Brain Connectomics Markers for Response Prediction to Transcranial Magnetic Stimulation in Cocaine Use Disorder

doi:10.21203/rs.3.rs-4312563/v1

Cocaine use disorder (CUD) is a worldwide health problem with limited effective treatment options. The therapeutic potential of repetitive transcranial magnetic stimulation (rTMS) is gaining more attention following evidence of its role on craving reduction in CUD. However, the heterogeneity of results underscores a pressing need for biomarkers of treatment outcome. We asked whether brain connectomics together with clinical assessments can predict response to add-on rTMS therapy for CUD better than solely conventional clinical assessments.

A total of 36 patients with CUD underwent an open-label acute phase of receiving two daily sessions of 5-Hz rTMS on the left dorsolateral prefrontal cortex (LDLPFC). Subsequently, 19 and 14 patients continued to an open-label maintenance phase of two weekly rTMS sessions for 3 and 6 months, respectively. Pre and post treatment resting-state brain functional connectivity as well as two clinical scores related to craving were measured to predict the subsequent response to rTMS therapy. Two conventional clinical scores, namely Cocaine craving questionnaires (CCQ) and visual analogue scale (VAS) were used as craving level assessments. We used a priori seed-driven connectivity of DLPFC and anterior cingulate cortex (ACC) together with the connectivity from a whole-brain multi-voxel pattern analysis at each time point to predict the reduction in craving after rTMS.

The combination of connectivity changes and baseline craving severity measures improved the prediction of individual craving compared to the prediction with only initial craving severity. The predictive model from the combination of neuromarkers could explain 45 to 97 percent of variance in craving changes assessed by two different clinical scores. We used leave-one-subject-out cross-validation to support the generalizability of our findings. Our results indicate that employing neuromarkers from resting-state functional connectivity of pre and post condition of CUD patients receiving add-on rTMS therapy increases the power of predicting craving changes and support the idea that neuromarkers may offer improvements in precision medicine approaches.

Health sciences/Biomarkers/Predictive markers

Physical sciences/Engineering/Biomedical engineering

resting state functional magnetic resonance imaging (rsfMRI)

brain connectomics

response prediction model

repetitive transcranial magnetic stimulation (rTMS)

craving

dorsolateral prefrontal cortex (DLPFC)

anterior cingulate cortex (ACC)

seed-based connectivity

multi voxel pattern analysis (MVPA)

neuromarkers

Cocaine use disorder (CUD) is considered a worldwide public health problem, with significant contribution to over 12% of mortality, and one of the leading cause of preventable death ¹. This disorder is characterized by patterns of compulsive substance searching, intense craving, impulsive decision making, and psychological behaviors such as depression and anxiety ^2–4. Despite the significant morbidity associated with CUD, there is not yet an effective psychological or pharmacological therapy for it. In recent years, transcranial magnetic stimulation (TMS) has emerged as a non-invasive neuromodulating approach which had positive outcomes in substance use disorders ^5–11^,¹². Likewise, repetitive transcranial magnetic stimulation (rTMS) and theta burst stimulation (TBS) have been used as add-on therapies for treatment of CUD and showed promising outcomes such as reducing craving and substance use ¹⁰^,^13–15.

Craving is a core clinical symptom of cocaine addiction that can be addressed as treatment target. Craving is defined as a pressing and urgent desire for an addictive behavior, motivated by internal and external cues ¹. There are currently various experimental measurements of this state: cocaine craving questionnaires (CCQ), visual analogue scale (VAS), and non-verbal physiological measures ^16,17. Substance craving is an important risk factor of relapse in cocaine addiction ^18,19, and therefore, reducing craving has been targeted to curb the addictive behavior ¹.

Neuroimaging studies in humans have shown that craving is mainly involved in the activation of reward and motivation circuits as well as executive control network (ECN) ⁹^,⁷. The ECN most prominently involves the prefrontal cortex (PFC), including dorsolateral prefrontal cortex (DLPFC) and anterior cingulate cortex (ACC) ²⁰. It has been shown that chronic cocaine use can cause damages to the structure and function of the prefrontal cortex (PFC). Previous studies reported on microstructural changes in frontostriatal circuits ²¹ and impairment of the executive control functions ¹⁰ in CUD. There is also evidence of the gray matter density increase in frontal and temporal regions of the brain after 1 month of abstinence in CUD ²². In this regard, the left dorsolateral prefrontal cortex (LDLPFC) has been considered as a possible target for the application of rTMS in CUD patients. LDLPFC is one of the main hubs of the executive control network in the human brain ⁹ and plays a critical role in the addictive cycle, craving and impulsive behavior ⁵^,²³. The application of rTMS to the left dorsolateral prefrontal cortex has shown promising outcomes in CUD such as reduction in cocaine use and symptoms ¹⁰^,^12,24.

Although rTMS of the LDLPFC could be considered as an additional treatment in CUD, the ability to predict the heterogenous treatment responses is still a challenge. Identifying clinically meaningful and accessible predictors can help with the prediction of treatment response in different individuals as well as the overall treatment design. Several biomarkers achieved from the brain neuroimaging studies (neuromarkers) have previously shown the power in predicting the treatment response in brain disorders ^25–30. Recent studies suggest that neuromarkers can provide higher predictive power than the conventional clinical and demographic measures in rTMS outcomes ^31–33. Among these biomarkers, there are well-replicated evidence of correlational relationship between neuromarkers in LDPFC and ACC and clinical response in CUD ⁷^,³⁴^,³⁵.

In this context, in the current study, we asked whether the neuromarkers from magnetic resonance imaging can help to predict the response to rTMS in CUD. We were particularly interested in the predictors from resting-state functional magnetic resonance imaging (rsfMRI), a widely available neuroimaging technique used to assess the individual’s brain activity at rest. Here, we aim to investigate the power of univariate as well as multivariate connectomics from rsfMRI in prediction of short term and long term rTMS impacts in CUD. First, since the LDLPFC and ACC have been regarded as the loci of dysfunction inside the ECN in CUD, we hypothesized that rTMS outcome would be associated with the seed-driven resting state functional connectivity of these regions. Then, in a data-driven approach, we examined the power of the predictors from resting-state multi-voxel pattern analysis (MVPA) in predicting the rTMS response. Treatment outcome in our study was considered as pre-post change in craving measures (CCQ and VAS), for each of the three time points during the treatment – two weeks, three months, and six months after the start of the treatment.

A. Participants and experimental procedure

The data used in this study was acquired and has been made available on OpenNeuro (https://openneuro.org/datasets/ds003037). This data was obtained from a clinical trial conducted at the Clinical Research Division of the National Institute of Psychiatry in Mexico City, Mexico.

Thirty-six CUD patients (6f/30m) were recruited for the 2-week acute phase of rTMS treatment. Specifically, each patient received 2 daily sessions of 5-Hz rTMS (5,000 pulses) on the left DLPFC for two weeks. All these patients first underwent a double-blind phase were twenty-four of them were actively treated by rTMS. The other twelve patients received two weeks of open-label rTMS after this blind phase. After this acute phase, 19, and 14 participants underwent twice-a-week maintenance phase of rTMS application for 3 months, and 6 months, respectively. Here, we aim to consider patients from these stages who received real rTMS during short and long periods, and investigate the effects of the rTMS application in these time intervals. The cocaine users were included in the study based on the criteria: 1) age 18–50 years, and 2) high cocaine consumption for at least 1 year. Clinical and MRI data were obtained for each patient at baseline (bs), after 2 weeks (2w), after 3 months (3m), and after 6 months (6m). An overview of the demographic data is shown in Table 1. Detailed information about recruitment, study design, and a complete demographics can be found in a previous publication by Garza-Villarreal et al ¹¹.

It should be noted that the original study on this dataset ¹¹ was designed with 44 CUD patients that first underwent a 2-week double-blind randomized controlled trial, with 24 patients in the active group and 20 in the sham group. After the 2-week phase, 12 patients from the sham group underwent real rTMS sessions for 2 weeks. Since our main goal was to study the impacts of the rTMS treatment in the CUD patients, we did not focus on the double-blind trial, and included all patients who received real rTMS for 2 weeks (24 + 12: 36 patients at baseline and 2w). Following this phase, 19 and 14 patients out of the 36 patients participated in the 3m and 6m time points, respectively, and were included in our study, as well.

B. Clinical assessment

The primary clinical score for outcome measurement was craving, including Cocaine Craving Visual Scale and Cocaine Craving Questionnaire Now as described below.

Cocaine Craving Visual Scale (VAS)

This score is used to evaluate the patients’ craving at the moment of assessment. To do so, a 10 cm line with “no craving” and “the most intense craving” at each end of it, is shown to the patients, and they were asked to mark their craving level at the moment of assessment.

Cocaine Craving Questionnaire Now (CCQ-N): The CCQ is a 45-item questionnaire that explores cocaine craving among patients. The CCQ-Now asks participants about their craving at the moment of assessment. The items are related to the following contents: desire to use cocaine, intention and planning to use cocaine, anticipation of positive outcome, the anticipation of relief from withdrawal or dysphoria, and lack of control overuse ⁴.

The detailed statistics of clinical measurements are provided in Table 1.

C. Transcranial magnetic stimulation

During the 10 weekdays of acute phase, 5-Hz rTMS was delivered every day, using a MagPro R30 + Option magnetic stimulator and a figure-eight B65-A/P coil (MagVenture, Alpharettta, GA). Stimulation was delivered with 50 trains of 50 pulses at 100% motor threshold to the left DLPFC. To identify the cortical target for stimulation, vitamin E capsule fiducials were used during the MRI acquisition.

D. Imaging information

MRI scans were acquired using a Philips Ingenia 3T MR system (Philips Healthcare, Best, The Netherlands, and Boston, MA, USA), with a 32-channel Head coil.

T1w images were acquired using a three-dimensional FFE SENSE sequence, TR/TE = 7/3.5ms, filed of view = 240mm², matrix = 240×240mm, number of slices = 180, and voxel size = 1×1×1mm. Resting state functional MRI (rsfMRI) sequences were acquired using a gradient recalled (GE) echo planner imaging (EPI) sequence with 36 axial slices with the following parameters: TR = 2000ms, TE = 30.001ms, flip angle = 75°, matrix = 80×80×37, FOV = 240mm², and voxel size of 3×3×3.3mm. The whole scanning last for 10min and a total of 300 volumes were acquired. During this time, the participants were asked to keep their eyes open, and relax, not thinking about anything in particular ⁴.

E. Preprocessing of anatomical images

T1w image of each subject was preprocessed using the SPM-based CONN Toolbox (RRID:SCR_009550, version 20b, doi:10.56441/hilbertpress.2048.3738). First, each subject’s T1 images were segmented for Grey/White/CSF parts and then were normalized to the standard MNI space. Then, the standard Harvard-Oxford Cortical atlas (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Atlases) was registered to each scan to obtain the related connectivity measures for each subject, and use in further processing.

F. Preprocessing of resting state data

Resting state fMRI data were preprocessed using the default preprocessing pipeline in CONN. Specifically, spatial preprocessing of functional volumes consisted of: realignment, slice timing correction, potential outlier scan detection, normalization to MNI space, and smoothing with an 8 mm full width half maximum Gaussian filter. The default denoising pipeline consisted of regressing out the potentially confounding components of white matter and cerebrospinal fluid, head motion, and identified outlier scans. Finally, the data were band-pass filtered to preserve the frequencies in the 0.01 to 0.08 Hz. The same preprocessing steps were done for the MRI data from two conditions (bs and 2w/3m/6m).

G. Processing

We implemented two main analyses to extract the predictors for the response prediction model. The methodological overview of the processing is displayed in Fig. 1. Thirty-six CUD patients were randomly selected for rTMS therapy. Baseline craving severity as well as the rsfMRI data were obtained for each subject at baseline, and three time points during the treatment – 2 weeks (2w), 3 months (3m), and 6 months (6m) after the start of the treatment. All rsfMRI and anatomical images passed through preprocessing steps, and then two main analyses were performed on them: seed-based connectivity (SBC) of LDLPFC and ACC, as well as whole brain multi-voxel pattern analysis (MVPA). LDLPFC and ACC were chosen for SBC as they are among the well-reported loci of dysfunction in executive control network in CUD. In the next step, second-level group analysis was performed in the CONN toolbox to obtain connectomics-based predictors at each time point. The significant cluster of voxels which had the largest correlation between the connectivity changes and craving changes were identified. We employed leave-one-subject-out cross-validation in our study to increase the generalizability of the results. The two neuromarkers along with the baseline craving scores were entered into a linear ordinary least square model to predict the changes in the craving scores. The performance of the predictive models was assessed using adjusted R-squared and mean absolute error.

1) Baseline craving severity:

The baseline severity of clinical condition of neuropsychiatric patients is a conventional predictor of individual responsivity to different treatments in psychiatric disorders ^25,32,31,36. Therefore, we first examined the correlational relationship between baseline VAS and CCQ-N and the changes in these scores at three time points during the treatment.

2) Seed-based analysis:

Using Seed-to-voxel Bivariate Correlation (SBC) analysis in CONN ³⁷, the correlation maps of left DLPFC as well as ACC with the rest of the brain voxels were obtained. Specifically, the first-level correlation maps of the seed are produced by computing the Pearson’s correlation coefficients between the BOLD time course of the seed and the time courses of all other brain voxels. The resultant correlation coefficients – r-maps – were then converted to z-scores using Fisher transformation, and submitted to the second-level General Linear Model analysis. Here, we defined the average stimulation target cone mask ¹¹ as the seed mask for LDLPFC. This mask is obtained from averaging over individual stimulation targets, with the maximum value in the average target location. This mask has a large number of overlapping voxels with the left DLPFC (Brodmann areas 9 and 46), the major locus of dysfunction in CUD ⁷. The mask of ACC – extracted from the Harvard-Oxford Cortical atlas (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Atlases) – was selected as the other seed in our SBC analysis.

Fisher transformed r-maps of the seed regions were used in a second-level analysis of covariance (ANCOVA) to regress the changes in DLPFC/ACC connectivity (pre minus post) based on the individual change in the clinical score (pre minus post condition) as a covariate of interest, controlling for the score at baseline (testing whether the correlation between connectivity-change and score-change differ from zero after controlling for the influence of baseline score value). The predictive model is compared to the null model using a standard likelihood ratio test. Results are estimated using height threshold uncorrected p < 0.01 and cluster-level threshold uncorrected p < 0.01.

To have a robust predictive model with the ability to be generalized to new cases, we performed a leave-one-subject-out cross-validation (LOSOCV) to extract the connectivity values of the seed with the largest significant cluster of voxels in the brain. In each iteration, the data from N-1 subjects were used to estimate the first largest cluster of voxels that has significant association with changes in the craving scores. Then, the average connectivity within this cluster was computed for the left-out subject, and this procedure was repeated for each participant. Doing so, the potential biases from selected voxels in the significant cluster is minimized ²⁵.

3) Multi-voxel pattern analysis:

In this part, an agnostic data-driven approach known as multi-voxel pattern analysis (MVPA) ³⁸ in CONN ³⁹ was used to extract the second set of predictors. The pairwise connectivity pattern between each voxel of the brain mask and all other voxels of the brain were computed, and then, the dimensionality of this multi-voxel pattern was reduced using principal component analysis (PCA). This was done to maximize the explained variability of the extracted patterns with a lower number of spatial components. Here, we selected only the first of four main components, and performed multivariate analysis to investigate the relationship between this component’s connectivity changes and the changes in the clinical scores, controlling for the baseline scores. The first largest cluster whose connectivity patterns were significantly correlated with craving changes was identified (height threshold uncorrected p < 0.01; cluster-level threshold uncorrected p < 0.01). The same LOSOCV procedure as in SBC analysis was used to extract subject-specific cluster mask and MVPA connectivity values, iteratively for each subject.

4) Statistical analyses:

To assess whether baseline craving score predicted change in the craving score, we used independent linear least square models using baseline score values as predictors. To test whether functional connectivity changes from SBC and MVPA analysis predicted the changes in craving scores (pre minus post), we ran linear regression models using these connectomics measures together with baseline score as predictors.

For all predictive models, we used statistical ordinary least squares (OLS) model in python. To numerically assess the quality of the predictive models, we used R² values. These values represent the explained percent variance in clinical score changes. Here, we calculated Adjusted R² (Adj. R²) which minimizes the biases due to model selection, and helps to better compare the model with different numbers of the degrees of freedom and predictors ⁴⁰. The mean absolute error (MAE) represents the difference between the observed and predicted values, i.e., craving changes. In deciding between different models, the model with higher R² and lower MAE is more favorable. Model performance was also assessed with maximum value of residuals. The maximum residual represents the maximum deviation from the observed score among the predicted values by the model.

LOSOCV of SBC and MVPA correlation analysis were performed in MATLAB using “spm_crossvalidation” code provided in the CONN toolbox. All other statistics were performed in python using statsmodels module version 0.14.0.

Ethics approval

All the procedures were approved by the institutional Ethics Research Committee (CEI/C/070/2016) and registered at ClinicalTrials.gov (NCT02986438) ⁴.

A. Clinical assessments

Raincloud plots for distribution of clinical scores at different time points are demonstrated in Figure 2. The t-test analysis for comparing the mean score values between each of the two time points (baseline vs 2w/3m/6m) are given in Table 2. It is evident that the two scores have significantly different means between the two conditions (p-value < 0.05), except for VAS after 6 months.

Table 2. T-test results for comparing the mean values of the two scores between two time points. VAS: visual analog scale; CCQ: cocaine craving questionnaire; 2w: 2 weeks; 3m: 3 months; 6m: 6 months.

T-test Results	Statistic; p-value
Score/Time Point	2w vs bs	3m vs bs	6m vs bs
VAS	3.19; 0.00	3.96; 0.00	0.75; 0.45
CCQ-NOW	2.37; 0.02	3.08; 0.00	2.12; 0.04

B. Baseline craving severity

Pearson’s correlation values between baseline VAS and CCQ-N and the changes in these scores at three time points during the treatment are presented in Figure 3. The results indicate the highly linear relationship between baseline and changes in scores, especially for scores at 3 months after the start of the treatment. In this regard, we hypothesized that the baseline craving severity can explain the variability in this score, and used it as one of the predictors in our response prediction model.

C. Resting-state functional magnetic resonance imaging

1) LDLPFC/ACC seed‑based correlation analysis

The group-level connectivity maps between left DLPFC and ACC, and the rest of the brain regions is illustrated in Figure 4. The seed-to-voxel analysis of left DLPFC/ACC resulted in multiple clusters with significant differences in the correlations of LDLPFC/ACC between bs and 2w/3m/6m conditions that were associated with the changes in the VAS and CCQ-N scores, respectively (cluster-level uncorrected p < 0.01; voxel threshold uncorrected p < 0.01). The first largest significant clusters obtained from this analysis are presented in Table 3 and Figure 5.

Table 3. The largest significant cluster’s breakdown, resulted from second-level analysis of SBC.

	Significant Clusters
Score/Time Point		2 Weeks	3 Months	6 Months
VAS	MNI coordinates(x,y,z)	-14 -18 76	-14 -92 22	14 -16 46
	Number of voxels	722	1797	2971
	Including voxels of:	Precentral Gyrus Left and Right Postcentral Gyrus Left and Right Precuneous Cortex	Occipital Pole Left Lateral Occipital Cortex, inferior and superior division Left Occipital Fusiform Gyrus Left Cuneal Cortex Left	Precentral Gyrus Left and Right Postcentral Gyrus Left and Right Supplementary Motor Cortex- Left and Right Superior Frontal Gyrus Left Paracingulate Gyrus Left Middle Frontal Gyrus Left
CCQ-N	MNI coordinates(x,y,z)	4 -10 -4	0 -80 -32	50 -72 16
	Number of voxels	437	380	461
	Including voxels of:	Precuneous Cortex Thalamus Left and Right Cingulate Gyrus, posterior division Brain-Stem Pallidum Left Accumbens Left Putamen Left	Cerebellum Crus1, 2 Left and Right Cerebellum 4 5 Left Lingual Gyrus Left and Right Vermis 3,4,5,6,7 Cerebellum 6 Left and Right Brain-Stem Parahippocampal Gyrus, posterior division Left	Lateral Occipital Cortex, inferior division Right Lingual Gyrus Left and Right Lateral Occipital Cortex, superior division Right Middle Temporal Gyrus, temporooccipital part Right Cuneal Cortex Right Intracalcarine Cortex Right Cerebellum 6 Left and Right Vermis 6

2) MVPA analysis

The whole brain MVPA analysis resulted in multiple clusters with significant difference in connectivity of the first component between bs and 2w/3m/6m conditions that was associated with the changes in VAS and CCQ-N scores (voxel uncorrected threshold p < 0.01, cluster uncorrected threshold p < 0.01). The largest significant clusters are presented in Table 4 and Figure 6.

Table 4. The largest significant cluster’s breakdown, resulted from second-level analysis of MVPA.

Significant Clusters
Score/Time Point		2 Weeks	3 Months	6 Months
Percent Explained Variance		79.66% of GM and 73.76% of WM variance	75.48% of GM and 69.04% of WM variance	73.98% of GM and 67.95% of WM variance
VAS	MNI coordintes (x,y,z)	-14 -62 -20	36 -72 -14	36 -44 -42
	Number of voxels	284	861	226
	Including voxels of:	Cerebellum 6 Left and Right Cerebellum 4 5 Left Vermis 4, 5, 6, 7, 8, 9 Cerebellum 8 Left Cerebellum Crus1 Left	Occipital Pole Left and Right Occipital Fusiform Gyrus Left and Right Lingual Gyrus Left and Right Lateral Occipital Cortex, inferior division Left and Right Cerebellum 6 Right Cerebellum Crus1 Left Intracalcarine Cortex Right Temporal Occipital Fusiform Cortex Right	Cerebellum 10 Right Cerebellum 8 Left and Right Cerebellum 9 Left and Right Cerebellum Crus 1 2 Left and Right Cerebellum 7b left and Right Cerebellum 6 Right Vermis 8
CCQ-N	MNI coordinates (x,y,z)	24 -64 -34	-18 -50 -34	-20 -32 -50
	Number of voxels	923	491	207
	Including voxels of:	Cerebellum 8 Left and Right Cerebellum 6 Left and Right Lingual Gyrus Left and Right Cerebellum Crus1 2 Right Cerebellum 4 5 Left and Right Vermis 4, 5, 6, 8 Cerebellum 7b Right	Cerebellum 6, 8, 9, 10 Left Vermis 4, 5, 6, 7, 8, 9, 10 Cerebellum 8 Right Cerebellum Crus1 2 Right	Cerebellum 8, 9 Left

D. Prediction model for craving score changes:

Here, we used ordinary least square models for prediction of changes in the clinical scores between the baseline and 2w/3m/6m conditions. Specifically, for each of the VAS and CCQ-N scores, the connectivity from the SBC and MVPA analysis together with the baseline scores were used in a linear model to explain the variance in the score changes. The detailed percentage of the variance explained by each model for both scores is specified in Tables 5 and 6. Scatter plots for VAS values computed from the predictive model for VAS changes, and the scatter plots of the predicted and observed values are shown in Figure 7. The similar scatter plots for predictive model of CCQ changes are shown in Figure 8.

The baseline craving VAS score accounts for 58%, 95%, and 44% of the variance in the VAS changes at 2 weeks, 3 months, and 6 months after rTMS application, respectively (Table 5). The baseline craving CCQ-N score accounts for 27%, 61%, and 43% of the variance in CCQ-N changes at 2 weeks, 3 months, and 6 months after rTMS application, respectively (Table 6). As is evident, the combination of the fMRI connectomics measures considerably improves the model prediction compared to using only the baseline scores.

Table 5. Prediction models for changes in VAS scores using connectivity changes.

Time Point	2 Weeks		3 Months		6 Months
Predictor from changes in connectivities	Baseline score	Combined model (Baseline score + ACC_SBC+MVPA)	Baseline score	Combined model (Baseline score + ACC_SBC+MVPA)	Baseline score	Combined model (Baseline score + ACC_SBC+MVPA)
Adj. R²	0.58	0.66	0.95	0.97	0.44	0.85
MAE	0.95	0.90	0.39	0.39	3.06	1.24
Max Residual	5.07	4.48	2.85	1.52	6.23	3.39

Table 6. Prediction models for changes in CCQ-N scores using connectivity changes

Time Point	2 Weeks		3 Months		6 Months
predictor	Baseline score	Combined (Baseline score + DLPFC_SBC+MVPA)	Baseline score	Combined (Baseline score + DLPFC_SBC+MVPA)		Baseline score	Combined (Baseline score + DLPFC_SBC+MVPA)
Adj. R²	0.27	0.45	0.61	0.82	0.43		0.71
MAE	24.63	23.25	24.01	16.39	27.36		14.62
Max Residual	99.40	64.88	79.16	39.86	68.89		70.73

The aim of the current study was to investigate whether changes in resting state functional connectivity of brain regions implicated in CUD, could explain the changes in craving scores in cocaine users receiving the add-on rTMS therapy for short as well as long term periods. We used neuromarkers from rsfMRI technique with two seed-driven as well as data-driven approaches to predict response to rTMS treatment for CUD patients, and assessed whether the combination of these neuromarkers and baseline severity can predict the variability in craving better than only using baseline clinical scores. The predictions were performed using the clinical and neuroimaging data for three time points during the add-on rTMS treatment (2 weeks, 3 months, 6 months). Here, we considered two cocaine craving scores as clinical measures – VAS and CCQ-Now. First assessment of baseline scores showed high correlation between the changes in these scores and their baseline values. Therefore, the baseline scores were considered as one of the first predictors in our prediction models.

Baseline VAS accounted for 58%, 95%, and 44% of the variance in VAS changes after 2 weeks, 3 months, and 6 months of rTMS application, whereas the combination of rsfMRI neuromarkers and baseline VAS accounted for 66%, 97%, and 85% of the variance in clinical benefits after 2 weeks, 3 months, and 6 months, respectively. Likewise, baseline CCQ-N accounted for 27%, 61%, and 43% of the variance in CCQ-N changes after 2 weeks, 3 months, and 6 months rTMS, while rsfMRI neuromarkers together with baseline CCQ-N accounted for 45%, 82%, and 71% of the variance in craving reduction after 2 weeks, 3 months, and 6 months, respectively.

Our models were mainly successful in predicting the direction of craving changes, and adding the neuromarkers improved their performance considerably, especially after 3 months. Overall, the proposed modeling approach had better performance (higher adj. R², lower MAE, and confidence interval of predicted values) for prediction of craving with the VAS. The proposed models showed some limitations in predicting the VAS changes for very small or zero values. However, the residuals in these cases were small (in the interval [0,2]). To compute the final scores obtained from the model we clipped the VAS values to [0,10] to prevent unacceptable negative outputs.

A. SBC results

In our study, we used resting-state fMRI predictors obtained from two approaches: seed-based and data-driven methods. Seed-driven connectivity changes in regions implicated in CUD and craving, were chosen as one of the predictors in our prediction model. Frequent evidence from literature on dysfunction of ACC and left DLPFC in craving as well as CUD ⁸, ¹⁰^,^41–44, motivated us to consider seed-based connectivity changes of these two regions as one of the connectomics-based predictors. We used ACC seed-based connectivity for VAS prediction. Multiple studies have consistently reported the structural and functional changes in this region in CUD ^45,46 and also its relationship with VAS ⁴⁷. Results showed that the connectivity of ACC with precentral gyrus, occipital cortex, and pre/post-central gyrus predicted greater reduction in VAS for 2w, 3m, and 6m after rTMS application, respectively.

As a more accurate technique with detailed scales to assess craving, we chose CCQ. Seed-based connectivity of LDLPFC were used for CCQ-N prediction. We hypothesized that this score would be more related to the connectivity of DLPFC, the well-recognized loci of dysfunction in CUD. Connectivity of LDLPFC with the cluster including precuneus, thalamus, nucleus accumbens, and posterior cingulate gyrus predicted greater response to rTMS after 2 weeks. The corresponding clusters after 3, and 6 months contained regions from left and right cerebellum, and lateral occipital, respectively. The findings from these significant clusters are consistent with evidence from prior studies which indicate the role of regions from frontal lobe ⁴⁸, nucleus accumbens ¹²^,⁴³, occipital-temporal ⁴⁸^,⁴³^,⁴⁹, and cerebellum ⁴¹^,⁵⁰^,⁵¹ in craving as well as cocaine addiction.

B. MVPA results

As second connectomics neuromarker we employed MVPA analysis to examine the whole brain voxels without any a priori seed. This analysis considers the connectivity between all voxel pairs in the brain, and interestingly, the results for both craving scores revealed the major component identified by this analysis in cerebellum and occipital areas. Previous studies implicated the role of cerebellum and vermis parts in CUD/craving ⁴¹^,⁵⁰^,⁵¹, and our agnostic approach confirmed these results. Moreover, the revelation of clusters in cerebellum were common for both craving scores (VAS and CCQ-N) for all time points, which indicates the predictive power of neuromarkers from these less noted.

Several functional imaging studies suggested the role of cerebellum in stimulant-associated behaviors, and indicated the relationship between the cerebellar activity during cocaine craving, and recall of cocaine-use experience. In one study BOLD fMRI analysis in cocaine users showed that cocaine-related cues selectively induce BOLD activation in vermis lobules 2-3 and 8-9 ⁵⁰. In this context, our study can be regarded as evidence for the locations in cerebellum and vermis that have the strongest association with the changes in craving. Our agnostic MVPA showed the association between connectivity changes in cerebellum 4-6 left and vermis 4-9, and VAS changes, 2 weeks after rTMS. Similarly right parts of cerebellum 6-10, left parts of cerebellum 7-9, and vermis 8 were identified after 6 months (Table 4). For CCQ-N changes, surprisingly all the identified regions from MVPA are commonly localized in cerebellum 8 and 9 left for all three time points. Vermis 4-8 were also identified in 2-week and 3-month results.

C. Prediction

Several studies provided evidence on the usefulness of neuromarkers in predicting future responsivity to different treatments. Such neuroimaging studies differ based on their purpose. Some studies consider only the correlational relationships as prediction, while others try to use different modeling approaches to link the neuroimaging and behavioral data. There are several studies that incorporated neuromarkers for prediction of rTMS outcomes in addiction ⁵², Schizophrenia ⁵³, Alzheimer's disease ⁵⁴, and depression ³²^,³¹^,⁵⁵^,⁵⁶. There are also studies that used neuromarkers to predict different types of treatments in brain disorders such as depression ²⁶^,²⁷, schizophrenia ²⁸, dyslexia ²⁹, generalized anxiety ³⁰. As an example, using the combination of SBC, MVPA, and diffusion MRI neuromarkers, Whitfield-Gabrieli et al. ²⁵ could predict the treatment response to social anxiety disorder. Recently P. Morris et al. ⁵⁷ demonstrated that SBC analysis of resting state fMRI data can provide mechanistic prediction for future changes in sedentary behavior.

Although most of the studies model the relationship between neuroimaging data and outcome measures, the findings mainly fail when applied to new datasets. The ability to being generalized to new cases is one of the main features in choosing neuromarkers for clinical practice. Due to our small sample size, we used leave-one-subject-out cross-validation to ensure about the generalizability of our predictive models. Specifically, in each approach, we extracted the connectivity measures for each individual patient from the cluster mask obtained from group analysis performed for the rest of the patients. Such procedure resulted in robust models in previous studies ²⁵^,⁵⁷, and doing so, we avoid non-independence in our data analysis ⁵⁸ and prevent the problem of double dipping in neuroscience ⁵⁹.

We hypothesized that the changes in neuromarkers together with rTMS intervention would be reflected in craving level. Therefore, in our main set of analyses we considered the connectivity “changes” in the predictive model. However, from clinical point of view, it is worth to assess whether the “baseline” brain condition of patients can predict future treatment response. In this regard, in our second set of analyses, we tried to assess the power of the connectomics measures from baseline fMRI data in outcome prediction (Table S1 & S2). Our findings show that baseline functional connectivity is also predictive for both VAS and CCQ-N, with comparable adjusted-R² values to the model with changes in connectivity. The results were slightly better in VAS prediction model for 2w and 3m (higher adj. R²and lower MAE), but generally lower in CCQ-N prediction model. This may be explained by the differences between craving measurements, as the VAS is a more visual cue whereas the CCQ-N is a questionnaire.

In order to predict the outcome for long term treatment, one should consider the whole course of changes in clinical and connectivity during the treatment. Here, we limited our analysis in each time point to the comparison between the scores in that time point and baseline. Despite our small sample size, the employed method showed that the linear model of combination of clinical scores and connectomics can be predictive in rTMS treatment. However, to achieve a more comprehensive model for this purpose, it is suggested to employ neuroimage and behavioral data from different time points all together into the model. Incorporating data from other neuroimage modalities as well as various behavioral scores may also improve the predictions. In addition, using nonlinear models of changes such as exponential decay ⁵⁶ or machine learning-based models ³¹^,⁶⁰ may be more beneficial in prediction of outcomes.

We assessed the predictive power of our models using adjusted R², mean square of errors, and the maximum residual value. The R² value compares the unexplained variance (variance of the model’s error) with the total variance of the data. We used this unitless and understandable measure so that our models would be comparable to other studies; however, this value have some limitation. This measure is so sensitive to outlier and extreme values, and as a result, different models with different configuration of predicted values can end to similar R² values. This fact is evident in plots for VAS prediction at 3m in Figure 7. We see that despite different predictions for the same observed value, the R² value of the combined model is close to 1. Due to this limitation, we used MAE as other metric for evaluation of our models. This metric shows the differences in the units of clinical metrics and makes the comparison and interpretation of models easier. Maximum residual value is also used to guide us in quantifying models. For models with high R² values such as the one mentioned earlier, it is evident that despite the different predictions for the same VAS values, the maximum residual is much smaller than the total variation that the VAS metric can have (maximum residual of 1.52 in range of 0-10).

In the present study, we aimed to develop neuromarkers from rsfMRI for predicting the rTMS treatment response. There are some specifications in our study that come as follows:

We hypothesized that the heterogeneity in patients’ brains is the substrate for heterogeneity in their response to treatment, and defined our predictors based on brain biomarkers. Therefore, our method differs from other studies that try to predict the connectivity changes from changes in behavior.

Although we tried to link the neuroimaging data to behavioral data for rTMS treatment, one should bear in mind that this treatment was an add-on treatment to the conventional pharmacological therapies that CUD patients received. Therefore, there can be pharmacological and other effects.

Here, we showed that the individual differences in functional connectivity changes from SBC and MVPA analysis can predict future changes in craving in cocaine users receiving add-on rTMS treatment. This study provide evidence for the power of neuromarkers in craving predictions, and highlights the idea that neurmarkers may offer improvements in precision medicine approaches.

Consent to participate

The study was approved by the local ethics committee and performed at the Instituto Nacional de Psiquiatría ‘Ramón de la Fuente Muñiz’ in Mexico City, Mexico. All participants gave written informed consent before participation in any study procedures, all of which conformed to the Declaration of Helsinki for research involving human subjects.

Competing interest

The authors declare no competing interest.

Data availability

All raw data are freely available in BIDS format: https://openneuro.org/datasets/ ds003037/versions/1.0.0.

Author contributions

N.G. Conceptualization, design, analysis, interpretation of data, manuscript writing, E.A.G. Conceptualization, interpretation of data, substantial revision, H.S.Z. Conceptualization, interpretation of data, substantial revision.

Gay A, Cabe J, De Chazeron I, et al. Repetitive Transcranial Magnetic Stimulation (rTMS) as a Promising Treatment for Craving in Stimulant Drugs and Behavioral Addiction: A Meta-Analysis. J Clin Med. 2022;11(3). doi:10.3390/jcm11030624
Kilts CD, Schweitzer JB, Quinn CK, et al. Neural activity related to drug craving in cocaine addiction. Arch Gen Psychiatry. 2001;58(4):334–341. doi:10.1001/archpsyc.58.4.334
Fryer JR. RG, Heaton PS, Levitt SD, Murphy KM. Measuring Ceack Cocaine and its Impact. Econ Inq. 2013;51(3):1651–1681. doi:https://doi.org/10.1111/j.1465-7295.2012.00506.x
Angeles-Valdez D, Rasgado-Toledo J, Issa-Garcia V, et al. The Mexican magnetic resonance imaging dataset of patients with cocaine use disorder: SUDMEX CONN. Sci Data. 2022;9(1):1–10. doi:10.1038/s41597-022-01251-3
Camprodon JA, Martínez-Raga J, Alonso-Alonso M, Shih MC, Pascual-Leone A. One session of high frequency repetitive transcranial magnetic stimulation (rTMS) to the right prefrontal cortex transiently reduces cocaine craving. Drug Alcohol Depend. 2007;86(1):91–94. doi:https://doi.org/10.1016/j.drugalcdep.2006.06.002
Politi E, Fauci E, Santoro A, Smeraldi E. Daily Sessions of Transcranial Magnetic Stimulation to the Left Prefrontal Cortex Gradually Reduce Cocaine Craving. Am J Addict. 2008;17(4):345–346. doi:https://doi.org/10.1080/10550490802139283
Hanlon CA, Dowdle LT, Scott Henderson J. Modulating neural circuits with transcranial magnetic stimulation: Implications for addiction treatment development. Pharmacol Rev. 2018;70(3):661–683. doi:10.1124/pr.116.013649
Hanlon CA, Dowdle LT, Austelle CW, et al. What goes up, can come down: Novel brain stimulation paradigms may attenuate craving and craving-related neural circuitry in substance dependent individuals. Brain Res. 2015;1628(Pt A):199–209. doi:10.1016/j.brainres.2015.02.053
Hanlon CA, Dowdle LT, Moss H, Canterberry M, George MS. Mobilization of Medial and Lateral Frontal-Striatal Circuits in Cocaine Users and Controls: An Interleaved TMS/BOLD Functional Connectivity Study. Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2016;41(13):3032–3041. doi:10.1038/npp.2016.114
Terraneo A, Leggio L, Saladini M, Ermani M, Bonci A, Gallimberti L. Transcranial magnetic stimulation of dorsolateral prefrontal cortex reduces cocaine use: A pilot study. Eur Neuropsychopharmacol J Eur Coll Neuropsychopharmacol. 2016;26(1):37–44. doi:10.1016/j.euroneuro.2015.11.011
Garza-Villarreal EA, Alcala-Lozano R, Fernandez-Lozano S, et al. Clinical and Functional Connectivity Outcomes of 5-Hz Repetitive Transcranial Magnetic Stimulation as an Add-on Treatment in Cocaine Use Disorder: A Double-Blind Randomized Controlled Trial. Biol Psychiatry Cogn Neurosci Neuroimaging. 2021;6(7):745–757. doi:10.1016/j.bpsc.2021.01.003
Bolloni C, Panella R, Pedetti M, et al. Bilateral Transcranial Magnetic Stimulation of the Prefrontal Cortex Reduces Cocaine Intake: A Pilot Study. Front psychiatry. 2016;7:133. doi:10.3389/fpsyt.2016.00133
Steele VR, Maxwell AM, Ross TJ, Stein EA, Salmeron BJ. Accelerated Intermittent Theta-Burst Stimulation as a Treatment for Cocaine Use Disorder: A Proof-of-Concept Study. Front Neurosci. 2019;13:1147. doi:10.3389/fnins.2019.01147
Sanna A, Fattore L, Badas P, Corona G, Cocco V, Diana M. Intermittent Theta Burst Stimulation of the Prefrontal Cortex in Cocaine Use Disorder: A Pilot Study. Front Neurosci. 2019;13:765. doi:10.3389/fnins.2019.00765
Young JR, Galla JT, Appelbaum LG. Transcranial Magnetic Stimulation Treatment for Smoking Cessation: An Introduction for Primary Care Clinicians. Am J Med. 2021;134(11):1339–1343. doi:https://doi.org/10.1016/j.amjmed.2021.06.037
Rosenberg H. Clinical and laboratory assessment of the subjective experience of drug craving. Clin Psychol Rev. 2009;29(6):519–534. doi:https://doi.org/10.1016/j.cpr.2009.06.002
Tiffany ST, Wray JM. The clinical significance of drug craving. Ann N Y Acad Sci. 2012;1248(1):1–17. doi:https://doi.org/10.1111/j.1749-6632.2011.06298.x
Paliwal P, Hyman SM, Sinha R. Craving predicts time to cocaine relapse: Further validation of the Now and Brief versions of the cocaine craving questionnaire. Drug Alcohol Depend. 2008;93(3):252–259. doi:https://doi.org/10.1016/j.drugalcdep.2007.10.002
Sinha R, Garcia M, Paliwal P, Kreek MJ, Rounsaville BJ. Stress-Induced Cocaine Craving and Hypothalamic-Pituitary-Adrenal Responses Are Predictive of Cocaine Relapse Outcomes. Arch Gen Psychiatry. 2006;63(3):324–331. doi:10.1001/archpsyc.63.3.324
Heinonen J, Numminen J, Hlushchuk Y, Antell H, Taatila V, Suomala J. Default mode and executive networks areas: Association with the serial order in divergent thinking. PLoS One. 2016;11(9):1–16. doi:10.1371/journal.pone.0162234
Rasgado-toledo J, Issa-garcia V, Alcalá-lozano R, Garza-villarreal EA. Cortical and subcortical connections change after repetitive transcranial magnetic stimulation therapy in cocaine use disorder and predict clinical outcome. 2023.
Hanlon CA, Dufault DL, Wesley MJ, Porrino LJ. Elevated gray and white matter densities in cocaine abstainers compared to current users. Psychopharmacology (Berl). 2011;218(4):681–692. doi:10.1007/s00213-011-2360-y
Ernst M, Ph D, London ED, Ph D. Neural Systems and Cue-Induced Cocaine Craving. 1999;(01).
Martinez D, Hu M, Zangen A, Levin F, Foltin R, Nunes E. A Pilot Study of Transcranial Magnetic Stimulation of the Medial Prefrontal and Cingulate Cortices and Cocaine Self-Administration. Brain Stimul Basic, Transl Clin Res Neuromodulation. 2019;12(2):555. doi:10.1016/j.brs.2018.12.834
Whitfield-Gabrieli S, Ghosh SS, Nieto-Castanon A, et al. Brain connectomics predict response to treatment in social anxiety disorder. Mol Psychiatry. 2016;21(5):680–685. doi:10.1038/mp.2015.109
Fu CHY, Steiner H, Costafreda SG. Predictive neural biomarkers of clinical response in depression: a meta-analysis of functional and structural neuroimaging studies of pharmacological and psychological therapies. Neurobiol Dis. 2013;52:75–83. doi:10.1016/j.nbd.2012.05.008
Siegle GJ, Carter CS, Thase ME. Use of FMRI to predict recovery from unipolar depression with cognitive behavior therapy. Am J Psychiatry. 2006;163(4):735–738. doi:10.1176/ajp.2006.163.4.735
Kumari V, Peters ER, Fannon D, et al. Dorsolateral prefrontal cortex activity predicts responsiveness to cognitive-behavioral therapy in schizophrenia. Biol Psychiatry. 2009;66(6):594–602. doi:10.1016/j.biopsych.2009.04.036
Hoeft F, McCandliss BD, Black JM, et al. Neural systems predicting long-term outcome in dyslexia. Proc Natl Acad Sci U S A. 2011;108(1):361–366. doi:10.1073/pnas.1008950108
Whalen PJ, Johnstone T, Somerville LH, et al. A functional magnetic resonance imaging predictor of treatment response to venlafaxine in generalized anxiety disorder. Biol Psychiatry. 2008;63(9):858–863. doi:10.1016/j.biopsych.2007.08.019
Cash RFH, Cocchi L, Anderson R, et al. A multivariate neuroimaging biomarker of individual outcome to transcranial magnetic stimulation in depression. Hum Brain Mapp. 2019;40(16):4618–4629. doi:10.1002/hbm.24725
Harika-Germaneau G, Wassouf I, Le Tutour T, et al. Baseline Clinical and Neuroimaging Biomarkers of Treatment Response to High-Frequency rTMS Over the Left DLPFC for Resistant Depression. Front Psychiatry. 2022;13(May):1–12. doi:10.3389/fpsyt.2022.894473
Koutsouleris N, Wobrock T, Guse B, et al. Predicting response to repetitive transcranial magnetic stimulation in patients with schizophrenia using structural magnetic resonance imaging: A multisite machine learning analysis. Schizophr Bull. 2018;44(5):1021–1034. doi:10.1093/schbul/sbx114
Eduardo A. Garza-Villarreal. Clinical and functional connectivity outcomes of 5-Hz repeated transcranial magnetic stimulation as an add-on treatment in cocaine use disorder: a double-blind randomized controlled trial. medRxiv. 2020;(165):1–13.
Andréa L. Hobkirka, Ryan P. Bella, Amanda V. Utevskyb, Scott Huettelb and CSM. Reward and executive control network resting-state functional connectivity is associated with impulsivity during reward-based decision making for cocaine users. Physiol Behav. 2016;176(1):139–148. doi:10.1016/j.drugalcdep.2018.09.013.Reward
Chen CH, Ridler K, Suckling J, et al. Brain Imaging Correlates of Depressive Symptom Severity and Predictors of Symptom Improvement After Antidepressant Treatment. Biol Psychiatry. 2007;62(5):407–414. doi:10.1016/j.biopsych.2006.09.018
Alfonso Nieto-Castanon. Handbook of Functional Connectivity Magnetic Resonance Imaging Methods in CONN. Boston, MA: Hilbert Press; 2020.
Norman KA, Polyn SM, Detre GJ, Haxby J V. Beyond mind-reading: multi-voxel pattern analysis of fMRI data. Trends Cogn Sci. 2006;10(9):424–430. doi:10.1016/j.tics.2006.07.005
Nieto-Castanon A. Brain-wide connectome inferences using functional connectivity MultiVariate Pattern Analyses (fc-MVPA). PLOS Comput Biol. 2022;18(11):1–28. doi:10.1371/journal.pcbi.1010634
Haslwanter T. An Introduction to Statistics with Python. 1st ed. Springer Cham; 2016. doi:https://doi.org/10.1007/978-3-319-28316-6
Bonson KR, Grant SJ, Contoreggi CS, et al. Neural Systems and Cue-Induced Cocaine Craving. Neuropsychopharmacology. 2002;26(3):376–386. doi:10.1016/S0893-133X(01)00371-2
Bellamoli E, Manganotti P, Schwartz RP, Rimondo C, Gomma M, Serpelloni G. rTMS in the treatment of drug addiction: an update about human studies. Behav Neurol. 2014;2014:815215. doi:10.1155/2014/815215
Breiter HC, Gollub RL, Weisskoff RM, et al. Acute effects of cocaine on human brain activity and emotion. Neuron. 1997;19(3):591–611. doi:10.1016/s0896-6273(00)80374-8
Prisciandaro JJ, McRae-Clark AL, Myrick H, Henderson S, Brady KT. Brain activation to cocaine cues and motivation/treatment status. Addict Biol. 2014;19(2):240–249. doi:10.1111/j.1369-1600.2012.00446.x
Camchong J, MacDonald AW 3rd, Nelson B, et al. Frontal hyperconnectivity related to discounting and reversal learning in cocaine subjects. Biol Psychiatry. 2011;69(11):1117–1123. doi:10.1016/j.biopsych.2011.01.008
Pando-Naude V, Toxto S, Fernandez-Lozano S, Parsons CE, Alcauter S, Garza-Villarreal EA. Gray and white matter morphology in substance use disorders: a neuroimaging systematic review and meta-analysis. Transl Psychiatry. 2021;11(1):29. doi:10.1038/s41398-020-01128-2
Zhao K, Fonzo GA, Xie H, et al. A generalizable functional connectivity signature characterizes brain dysfunction and links to rTMS treatment response in cocaine use disorder. medRxiv Prepr Serv Heal Sci. May 2023. doi:10.1101/2023.04.21.23288948
Hester R, Garavan H. Executive Dysfunction in Cocaine Addiction: Evidence for Discordant Frontal, Cingulate, and Cerebellar Activity. J Neurosci. 2004;24(49):11017–11022. doi:10.1523/JNEUROSCI.3321-04.2004
Müller CP, De Souza Silva MA, Huston JP. Double dissociating effects of sensory stimulation and cocaine on serotonin activity in the occipital and temporal cortices. Neuropharmacology. 2007;52(3):854–862. doi:10.1016/j.neuropharm.2006.10.002
Anderson CM, Maas LC, Frederick B deB, et al. Cerebellar Vermis Involvement in Cocaine-Related Behaviors. Neuropsychopharmacology. 2006;31(6):1318–1326. doi:10.1038/sj.npp.1300937
Miquel M, Gil-Miravet I, Guarque-Chabrera J. The Cerebellum on Cocaine. Front Syst Neurosci. 2020;14. doi:10.3389/fnsys.2020.586574
Noël X, Sferrazza R, Van Der Linden M, et al. Contribution of frontal cerebral blood flow measured by (99m)Tc-Bicisate spect and executive function deficits to predicting treatment outcome in alcohol-dependent patients. Alcohol Alcohol. 2002;37(4):347–354. doi:10.1093/alcalc/37.4.347
Koutsouleris N, Wobrock T, Guse B, et al. Predicting Response to Repetitive Transcranial Magnetic Stimulation in Patients With Schizophrenia Using Structural Magnetic Resonance Imaging: A Multisite Machine Learning Analysis. Schizophr Bull. 2017;44(5):1021–1034. doi:10.1093/schbul/sbx114
Zhang S, Liu L, Zhang L, et al. Evaluating the treatment outcomes of repetitive transcranial magnetic stimulation in patients with moderate-to-severe Alzheimer’s disease. Front Aging Neurosci. 2022;14:1070535. doi:10.3389/fnagi.2022.1070535
Ge R, Humaira A, Gregory E, et al. Predictive Value of Acute Neuroplastic Response to rTMS in Treatment Outcome in Depression: A Concurrent TMS-fMRI Trial. Am J Psychiatry. 2022;179(7):500–508. doi:10.1176/appi.ajp.21050541
Berlow YA, Zandvakili A, Brennan MKC, Williams LM, Price LH, Philip NS. Modeling the antidepressant treatment response to transcranial magnetic stimulation using an exponential decay function. Sci Rep. 2023;13(1):1–9. doi:10.1038/s41598-023-33599-w
Morris TP, Kucyi A, Anteraper SA, et al. Resting state functional connectivity provides mechanistic predictions of future changes in sedentary behavior. Sci Rep. 2022;12(1):1–11. doi:10.1038/s41598-021-04738-y
Esterman M, Tamber-Rosenau BJ, Chiu YC, Yantis S. Avoiding non-independence in fMRI data analysis: leave one subject out. Neuroimage. 2010;50(2):572–576. doi:10.1016/j.neuroimage.2009.10.092
Kriegeskorte N, Simmons WK, Bellgowan PSF, Baker CI. Circular analysis in systems neuroscience: the dangers of double dipping. Nat Neurosci. 2009;12(5):535–540. doi:10.1038/nn.2303
Chekroud AM, Bondar J, Delgadillo J, et al. The promise of machine learning in predicting treatment outcomes in psychiatry. World Psychiatry. 2021;20(2):154–170. doi:10.1002/wps.20882

No competing interests reported.

Ghazi.Supplementarymaterials.docx

Brain Connectomics Markers for Response Prediction to Transcranial Magnetic Stimulation in Cocaine Use Disorder

Status:

Version 1

Abstract

Figures

I. INTRODUCTION

II. MATERIALS AND METHODS

III. RESULTS

A. Clinical assessments

B. Baseline craving severity

C. Resting-state functional magnetic resonance imaging

D. Prediction model for craving score changes:

IV. DISCUSSION

Declarations

Consent to participate

References

Additional Declarations

Supplementary Files

Status:

Version 1