regions that classify treatment response with the percent contribution of each voxel to the SVM decision function, superimposed onto the MNI152 Template

Table 1.

A summary of the ablation experiment with SVM model performance per data type

Intensity Direction Combined
Accuracy 0.913 0.688 0.788
AUROC 0.918 0.808 0.866
F1 Score 0.932 0.795 0.835
MCC 0.812 0.301 0.539
Sensitivity 0.960 0.970 0.860
Specificity 0.833 0.217 0.667
FPR 0.167 0.783 0.333
FNR 0.040 0.030 0.140
Open in a new tab

AUROC – Area Under the Receiver Operating Characteristic Curve, MCC – Matthews Correlation Coefficient, FPR – False Positive Rate, FNR – False Negative Rate

Electric Field characteristics to Classify Treatment Response

Furthermore, we examined the brain voxels that classify tDCS responders (i.e., the feature weights learned during training, Inline graphic from Eq. 1). As shown in Fig. 5, the median current magnitude was significantly higher in responders than in non-responders within these brain regions (Inline graphic, see Fig. 5C), indicating a strong positive relationship between current density magnitude and treatment response. To quantify the difference in current density magnitudes between responders and non-responders, we calculated the effect size using Hedges’ Inline graphic. The effect size was 2.75, with a 95% confidence interval between 2.08 and 4.46 (Inline graphic; see Fig. 5D), indicating a large and reliable difference between the two groups. These results suggest that current magnitude is a robust classifier of tDCS treatment response, and that higher current magnitudes within essential brain regions may be associated with a greater likelihood of response.

Fig. 5.

Fig. 5

Open in a new tab

Plots to demonstrate the current density characteristics within regions predictive of tDCS responders. (A) Histogram of current magnitude (bin width of 0.0013 Am-2), with the y-axis representing the number of observations in each bin divided by the total number of observations, where the sum of all bar heights is equal to 1. (B) Cumulative histogram of current magnitude with the height of each step equal to the cumulative number of observations in the bin over the total number of observations in each bin and all previous bins where the height of the last bar is equal to 1. (C) Logistic regression of class labels vs. median current magnitude. Each dot represents a single participants median current magnitude. (D) The Hedges’ g between responders and non-responders is shown in a Gardner-Altman estimation plot. Each dot represents a single participants median current magnitude. The mean difference is plotted on a floating axis as a bootstrap sampling distribution. The mean difference is depicted as a dot; the 95% confidence interval is indicated by the ends of the vertical error bar

Regional contributions toward classification of tDCS response

To visualize the regions of the brain contributing to treatment response, Fig. 6 illustrates the top ten regions of interest (ROIs) from the Harvard-Oxford atlas, based on the average normalized weight per voxel within each ROI. The distribution of percent contribution across Harvard-Oxford ROIs is shown in Fig. 6B and C. The top ranked ROIs that classified depressive symptom improvements were largely located in the prefrontal and medial temporal lobes, as shown in Fig. 6A. These ROIs were labeled as the: (1) Left Superior Frontal Gyrus, (2) Right Superior Frontal Gyrus, (3) Right Supplemental Motor Area, (4) Left Middle Frontal Gyrus, (5) Left Heschl’s Gyrus, (6) Left Supplementary Motor Area, (7) Right Posterior Parahippocampal Gyrus, (8) Right Amygdala, (9) Right Middle Frontal Gyrus, (10) Left Central Operculum (Fig. 6). Overall, these brain regions have been implicated in emotion regulation, cognitive control, and sensory processing, all of which are disrupted in depression.

Fig. 6.

Fig. 6

Open in a new tab

(A) Visualization of ROIs from the Harvard-Oxford atlas ranked based on their contribution toward predictions of treatment response. (B) Rank, label, and mean percent contribution per voxel of each ROI. (C) A bar graph to represent the average percent contribution per voxel within each ROI of the Harvard-Oxford atlas. Within the bar graph, the top ten ROIs are highlighted in red

Dose optimization performance

To correct the differences in current density between responders and non-responders, a GMM (i.e., Eq. 2) was used to compute and optimize the likelihood of treatment response given the estimate current density within the brain. An exhaustive search of the tDCS optimization space was used to identify the global optimum dosing parameters for non-responders to match the current profile of treatment responders. Following optimization, original non-responders were significantly closer to the responder current density mean by 2.34 pooled standard deviations compared to pre-optimization fixed dosing of non-responders (see Fig. 7A, Inline graphic). GMM dose optimization also achieved an average response likelihood of 99.981% (see Fig. 7B), indicating that the optimized doses were highly likely to lead to a positive treatment response. Furthermore, the optimized doses shared 60.9% (7.4 bits) of normalized mutual information with the mean current distribution for responders, indicating that the optimized doses were similar to those given to responders (see Fig. 7C). Regression of the mean optimized current density vector demonstrated strong feature-wise coherence (see Fig. 7F, Inline graphic), which suggest that the optimized doses exhibited patterns of electrical current indicative of treatment response. Additionally, the GMM optimization decreased dosing variability by 91.9%, suggesting that the optimized doses were more consistent than non-optimized fixed doses. Furthermore, the optimized doses exhibited 203.4% greater average feature-wise dot product with the current density vector of responders, indicating that the optimized doses were more similar to those given to responders than conventional non-responder fixed doses (see Fig. 7D-E). Finally, the optimized doses of the non-responder group were passed back through the original SVM discriminator and 100% of these optimized doses were predicted to produce treatment responders, indicating a high level of classification accuracy and effectiveness of the optimization strategy. Overall, these results demonstrate that dose optimization maximized the likelihood of treatment response by consistently matching the current density distribution of treatment responders throughout the brain.

Fig. 7.

Fig. 7

Open in a new tab

Conventional fixed dosing compared to optimized dosing. (A) Dosing variability measured by root mean squared error (RMSE) compared to the responder current density within the brain for fixed versus optimized doses, respectively. Black dots represent the mean values. Error bars represent ± 1 SD from the mean values. Histograms represent the normal distribution of the sample. (B) Response likelihood of responder (green), non-responder (red), and optimized (blue) models. Contour lines represent a 3D Gaussian distribution of the first and second principal components (i.e., PC1 and PC2) for responders. Histograms represent the smoothed distribution of PC1 and PC2 for estimated current density. (C) Estimated current density reaching the brain for responder (green), non-responder (red), and optimized (blue) dosing. Dots represent the median of each distribution. (D) 3D dot product of the mean current density vector for the fixed versus responder mean. (E) 3D dot product of current density vectors for average optimized dose versus average responder dose. (F) Scatter plot of the voxel-wise mean current density of optimized doses versus the voxel-wise mean current density of tDCS responders

Discussion

The results of this study suggest that machine learning algorithms applied to computational models of tDCS can effectively classify and optimize treatment response in individuals with major depression. Specifically, the current study applied a machine learning approach to (1) classify individual response to tDCS application, (2) investigate the features of the electric field that are the best classifiers of treatment response, and (3) optimize tDCS parameters to maximize the likelihood of treatment response. Our findings highlight the potential for machine learning algorithms to identify classification markers of response and prospectively adjust dosing parameters for optimal treatment outcomes.

Response classification and current characteristics

The classifcation capabilities of this approach may be particularly useful in identifying individuals who are likely to benefit from tDCS and in developing personalized treatment approaches. Furthermore, our findings suggest that the magnitude of the electric field is the most effective classifier of treatment response, outperforming current density vector and direction. Specifically, responders were found to have greater current density magnitudes within the voxels that discriminate tDCS responders from non-responders. This finding suggests that the overall magnitude of current delivered may be a critical factor in the therapeutic effects of tDCS. Other current modeling studies have observed similar dose-response relationships to further support this notion (Amorim 2000; Seibt et al. 2015; Chauhan et al. 2018; Göksu et al. 2018; Antonenko et al. 2019; Suen et al. 2020). However, it is important to note that the direction of current may still play an important role in determining response. Both normal and tangential current have been demonstrated as essential components of current (Radman et al. 2009b; Rahman et al. 2013; Lafon et al. 2017; Rawji et al. 2018). Thus, future studies should integrate both current magnitude and directional components for a more comprehensive understanding of how tDCS modulates brain activity and improves treatment efficacy.

The use of machine learning algorithms in the context of tDCS treatment for major depression has several potential benefits. By identifying classification markers of response, machine learning algorithms may help to reduce reliance on fixed dosing or trial-and-error approaches currently favored in tDCS treatment, which can lead to suboptimal treatment outcomes and prolong the duration of treatment. Additionally, machine learning algorithms can be used to prospectively adjust dosing parameters to improve treatment outcomes, potentially leading to more efficient and effective treatment.

Regions of importance

To that end, our study identified specific ROIs that were effective classifiers of response to tDCS. These ROIs were primarily located in the prefrontal and medial temporal lobes, with the bilateral superior frontal gyri, parahippocampal gyrus, and amygdala being within the top 10 best classification ROIs. The prefrontal cortex (PFC) is involved in regulating mood and emotion (Drevets 2007; Pizzagalli 2014), as well as cognitive functions such as decision-making, problem-solving, and attention. Studies have shown that individuals with MDD have decreased activity in the PFC (Grimm et al. 2008; Diener et al. 2012), particularly in the dorsolateral prefrontal cortex (DLPFC) and anterior cingulate cortex (ACC). The importance of frontal regions is also likely a result of those being the areas targeted by the stimulation. The amygdala, on the other hand, is a key brain region involved in processing emotions, particularly fear and anxiety. Research has suggested that individuals with MDD may have increased activity in the amygdala, leading to heightened emotional reactivity and sensitivity to negative stimuli (Whalen et al. 2002; Diener et al. 2012). Our findings of the best classification ROIs are consistent with previous research suggesting that prefrontal and medial temporal lobes play a critical role in the pathophysiology of depression and the mechanism of action of tDCS (Whalen et al. 2002; Goldstein et al. 2007; Grimm et al. 2008; Diener et al. 2012; Lindquist et al. 2017).

Dose optimization

The use of GMM optimization to personalize dosing for individual patients was also effective in computationally improving treatment response. Our results showed that GMM optimization led to significantly increased statistical response likelihood and decreased dosing variability compared to conventional fixed dosing for non-responders. Additionally, after receiving the optimized dose using our GMM method, the non-responder group were passed back through the original SVM discriminator, and 100% of them were classified to produce the same level of current distribution as the responder group. These findings suggest that the optimized dosing approach may be a promising method for improving the efficacy of tDCS for the treatment of depression.

Study limitations

Despite the promising findings of this study, several limitations should be noted. First, the sample size of this study was relatively small, which may limit the generalizability of our findings. Future studies with larger sample sizes are needed to validate our results and to determine the clinical applicability of our findings. Additionally, the current study used a retrospective design, which limits our ability to draw causal inferences about the relationship between tDCS parameters and treatment response. Prospective and controlled precision-dosing studies are needed to confirm the causal relationship between tDCS parameters and treatment response. Another limitation of our study is the use of computational models to estimate the electric field distribution. Although computational models are widely used in tDCS research, they are based on simplified assumptions (e.g., tissue segmentation, isotropic conductivity values, ideal electrode placement, etc.) and may not fully capture the complexity of the electrical properties of the brain. Utilizing neuronavigated electrode placement and modeling white matter anisotropy might improve these results in future studies. Furthermore, our current optimization approach primarily accounts for physical differences, such as anatomical variations, in a multidomain problem where functional and physiological factors may also play critical roles. Accounting for additional domains, such as combining structural MRI (sMRI) and functional MRI (fMRI) data, would likely improve the predictive accuracy and efficacy of individualized tDCS dosing. Finally, the need for MRI scans to perform personalized treatment planning could be cost prohibitive in a clinical setting; if the results of the study prove marginal, the cost of MRI would not justify this approach. Thus, our findings should be interpreted with caution and validated using other methods, such as in-vivo current density imaging and randomized control trials (Kasinadhuni et al. 2017; Göksu et al. 2018).

Conclusion

In summary, the ability of the current approach to identify classification markers of tDCS response and adjust dosing parameters accordingly may enable clinicians to optimize treatment outcomes for individual patients. This approach may also help to address the issue of variability in treatment response across individuals, which is a common challenge in the treatment of depression. The optimized dosing approach achieved an average response likelihood of 99.981% and significantly decreased dosing variability by 91.9% compared to conventional non-responder fixed doses. Furthermore, our findings suggest that the magnitude of the electric field within the brain is an effective classifier of treatment response, demonstrating the importance of optimizing tDCS parameters to account for individual anatomical differences. Our findings highlight the potential for machine learning algorithms to improve the effectiveness of tDCS treatment for depression. However, further research is needed to validate our findings and to determine the clinical applications of this approach. Nonetheless, these findings have important implications for the development of personalized tDCS dosing regimens for the treatment of depression.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1 (29.2KB, docx)

Acknowledgements

We gratefully acknowledge the support of NVIDIA AI Tech Center (NVAITC) to this research project, especially the great support with GPU and MPI technology from Oded Green (NVAITC), Jingchao Zhang (NVAITC), and Kaleb Smith (NVAITC).

Author contributions

AA: contributed to conceptualization, investigation, methodology, project administration, formal analysis, resources, validation, writing – original draft, writing – review & editing. PS: contributed to conceptualization, investigation, methodology, project administration, formal analysis, resources, validation, writing – original draft, writing – review & editing. ZH: contributed to conceptualization, investigation, data curation, formal analysis. JLW: contributed formal analysis, data curation, writing – review & editing. SES: contributed to conceptualization, investigation, methodology. AI: contributed to conceptualization, investigation, methodology, project administration, formal analysis, resources, validation, writing – review & editing. RF: contributed to conceptualization, investigation, methodology, project administration, formal analysis, resources, validation, writing – review & editing. ARB: contributed to funding acquisition, conceptualization, investigation, methodology, project administration, writing – review & editing. AJW: contributed to conceptualization, investigation, methodology, project administration, writing – original draft, writing – review & editing.

Funding

This work was supported by the National Institutes of Health/National Institute on Aging (RF1AG071469), the National Science Foundation (1842473), the University of Florida McKnight Brain Institute and the McKnight Brain Research Foundation. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health and the National Science Foundation. Supported by a grant (2012/20911-5) from the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP), an independent public foundation, the income of which is provided by the state of São Paulo; by grants from the 2013 NARSAD Young Investigator from the Brain and Behavior Research Foundation (20493, to Dr. Brunoni), the 2013 FAPESP Young Researcher from the São Paulo State Foundation (20911-5, to Dr. Brunoni), and the National Council for Scientific and Technological Development (CNPq; 470904 and 30167-PQ2, to Dr. Brunoni); by the Associação Beneficente Alzira Denise Hertzog da Silva (to the LIM27 neuroscience laboratory); and by scholarships from the Brazilian Coordination for the Improvement of Higher Ed.

Data availability

The processed data of this study are available upon request from the corresponding author. The raw data are not publicly available due to potential identifying information that could compromise participant privacy. Source data are provided with the paper in a public repository (https://github.com/aa14av/BioElecMed_Albizu_2024/).

Declarations

Ethical approval and consent to participate

The trial received local and national ethics committee approval (CAAE:10173712.3.0000.0076) of the University Hospital and Clinics of the University of São Paulo and is registered in clinicaltrials.gov (NCT01894815). All study participants provide written, informed consent for participation in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  1. Albizu A, Fang R, Indahlastari A, O’Shea A, Stolte SE, See KB, Boutzoukas EM, Kraft JN, Nissim NR, Woods AJ. Machine learning and individual variability in electric field characteristics predict tDCS treatment response. Brain Stimul. 2020 Nov-Dec;13(6):1753-1764. 10.1016/j.brs.2020.10.001 [DOI] [PMC free article] [PubMed]
  2. Albizu A, Indahlastari A, Huang Z, Waner J, Stolte SE, Fang R et al. Machine-learning defined precision tDCS for improving cognitive function. 2023 [cited 2023 Jun 16]; 10.1016/j.brs.2023.05.020 [DOI] [PMC free article] [PubMed]
  3. Albizu A, Indahlastari A, Woods AJ. Non-invasive Brain Stimulation. In: Encyclopedia of Gerontology and Population Aging [Internet]. Springer International Publishing; 2019 [cited 2021 Mar 29]. pp. 1–8. https://link.springer.com/referenceworkentry/10.1007/978-3-319-69892-2_682-1
  4. Alvarez-Alvarado S, Boutzoukas EM, Kraft JN, O’Shea A, Indahlastari A, Albizu A, et al. Impact of Transcranial Direct current stimulation and cognitive training on frontal lobe neurotransmitter concentrations. Front Aging Neurosci. 2021;13:711. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Amorim P. Mini International Neuropsychiatric Interview (MINI): validação de entrevista breve para diagnóstico de transtornos mentais. Brazilian J Psychiatry [Internet]. 2000 Sep [cited 2023 Mar 15];22(3):106–15. http://www.scielo.br/j/rbp/a/8BcLbLX3QHtPMsxjtFCHqKb/?lang=pt
  6. Antonenko D, Thielscher A, Saturnino GB, Aydin S, Ittermann B, Grittner U et al. Towards precise brain stimulation: Is electric field simulation related to neuromodulation? Brain Stimul. 2019. [DOI] [PubMed]
  7. Bares M, Brunovsky M, Kopecek M, Novak T, Stopkova P, Kozeny J, et al. The decrease of prefrontal theta QEEG cordance value predicts response to Venlafaxine treatment in patients with resistant depression. Eur Psychiatry. 2008;23:S74. [DOI] [PubMed] [Google Scholar]
  8. Boggio PS, Rigonatti SP, Ribeiro RB, Myczkowski ML, Nitsche MA, Pascual-Leone A et al. A randomized, double-blind clinical trial on the efficacy of cortical direct current stimulation for the treatment of major depression. Int J Neuropsychopharmacol [Internet]. 2008 Mar 1 [cited 2023 Mar 15];11(2):249–54. https://academic.oup.com/ijnp/article/11/2/249/767876 [DOI] [PMC free article] [PubMed]
  9. Bromet E, Andrade LH, Hwang I, Sampson NA, Alonso J, de Girolamo G et al. Cross-national epidemiology of DSM-IV major depressive episode. BMC Med [Internet]. 2011 Jul 26 [cited 2023 Mar 15];9(1):1–16. https://bmcmedicine.biomedcentral.com/articles/10.1186/1741-7015-9-90 [DOI] [PMC free article] [PubMed]
  10. Brunoni AR, Moffa AH, Sampaio-Junior B, Borrione L, Moreno ML, Fernandes RA et al. Trial of Electrical Direct-Current Therapy versus Escitalopram for Depression. N Engl J Med [Internet]. 2017 Jun 29 [cited 2023 Mar 10];376(26):2523–33. https://www.nejm.org/doi/full/10.1056/nejmoa1612999 [DOI] [PubMed]
  11. Brunoni AR, Sampaio B, Moffa AH, Borrione L, Nogueira BS, Aparício LVM et al. The Escitalopram versus Electric Current Therapy for Treating Depression Clinical Study (ELECT-TDCS): rationale and study design of a non-inferiority, triple-arm, placebo-controlled clinical trial. Sao Paulo Med J [Internet]. 2015 Jan 1 [cited 2023 Mar 28];133(3):252–63. http://www.scielo.br/j/spmj/a/kRH8B94VbvV4JWc3XntqsXj/?lang=en [DOI] [PMC free article] [PubMed]
  12. Brunoni AR, Valiengo L, Baccaro A, Zanão TA, De Oliveira JF, Goulart A et al. The Sertraline vs Electrical Current Therapy for Treating Depression Clinical Study: Results From a Factorial, Randomized, Controlled Trial. JAMA Psychiatry [Internet]. 2013 Apr 1 [cited 2023 Mar 10];70(4):383–91. https://jamanetwork.com/journals/jamapsychiatry/fullarticle/1568955 [DOI] [PubMed]
  13. Bulubas L, Padberg F, Bueno PV, Duran F, Busatto G, Amaro E et al. Antidepressant effects of tDCS are associated with prefrontal gray matter volumes at baseline: evidence from the ELECT-TDCS trial. Brain Stimul. 2019;12(5). [DOI] [PubMed]
  14. Chauhan M, Indahlastari A, Kasinadhuni AK, Schar M, Mareci TH, Sadleir RJ. Low-frequency conductivity Tensor Imaging of the human head in vivo using DT-MREIT: first study. IEEE Trans Med Imaging. 2018;37(4):966–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Clancy KJ, Baisley SK, Albizu A, Kartvelishvili N, Ding M, Li W. Lasting connectivity increase and anxiety reduction via transcranial alternating current stimulation. Soc Cogn Affect Neurosci. 2018;13(12):1305–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Diener C, Kuehner C, Brusniak W, Ubl B, Wessa M, Flor H. A meta-analysis of neurofunctional imaging studies of emotion and cognition in major depression. NeuroImage. 2012;61(3):677–85. [DOI] [PubMed] [Google Scholar]
  17. Blumberger DMMF, Tran LB, Fitzgerald PMMPF, Hoy KBbnsDp, Daskalakis ZJMPF. A Randomized double-blind sham-controlled study of Transcranial Direct current stimulation for treatment-resistant Major Depression. Front Psychiatry. 2012;3:74. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Drevets WC. Orbitofrontal Cortex Function and Structure in Depression. Ann N Y Acad Sci [Internet]. 2007 Dec 1 [cited 2023 Apr 4];1121(1):499–527. https://onlinelibrary.wiley.com/doi/full/10.1196/annals.1401.029 [DOI] [PubMed]
  19. Esmaeilpour Z, Marangolo P, Hampstead BM, Bestmann S, Galletta E, Knotkova H, Bikson M. Incomplete evidence that increasing current intensity of tDCS boosts outcomes. Brain Stimul. 2018 Mar-Apr;11(2):310-321. 10.1016/j.brs.2017.12.002 [DOI] [PMC free article] [PubMed]
  20. Fregni F, Boggio PS, Santos MC, Lima M, Vieira AL, Rigonatti SP, et al. Noninvasive cortical stimulation with transcranial direct current stimulation in Parkinson’s disease. Mov Disord. 2006;21(10):1693–702. [DOI] [PubMed] [Google Scholar]
  21. Fritsch B, Reis J, Martinowich K, Schambra HM, Ji Y, Cohen LG, Lu B. Direct current stimulation promotes BDNF-dependent synaptic plasticity: potential implications for motor learning. Neuron. 2010 Apr 29;66(2):198-204. 10.1016/j.neuron.2010.03.035 [DOI] [PMC free article] [PubMed]
  22. Gibson-Smith D, Bot M, Milaneschi Y, Twisk JW, Visser M, Brouwer Ia, et al. The prevalence and national burden of treatment-resistant depression and major depressive disorder in the United States. J Clin Psychiatry. 2015;77(September):22–7. [Google Scholar]
  23. Göksu C, Hanson LG, Siebner HR, Ehses P, Scheffler K, Thielscher A. Human in-vivo brain magnetic resonance current density imaging (MRCDI). Neuroimage. 2018 May 1;171:26-39. 10.1016/j.neuroimage.2017.12.075 [DOI] [PubMed]
  24. Goldstein JM, Seidman LJ, Makris N, Ahern T, O’Brien LM, Caviness VS et al. Hypothalamic Abnormalities in Schizophrenia: Sex Effects and Genetic Vulnerability. Biol Psychiatry. 2007. [DOI] [PubMed]
  25. Greenberg PE, Fournier A-A, Sisitsky T, Simes M, Berman R, Koenigsberg SH et al. The Economic Burden of Adults with Major Depressive Disorder in the United States (2010 and 2018) Key Points for Decision Makers. Pharmacoeconomics [Internet]. 2021 [cited 2023 Jun 8];39:653–65. 10.1007/s40273-021-01019-4 [DOI] [PMC free article] [PubMed]
  26. Grimm S, Beck J, Schuepbach D, Hell D, Boesiger P, Bermpohl F, et al. Imbalance between Left and Right Dorsolateral Prefrontal Cortex in Major Depression is linked to negative Emotional Judgment: an fMRI study in severe major depressive disorder. Biol Psychiatry. 2008;63(4):369–76. [DOI] [PubMed] [Google Scholar]
  27. Huang Y, Datta A, Bikson M, Parra LC. Realistic volumetric-approach to simulate transcranial electric stimulation-ROAST-a fully automated open-source pipeline. J Neural Eng. 2019 Jul 30;16(5):056006. 10.1088/1741-2552/ab208d [DOI] [PMC free article] [PubMed]
  28. Indahlastari A, Albizu A, Boutzoukas EM, O’Shea A, Woods AJ. White Matter Hyperintensities Affect Transcranial Electrical Stimulation in the Aging Brain. Brain Stimul [Internet]. 2020;14(1):69–73. https://linkinghub.elsevier.com/retrieve/pii/S1935861X20302965 [DOI] [PMC free article] [PubMed]
  29. Indahlastari A, Albizu A, O’Shea A, Forbes MA, Nissim NR, Kraft JN et al. Modeling Transcranial Electrical Stimulation in the aging brain. Brain Stimul. 2020;13(4). [DOI] [PMC free article] [PubMed]
  30. Indahlastari A, Hardcastle C, Albizu A, Alvarez-Alvarado S, Boutzoukas EM, Evangelista ND et al. A Systematic Review and Meta-Analysis of Transcranial Direct Current Stimulation to Remediate Age-Related Cognitive Decline in Healthy Older Adults. Neuropsychiatr Dis Treat [Internet]. 2021 Mar 29 [cited 2021 Mar 29];Volume 17:971–90. https://www.dovepress.com/a-systematic-review-and-meta-analysis-of-transcranial-direct-current-s-peer-reviewed-article-NDT [DOI] [PMC free article] [PubMed]
  31. James SL, Abate D, Abate KH, Abay SM, Abbafati C, Abbasi N, et al. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990–2017: a systematic analysis for the global burden of Disease Study 2017. Lancet. 2018;392(10159):1789–858. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Kalu UG, Sexton CE, Loo CK, Ebmeier KP. Transcranial direct current stimulation in the treatment of major depression: a meta-analysis. Psychol Med [Internet]. 2012 Sep [cited 2023 Mar 15];42(9):1791–800. https://www.cambridge.org/core/journals/psychological-medicine/article/transcranial-direct-current-stimulation-in-the-treatment-of-major-depression-a-metaanalysis/96254C1048E1706414248C27C4E9BCA5 [DOI] [PubMed]
  33. Kambeitz J, Goerigk S, Gattaz W, Falkai P, Benseñor IM, Lotufo PA, et al. Clinical patterns differentially predict response to transcranial direct current stimulation (tDCS) and escitalopram in major depression: a machine learning analysis of the ELECT-TDCS study. J Affect Disord. 2020; [DOI] [PubMed]
  34. Kanner AM. Structural MRI changes of the brain in Depression. Clin EEG Neurosci. 2004;35(1):46–52. [DOI] [PubMed] [Google Scholar]
  35. Kasinadhuni AK, Indahlastari A, Chauhan M, Schär M, Mareci TH, Sadleir RJ. Imaging of current flow in the human head during transcranial electrical therapy. Brain Stimul. 2017;10(4):764–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Kim S, Yang C, Dong SY, Lee SH. Predictions of tDCS treatment response in PTSD patients using EEG based classification. Front Psychiatry. 2022;13(1). [DOI] [PMC free article] [PubMed]
  37. Knotkova H, Rosedale M, Strauss SM, Horne J, Soto E, Cruciani RA et al. Using Transcranial Direct Current Stimulation to Treat Depression in HIV-Infected Persons: The Outcomes of a Feasibility Study. Front Psychiatry [Internet]. 2012 [cited 2021 May 11];3. https://pubmed.ncbi.nlm.nih.gov/22719732/ [DOI] [PMC free article] [PubMed]
  38. Kronberg G, Bridi M, Abel T, Bikson M, Parra LC. Direct Current Stimulation Modulates LTP and LTD: Activity Dependence and Dendritic Effects. Brain Stimul. 2017;10(1):51-58. 10.1016/j.brs.2016.10.001 [DOI] [PMC free article] [PubMed]
  39. Kronberg G, Rahman A, Sharma M, Bikson M, Parra LC. Direct current stimulation boosts hebbian plasticity in vitro. Brain Stimul. 2020 Mar-Apr;13(2):287-301. 10.1016/j.brs.2019.10.014 [DOI] [PMC free article] [PubMed]
  40. Lafon B, Rahman A, Bikson M, Parra LC. Direct Current Stimulation Alters Neuronal Input/Output Function. Brain Stimul [Internet]. 2017 Jan 1 [cited 2021 May 17];10(1):36–45. https://pubmed.ncbi.nlm.nih.gov/27717601/ [DOI] [PMC free article] [PubMed]
  41. Lindquist MA, Krishnan A, López-Solà M, Jepma M, Woo CW, Koban L, Roy M, Atlas LY, Schmidt L, Chang LJ, Reynolds Losin EA, Eisenbarth H, Ashar YK, Delk E, Wager TD. Group-regularized individual prediction: theory and application to pain. Neuroimage. 2017 Jan 15;145(Pt B):274-287. 10.1016/j.neuroimage.2015.10.074 [DOI] [PMC free article] [PubMed]
  42. Murphy DN, Boggio P, Fregni F. Transcranial direct current stimulation as a therapeutic tool for the treatment of major depression: Insights from past and recent clinical studies. Curr Opin Psychiatry [Internet]. 2009 May [cited 2023 Mar 15];22(3):306–11. https://journals.lww.com/co-psychiatry/Fulltext/2009/05000/Transcranial_direct_current_stimulation_as_a.11.aspx [DOI] [PubMed]
  43. Nandi T, Puonti O, Clarke WT, Nettekoven C, Barron HC, Kolasinski J et al. tDCS induced GABA change is associated with the simulated electric field in M1, an effect mediated by grey matter volume in the MRS voxel. Brain Stimul [Internet]. 2022;15(5):1153–62. 10.1016/j.brs.2022.07.049 [DOI] [PMC free article] [PubMed]
  44. Nemeroff CB. Prevalence and Management of Treatment-Resistant Depression. J Clin Psychiatry [Internet]. 2007 Jul 16 [cited 2023 Mar 10];68(suppl 8):15435. https://www.psychiatrist.com/jcp/depression/prevalence-management-treatment-resistant-depression [PubMed]
  45. Nielsen JD, Madsen KH, Puonti O, Siebner HR, Bauer C, Madsen CG, et al. Automatic skull segmentation from MR images for realistic volume conductor models of the head: Assessment of the state-of-the-art. NeuroImage. 2018;174:587–98. [DOI] [PubMed] [Google Scholar]
  46. Nitsche MA, Boggio PS, Fregni F, Pascual-Leone A. Treatment of depression with transcranial direct current stimulation (tDCS): a review. Exp Neurol. 2009;219(1):14–9. [DOI] [PubMed] [Google Scholar]
  47. Opitz A, Paulus W, Will S, Antunes A, Thielscher A. Determinants of the electric field during transcranial direct current stimulation. Neuroimage. 2015 Apr 1;109:140-50. 10.1016/j.neuroimage.2015.01.033 [DOI] [PubMed]
  48. Otte C, Gold SM, Penninx BW, Pariante CM, Etkin A, Fava M et al. Major depressive disorder. Nat Rev Dis Prim 2016 21 [Internet]. 2016 Sep 15 [cited 2023 Mar 9];2(1):1–20. https://www.nature.com/articles/nrdp201665 [DOI] [PubMed]
  49. Palazidou E. The neurobiology of depression. Br Med Bull. 2012;101(1):127–45. [DOI] [PubMed] [Google Scholar]
  50. Palm U, Schiller C, Fintescu Z, Obermeier M, Keeser D, Reisinger E, et al. Transcranial direct current stimulation in treatment resistant depression: a randomized double-blind, placebo-controlled study. Brain Stimul. 2012;5(3):242–51. [DOI] [PubMed] [Google Scholar]
  51. Pizzagalli DA. Depression, Stress, and Anhedonia: Toward a Synthesis and Integrated Model. https://doi.org/101146/annurev-clinpsy-050212-185606 [Internet]. 2014 Mar 28 [cited 2023 Apr 4];10:393–423. https://www.annualreviews.org/doi/abs/10.1146/annurev-clinpsy-050212-185606 [DOI] [PMC free article] [PubMed]
  52. Podda MV, Cocco S, Mastrodonato A, Fusco S, Leone L, Barbati SA, Colussi C, Ripoli C, Grassi C. Anodal transcranial direct current stimulation boosts synaptic plasticity and memory in mice via epigenetic regulation of Bdnf expression. Sci Rep. 2016 Feb 24;6:22180. 10.1038/srep22180 [DOI] [PMC free article] [PubMed]
  53. Radman T, Datta A, Ramos RL, Brumberg JC, Bikson M. One-dimensional representation of a neuron in a uniform electric field. In: 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE; 2009. pp. 6481–4. [DOI] [PMC free article] [PubMed]
  54. Radman T, Ramos RL, Brumberg JC, Bikson M. Role of cortical cell type and morphology in subthreshold and suprathreshold uniform electric field stimulation in vitro. Brain Stimul. 2009 Oct;2(4):215-28, 228.e1-3. 10.1016/j.brs.2009.03.007 [DOI] [PMC free article] [PubMed]
  55. Rahman A, Reato D, Arlotti M, Gasca F, Datta A, Parra LC, Bikson M. Cellular effects of acute direct current stimulation: somatic and synaptic terminal effects. J Physiol. 2013 May 15;591(10):2563-78. 10.1113/jphysiol.2012.247171 [DOI] [PMC free article] [PubMed]
  56. Rawji V, Ciocca M, Zacharia A, Soares D, Truong D, Bikson M, Rothwell J, Bestmann S. tDCS changes in motor excitability are specific to orientation of current flow. Brain Stimul. 2018 Mar-Apr;11(2):289-298. 10.1016/j.brs.2017.11.001 [DOI] [PMC free article] [PubMed]
  57. Reato D, Rahman A, Bikson M, Parra LC. Effects of weak transcranial alternating current stimulation on brain activity-a review of known mechanisms from animal studies. Front Hum Neurosci. 2013;7:687. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Seibt O, Brunoni AR, Huang Y, Bikson M. The pursuit of DLPFC: non-neuronavigated methods to Target the Left Dorsolateral pre-frontal cortex with symmetric Bicephalic Transcranial Direct Current Stimulation (tDCS). Brain Stimul. 2015;8(3):590–602. [DOI] [PubMed] [Google Scholar]
  59. Steinwart I, Christmann A. Support Vector machines. Springer Science & Business Media; 2008.
  60. Suen PJC, Doll S, Batistuzzo MC, Busatto G, Razza LB, Padberg F, Mezger E, Bulubas L, Keeser D, Deng ZD, Brunoni AR. Association between tDCS computational modeling and clinical outcomes in depression: data from the ELECT-TDCS trial. Eur Arch Psychiatry Clin Neurosci. 2021 Feb;271(1):101-110. 10.1007/s00406-020-01127-w [DOI] [PMC free article] [PubMed]
  61. Szymkowicz SM, McLaren ME, Suryadevara U, Woods AJ. Transcranial Direct Current Stimulation Use in the Treatment of Neuropsychiatric Disorders: A Brief Review. Psychiatr Ann. 2016 Nov;46(11):642-646. 10.3928/00485713-20161006-01 [DOI] [PMC free article] [PubMed]
  62. Trivedi MH, Fava M, Wisniewski SR, Thase ME, Quitkin F, Warden D et al. Medication Augmentation after the Failure of SSRIs for Depression. https://doi.org/101056/NEJMoa052964 [Internet]. 2006 Mar 23 [cited 2023 Mar 10];354(12):1243–52. https://www.nejm.org/doi/full/10.1056/NEJMoa052964 [DOI] [PubMed]
  63. Truong DQ, Magerowski G, Blackburn GL, Bikson M, Alonso-Alonso M. Computational modeling of transcranial direct current stimulation (tDCS) in obesity: Impact of head fat and dose guidelines. Neuroimage Clin. 2013 May 31;2:759-66. 10.1016/j.nicl.2013.05.011 [DOI] [PMC free article] [PubMed]
  64. Warden D, Rush AJ, Trivedi MH, Fava M, Wisniewski SR. The STAR*D Project Results: A Comprehensive Review of Findings. 2007 [cited 2023 Mar 15]; http://www.star-d.org [DOI] [PubMed]
  65. Whalen PJ, Shin LM, Somerville LH, McLean AA, Kim H. Functional neuroimaging studies of the amygdala in depression. Semin Clin Neuropsychiatry [Internet]. 2002 [cited 2023 Apr 6];7(4):234–42. https://pubmed.ncbi.nlm.nih.gov/12382206/ [DOI] [PubMed]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Material 1 (29.2KB, docx)

Data Availability Statement

The processed data of this study are available upon request from the corresponding author. The raw data are not publicly available due to potential identifying information that could compromise participant privacy. Source data are provided with the paper in a public repository (https://github.com/aa14av/BioElecMed_Albizu_2024/).

Articles from Bioelectronic Medicine are provided here courtesy of BMC

RESOURCES