Stratification Algorithm for Repetitive TMS in Stroke (START): Results from an Exploratory Crossover Study

Abstract

The role of contralesional motor cortices in paretic upper extremity (UE) motor recovery following stroke varies based on available structural reserve. However, an optimal measure of the reserve to stratify patients for different contralesional brain stimulation remains unknown. This study aimed to establish severity criteria distinguishing which patients benefit more from inhibitory contralesional motor cortex (M1) stimulation versus facilitatory contralesional dorsal premotor cortex (cPMd) stimulation. Twenty-four chronic stroke participants underwent three repetitive transcranial magnetic stimulation (rTMS) sessions: inhibitory 1Hz rTMS to contralesional M1, facilitatory 5Hz rTMS to cPMd, and sham rTMS. Motor performance on a reaching task (RT) was assessed pre- and post-stimulation. Baseline assessments included UE Fugl-Meyer (UEFM), corticospinal integrity (fractional anisotropy), and motor evoked potentials (MEPs). Classification and Regression Tree (CART) analysis identified UEFM 42 as the threshold distinguishing patients who improved with cM1 inhibition versus cPMd facilitation rTMS, with 91.6% and 83.3% accuracy, respectively. Participants with UEFM>42 showed greater RT gains with inhibitory rTMS than more severely impaired individuals (p=0.06), whereas those with UEFM≤42 demonstrated greater RT gains with facilitatory cPMd rTMS than sham (p=0.003). Less-severe participants had larger increases in ipsilesional MEPs following inhibitory rTMS (p=0.007), whereas more-severe (UEFM≤42) MEP-absent participants had larger reductions in interhemispheric inhibition (IHI) following facilitatory cPMd rTMS (p=0.028). Our findings support the bimodal theory and introduce the START (Stratification Algorithm for rTMS) framework, utilizing clinical impairment and white matter integrity to stratify response. While promising, the START algorithm requires further validation in larger samples to develop targeted and effective neuromodulation treatments.

1. Introduction

The role of the contralesional motor cortices in motor functional recovery of the paretic upper extremity (UE) after stroke remains a topic of considerable debate. Early studies have suggested that contralesional motor cortices exert excessive interhemispheric inhibition (IHI) on ipsilesional motor output via transcallosal pathways [1, 2]. This led to the development of various non-invasive brain stimulation approaches, including 1Hz repetitive transcranial magnetic stimulation (rTMS) and cathodal transcranial direct current stimulation (tDCS), aimed at reducing contralesional motor cortical excitability to mitigate excessive IHI effect [3, 4]. While initial studies reported promising results, later randomized controlled trials (RCTs) failed to demonstrate a significant advantage of pairing inhibitory contralesional stimulation with UE rehabilitation over UE rehabilitation delivered alone [5, 6].

The bimodal-balance recovery hypothesis offers a potential explanation for these mixed results. According to this theory, the influence of contralesional motor cortices on ipsilesional motor output varies based on the brain’s structural reserve [7]. The structural reserve is defined as “the quantity of strategic neural pathways and relays that are spared by the lesion and (that) can reallocate previous or outsource new functions [7].” According to this theory, when the structural reserve is high, contralesional motor cortices exert a predominantly inhibitory effect (i.e., stronger IHI). On the other hand, when the structural reserve is low, contralesional motor cortices may assume a more compensatory role to support paretic UE movement, such as via reduced IHI and/or unmasked alternate motor pathways [7]. The bimodal theory suggests that structural reserve can be quantified using clinical, anatomic, and functional measures of motor connectivity [7].

Several lines of evidence support the bimodal hypothesis. One study has shown that inhibitory tDCS of the contralesional motor cortex produces greater UE motor gains in stroke participants with milder corticospinal tract (CST) injury than in those with severe CST injury, indicating stronger IHI in the presence of milder damage [8]. Another study has found an inverted U-relationship between IHI and post-stroke UE motor impairment, characterized using the UE Fugl-Meyer (UEFM) scale (range 0 to 66, best), with a change point at UEFM 43 (95% confidence interval 40-46) [9, 10]. Among participants with UEFM > 43 (i.e., less impaired participants), greater motor impairment was associated with stronger IHI. Conversely, among participants with UEFM < 43 (i.e., more impaired participants), greater motor impairment was associated with reduced IHI. Furthermore, IHI decreased with clinically relevant rehabilitation motor gains in more impaired participants, whereas this effect was less pronounced in mild/moderately impaired participants (UEFM > 43). This suggests that with severe damage, contralesional motor cortices exert a reduced inhibitory effect to support paretic UE movement [10].

Emerging evidence suggests that the contralesional dorsal premotor cortex (cPMd) may be particularly well-positioned to support paretic UE movement in those with severe injury based on its rich callosal and bilateral descending motor projections [11]. Temporary inactivation of cPMd using TMS disrupts ongoing paretic UE movement more significantly in severely impaired patients than inactivation of either ipsilesional or contralesional primary motor cortex (M1) [11-14]. Additionally, proximal UE motor gains in moderate to severely impaired stroke survivors are associated with increased functional connectivity between ipsilesional M1 and cPMd [15]. These findings have led to the exploration of facilitatory stimulation of cPMd using rTMS and tDCS as potential brain stimulation therapy to promote paretic UE movement in severely impaired stroke survivors [14, 16].

Overall, the bimodal theory challenges the traditional 'one-size-fits-all' approach to brain stimulation. Instead, the theory posits that the available structural reserve should guide whether brain stimulation is used to inhibit or facilitate contralesional motor cortices [7]. However, an optimal measure of structural reserve to effectively stratify patients for different stimulation approaches remains unidentified. Therefore, we aimed to identify severity criteria that differentiate patients who achieve greater paretic UE gains with inhibitory contralesional stimulation from those who exhibit greater gains with facilitatory cPMd stimulation. In a randomized, sham-controlled crossover design, chronic stroke survivors with varying degrees of UE motor impairment underwent single sessions of 1Hz inhibitory rTMS to the contralesional M1, 5Hz facilitatory rTMS to the cPMd, and sham rTMS, with at least one week between sessions. Inhibitory 1Hz rTMS was applied to the ipsilesional motor hotspot in M1, while facilitatory 5Hz rTMS was targeted at a site 2.5 cm anterior to the contralesional motor hotspot, corresponding to the anatomical location of the cPMd [17]. At baseline, participants underwent assessments of UE motor impairment (UEFM) and T1-weighted MRI and Diffusion Tensor Imaging (DTI) to characterize lesion location and anatomic CST white matter integrity. We collected UE reaching time (RT) measures to assess motor gains, along with TMS parameters to evaluate IHI and ipsilesional excitability before and after each rTMS session. We hypothesized that clinical motor impairment (UEFM), CST integrity and lesion location, and residual ipsilesional physiology—either individually or in combination—would accurately predict which participants show greater RT improvements with 1Hz inhibitory rTMS of contralesional M1 versus 5Hz facilitatory rTMS of cPMd. We anticipated that participants with less severe impairment and damage would demonstrate greater paretic RT gains with 1Hz inhibitory rTMS of contralesional M1, while those with more severe impairment and damage would exhibit greater paretic RT gains with 5Hz facilitatory rTMS of cPMd. Ultimately, we would establish severity criteria to classify patients for targeted and effective contralesional rTMS treatments, guided by what we refer to as the START (Stratification Algorithm for rTMS).

2. Materials and Methods

2.1. Participants

Inclusion criteria required participants to be 21-80 years old, at least 6 months post-first clinically symptomatic (ischemic or hemorrhagic, cortical and/or subcortical) stroke, and to have sufficient shoulder and elbow flexion/extension to complete an upper limb reaching task (minimal motor criterion). Eligible participants were required to have reported difficulty in using the paretic UE in daily tasks and exhibit at least 10% slower reaching time with the paretic UE compared to the non-paretic UE to prevent ceiling effects. Exclusion criteria were contraindications to TMS or MRI, such as metal implants in the head; cardiac pacemaker or pace-generating implants; a history of seizures or a family history of medication-resistant epilepsy; severe cognitive impairment (Mini-Mental State Examination, MMSE < 24/30).

2.2. Design

Participants underwent three rTMS conditions in a randomized order, with at least a one-week interval between sessions: (1) inhibitory (1Hz) rTMS of contralesional M1; 2) facilitatory (5Hz) rTMS of cPMd; and 3) sham rTMS (Figure 1). At baseline, structural reserve was assessed using (1) the UEFM scale for clinical motor impairment, (2) DTI to quantify microstructural white matter tract integrity of corticospinal pathways, and (3) TMS to measure physiologic corticospinal excitability. Participants also completed practice on an RT motor task at baseline (details follow). The effects of each rTMS condition were determined based on changes from pre-test to post-test in performance on RT motor task and corticomotor excitability measured using TMS (Figure 1).

Figure 1.

Schematic experimental paradigm. Subjects had one visit for baseline and three visits for sessions of repetitive transcranial magnetic stimulation (rTMS). At baseline, measures of impairment (Upper Extremity Fugl-Meyer, UEFM) and damage (measured by transcranial magnetic stimulation, TMS, and Diffusion Tensor Imaging, DTI) were collected, and reaching test (RT) was practiced. Three experimental sessions were separated by at least a one-week interval. Each session included an rTMS regimen and a pre-test and post-test. Pre-test and post-test included RT and measures of neurophysiology, measured with. RTMS regimens were protocols for contralesional primary motor cortex (cM1) inhibition, contralesional premotor cortex (cPMD) facilitation, and sham delivered in a randomized order

2.3. Baseline Clinical, Imaging and Physiologic Assessments:

A comprehensive neurological examination was performed at baseline to confirm the diagnosis of the index clinical stroke. Baseline characteristics were recorded, including demographics (e.g., gender, age, pre-morbid hand dominance), stroke-related characteristics (e.g., stroke type [ischemic or hemorrhagic], lesion location [cortical and/or subcortical], side of paresis, and post-stroke duration), and cognitive function was evaluated using the MMSE.

A rehabilitation therapist administered the UEFM assessment, a reliable and valid measure of post-stroke UE motor impairment [18-23]. The UEFM evaluates the volitional movement of the upper limb, including the shoulder, elbow, forearm, wrist, and hand, both in and out of synergies. Each item is graded on a 3-point ordinal scale (0, cannot perform; 1, performs partially; and 2, performs fully), with a maximum total score of 66. We selected UEFM as a clinical measure of structural reserve due to its ability to assess both proximal and distal upper limb impairment across varying levels of impairment.

At baseline, DTI and TMS were performed to provide complementary assessments of corticospinal pathway reserve. DTI evaluates the structural anatomic integrity of white matter tracts, while TMS assesses the physiological excitability of residual motor pathways [24]. DTI metrics can be acquired even in cases of severe corticospinal injury when TMS-evoked motor responses, known as motor evoked potentials (MEPs), cannot be elicited (i.e., MEP−). On the other hand, when anatomic white matter tract integrity is largely preserved, TMS provides a granular measure of physiologic conduction [24-26].

The DTI protocol utilized high angular resolution diffusion imaging with the following parameters: axial slices with a field of view (FOV) of 256mm x 256mm x 104mm; 2mm³ isotropic voxels; TE/TR = 66/6200ms; 71 diffusion-weighting gradients; b = 1000s/mm²; and, 8 T2-weighted image volumes [27]. Images were processed using FSL [28] to generate whole-brain maps of fractional anisotropy (FA) [29]. Mean FA was calculated at the posterior limb of the internal capsule (PLIC) using a threshold FA value of 0.2 [30, 31]. The PLIC represents a major site of corticospinal convergence, and its white matter integrity strongly predicts UE motor recovery after stroke [24, 30]. PLIC was defined across three consecutive slices centered at the level of the foramen of Monro on the FA map (b0 space) [32, 33]. FA_Asymmetry between the ipsilesional and contralesional PLIC was calculated using equation (1). FA_Asymmetry values range from -1 to +1, with higher positive values indicating greater ipsilesional pathway damage (poor structural reserve) [24, 30]. Additionally, high-resolution T1-weighted MRI (3D-MPRAGE, 1 mm³ resolution) was collected to identify lesion location and compute lesion volume. Lesions were categorized as subcortical, cortical, or mixed.

$${\text{FA}}_{\text{Asymmetry}}=\frac{{\text{FA}}_{\text{contralesional}}-{\text{FA}}_{ipsilesional}}{{\text{FA}}_{\text{contralesional}}+{\text{FA}}_{ipsilesional}}$$ (1)

Single-pulse TMS was used to assess ipsilesional corticomotor excitability. A 70-mm figure-of-eight coil was connected to a TMS device delivering monophasic pulses (Magstim 200², The Magstim Company, UK). The coil was held tangentially over the scalp at a 45° angle, with the handle oriented backward and laterally. Coil placement was guided by MRI-based stereotaxy (Brainsight, Rogue Research, Montreal, QC, Canada), using individual MRI data and cranial landmarks for targeting. Participants were seated with their shoulders in slight flexion and abduction, elbows in 90° flexion, and forearms fully pronated, resting on a flat surface. Surface electromyography (EMG) (PowerLab 4/25T, AD Instruments, Colorado Springs, CO) was collected from the paretic Extensor Digitorum Communis (EDC), a key muscle for finger opening/extension, which is often weakened due to stroke [34]. EMG signals were acquired at 4KHz and analyzed offline using LabChart software (LabChart 8, ADInstruments CO., Sydney, AUS).

Before the TMS assessment, participants performed three maximum voluntary contractions (MVC) trials for both paretic and non-paretic EDC. Each trial consisted of a 3-5 second maximum isometric contraction, with ~30 seconds of rest between trials. The root-mean-square EMG of MVC was used to establish a target EMG amplitude for various parts of the TMS experiment and normalize MEPs.

Motor Evoked Potentials (MEPs) were recorded during slight active contraction of the EDC muscle (20% ±5% MVC) because active contraction increases the likelihood of eliciting MEPs, especially in cases of severe damage. TMS pulses were delivered to multiple scalp locations on a 1cm² resolution grid to identify the motor hotspot, where MEPs could be elicited ≥100μV (peak-to-peak) in size larger than pre-stimulus muscle activity (50ms window) at the lowest TMS intensity in at least 5 out of 10 trials [35]. The lowest TMS intensity required to generate criterion MEPs at the hotspot was called the active motor threshold (AMT), a measure of corticospinal excitability [36], expressed in % maximum stimulator output (MSO). The participant was further classified as MEP+ or MEP− based on their ability to produce criterion-sized MEPs with a maximum amplitude of TMS (100%MSO). MEP− status represents severe damage to descending corticomotor pathways [37].

2.4. rTMS Conditions

Participants underwent rTMS preparation following the same setup as for single-pulse TMS assessment collected at baseline. However, for rTMS, the 70mm figure-of-eight coil was connected to a device delivering biphasic pulses (MagStim^® Rapid² device, Magstim Company Ltd, UK). The MRI-based stereotaxy was also used for rTMS. The stimulation intensity was set to 90% AMT for the contralesional motor hotspot. Three rTMS conditions were administered: (1) Inhibitory rTMS (1Hz, contralesional M1): 1Hz pulses were delivered to the contralesional motor hotspot for 25 minutes (a total of 1500 pulses). This approach aims to reduce IHI drive from the contralesional M1 to the ipsilesional motor output [38-40]. (2) Facilitatory rTMS (5Hz, contralesional PMd): 5Hz pulses were applied 2.5cm anterior to the contralesional motor hotspot, based on previous rTMS studies. The stimulation lasted 10 minutes, delivered as five 60-second trains of 300 pulses each (a total of 1500 pulses) [26]. This high-frequency rTMS protocol can increase the excitability of higher-order cortices, including PMd and prefrontal regions [26, 41-44]. (3) Sham rTMS (placebo condition) utilized a placebo coil, miming the clicking noises generated with the active rTMS coil. This coil was positioned over the ipsilesional M1 motor hotspot and delivered 1Hz pulses for 25 minutes (a total of 1500 pulses). These three rTMS conditions (inhibitory, facilitatory, and sham) were designed to investigate the differential effects of stimulation on motor function in stroke survivors while ensuring consistency in number of pulses and session duration.

2.5. Motor and Physiology Outcome Assessments

Before and after rTMS, participants completed an RT motor task to evaluate UE reaching performance and underwent single-pulse TMS to assess IHI and corticospinal excitability. Details of the RT task can be found in previous publications [26, 45]. In summary, participants were seated in a high-back chair with the shoulder at 30° abduction, elbow at 90° flexion, forearm in a pronation position, and hand resting on a start button. Three target buttons (labeled 1, 2, and 3) were arranged in a semicircle at 80% of the participant’s maximum reaching distance. Each trial began with a visual cue (a number displayed on a computer screen corresponding to a target button). An auditory cue instructed the participants to reach forward “as fast as possible” and press the indicated button. Mechanical restraints (straps) were utilized to minimize trunk compensatory movements. Participants completed two blocks of 10 trials per side, with an order of testing randomized and counterbalanced across participants. The primary outcome was RT, defined as the time elapsed between the auditory cue and the correct button press. RT demonstrated excellent reliability, with an intraclass correlation (ICC) of 0.97 both within and between sessions, consistent with our earlier findings [26, 46]. To minimize practice-related confounds on rTMS outcomes, participants underwent RT task familiarization at baseline, where they practiced at least three blocks of 10 trials per side (Figure 1).

Single-pulse TMS was administered using the same protocol as at baseline. The ipsilesional motor hotspot was identified, and the AMT was determined. The contralesional motor hotspot was also localized, and the AMT and the resting motor threshold (RMT) were measured to assess IHI. MEPs were recorded from both ipsilesional and contralesional motor hotspots at 120% AMT. Measurements were collected across 10 trials while participants maintained slight contraction of the EDC muscle (20 ± 5% MVC). MEP amplitudes (mV) were normalized to MVC. IHI was measured using the ipsilateral silent period (iSP) method, as described in previous studies [26, 45, 47, 48]. Supra-threshold TMS pulses (150% RMT, 15 pulses) were applied to the contralesional motor hotspot while participants maintained a sub-maximal contraction of the paretic EDC (50% ± 5% MVC). These supra-threshold TMS pulses typically suppress ongoing ipsilateral EMG activity, resulting in what is referred to as the iSP. The onset of iSP was identified at the time point when rectified EMG drops below one standard deviation of baseline EMG for at least five data points, while the offset is defined at the point when EMG returns above this level [17, 25, 47, 48]. The extent of EMG suppression during iSP serves as an indicator of IHI from the targeted motor cortex. It is expressed as a percentage reduction in ongoing EMG relative to the baseline EMG over 100 ms using equation (2) [17, 47, 49]. Most physiologic measures demonstrated good to excellent reliability within and between sessions. ICCs were 0.94 (between) and 0.99 (within) for AMT, 0.83 (between) and 0.96 (within) for iSP, and 0.54 (between) and 0.70 (within) for MEP amplitude.

$$iSP=\frac{Mea{n}_{background}-Mea{n}_{iSP}}{Mea{n}_{background}}\times 100\%$$ (2)

2.6. Data and Statistical Analysis

The change in RT following cM1 inhibition rTMS and cPMd facilitation rTMS was calculated as a difference score, ΔRT_cM1-cPMd (in ms). A positive ΔRT_cM1-cPMd indicated greater RT improvement with cM1 inhibition, while a negative ΔRT_cM1-cPMd suggested greater improvement with cPMd facilitation. Participants were categorized based on which rTMS protocol yielded more improvement. The relationship between ΔRT_cM1-cPMd and baseline severity characteristics was evaluated using 1) Spearman's rank correlation (ρ) for continuous variables and 2) Mann-Whitney U test for dichotomous variables. Baseline severity characteristics included: 1) clinical motor impairment score (UEFM); 2) lesion location; 3) anatomic CST integrity (FA_Asymmetry); and 4) residual ipsilesional pathway physiology (MEP+/− status, AMT).

Classification and Regression Tree (CART) analysis was used to determine which baseline severity characteristic(s) best-predicted improvement with cM1 inhibition rTMS versus cPMd facilitation rTMS. CART generates a decision tree algorithm that sequentially applies key characteristic(s) to classify participants into outcome categories [50, 51]. Only characteristics that showed a significant relationship with ΔRT_cM1-cPMd were entered into the CART model. The resulting algorithm, termed START (Stratification Algorithm for Repetitive TMS), was expected to identify key severity characteristics and cut-off values that distinguish individuals responding better to each rTMS protocol. START was expected to support the hypothesis that less severe participants would improve more with cM1 inhibition rTMS, while more severe participants would benefit more from cPMd facilitation rTMS.

To ensure rigor, CART results were internally validated using two methods: (1) Bootstrap validation- The C statistic, a measure of classification power, was calculated from bootstrap samples. C statistic values range from 0.5 (random chance) to 1.0 (perfect classification) [52, 53]. (2) Inferential testing- a two-way repeated measures ANOVA examined RT changes across the three rTMS conditions in subgroups of participants classified by severity (Group x Condition). Significant interactions were investigated using post-hoc pairwise comparisons. To examine the effects of different rTMS conditions on neurophysiology within severity subgroups, non-parametric tests were applied: (1) Wilcoxon Signed Ranks Test for within-group comparisons, and (2) Mann-Whitney U test for between-group comparisons. The alpha level was set at 0.05. The effect size was represented by partial eta square (partial η²), with values interpreted as 0.01 (small effect), 0.06 (medium effect), and 0.14 (large effect) [54].

3. Results

A total of 26 chronic stroke participants were enrolled in the study. However, data from two participants were excluded due to the incidental finding of clinically asymptomatic infarcts in the intact hemisphere on 3T MRI (age 56 and 59, with UEFM scores of 25 and UEFM 31, respectively). The final analysis included 24 participants (9 females) with a mean age of 61.1 ±11 years, average post-stroke duration of 43.6 ± 52.9 months, and mean UEFM score of 41 ± 16.4 (range 7-63). Seventeen participants had subcortical lesions, while 7 had cortical or mixed lesions. MEPs were present in 17 participants (MEP+) and absent in 7 (MEP−). DTI data were available in 21 participants because two participants could not undergo 3T MRI due to claustrophobia and data of one participant were corrupted due to a technical issue. Detailed clinico-demographic characteristics are reported in Table 1.

Table 1.

Clinico-Demographic characteristics

Subject	Age/Gender/ Handedness	Time Since Stroke (Months)	Side of Paresis	MMSE (Max 30)	Lesion Location	Lesion Volume (cm3)	UEFM (Max 66)	Active MEP (+/−)	FAAsymmetry
1	74/M/R	108	L	26	Pontine	0.01	63	+	−0.0169
2	63/M/R	44	L	26	Pontumedullary junction	0.5	61	+	0.0372
3	63/M/R	10	R	30	Internal capsule, putamen, caudate nucleus	0.1	61	+	0.0270
4	61/F/R	9	R	30	Corona radiata, putamen, caudate, internal capsule	5.2	60	+	0.0355
5	57/M/L	14	L	28	Caudate, putamen, thalamus, internal capsule	0.3	60	+	−0.0049
6	58/M/R	11	R	27	Internal capsule	0.2	58	+	0.0270
7	72/F/R	13	R	27	Thalamus	0.5	57	+	−0.0081
8	57/M/R	53	R	29	Centrum semi-ovale, basal ganglia, caudate	2.4	52	+	0.1963
9	52/F/R	30	R	30	Thalamus, internal capsule	0.6	52	+	0.0494
10	69/M/R	45	L	24	Cortical (mesial frontal cortical, superior frontal gyrus)	9.4	48	+	−0.0007
11	54/M/R	37	R	29	Caudate, internal capsule	N/A*	46	−
12	70/M/R	23	R	30	Internal capsule, putamen	1.2	44	−	−0.0213
13	78/M/R	95	R	30	Caudate, corona radiata, internal capsule, putamen	1	42	+	0.2164
14	28/F/R	14	L	29	Cortical (frontal, parietal, temporal), corona radiata, internal capsule, thalamus, putamen	63.7	35	+	0.2742
15	77/M/R	20	R	29	Corona radiata, thalamus, internal capsule	3	34	+	0.4535
16	57/M/R	18	R	27	Corona radiata, putamen, internal capsule	1.2	32	+	0.1800
17	58/F/R	11	L	29	Cortical (temporal, frontal, parietal), basal ganglia, putamen, caudate, corona radiata, internal capsule, thalamus	12.5	31	−	0.2157
18	59/F/L	75	R	28	Putamen, internal capsule, corona radiata, thalamus	1.5	30	+	0.1307
19	51/F/R	33	L	29	Cortical (temporal, frontal, parietal), corona radiata, caudate, putamen, internal capsule, thalamus	82.6	27	+	0.3985
20	52/M/R	16	L	28	Cortical (temporal, parietal, frontal), thalamus, putamen, internal capsule, caudate	245.9	24	−
21	77/M/R	38	R	28	Basal ganglia, corona radiata, internal capsule	0.9	24	+	0.0537
22	58/M/R	259	L	29	Cortical (temporal, occipital, frontal, parietal), corona radiata, basal ganglia, caudate, putamen, thalamus, internal capsule	199.2	21	−	0.3807
23	67/F/R	31	L	30	Corona radiata, internal capsule	N/A*	16	−
24	56/F/R	40	L	29	Cortical (frontal, parietal temporal), corona radiata, caudate, thalamus, internal capsule, putamen	245.8	7	−	0.4035
Mean ± SD	61.1 ±11.0/ 9F/22 R	43.6 ± 52.9	13 (R)	28.4± 1.6	Cortical/Mixed 7 of 24	39.9 ± 80.7	41 ± 16.4	17 (+)	0.144 ± 0.155

^*No 3T imaging due to claustrophobia

Correlation analyses revealed significant associations between ΔRT_cM1-cPMd and the following: 1) clinical UE motor impairment score (UEFM: ρ = 0.580, p = 0.003); 2) anatomic CST integrity (FA_Asymmetry: ρ = −0.682, p = 0.001); and 3) ipsilesional physiologic excitability (AMT: ρ = −0.554, p = 0.021) (Figure 2). Participants with better UEFM scores, more favorable CST integrity, and higher ipsilesional excitability showed greater RT improvements following cM1 inhibition rTMS. Conversely, those with lower UEFM scores, poorer CST integrity, and reduced ipsilesional excitability benefitted more from cPMd facilitation rTMS. Additionally, ΔRT_cM1-cPMd did not significantly differ between participants with subcortical versus cortical/mixed lesions or between MEP+ and MEP− participants (U = 41.0, p = 0.240; and U = 52.0, p = 0.634, respectively). Therefore, lesion location and residual pathway physiology status (MEP+/−) were not associated with RT improvements across different rTMS conditions.

Figure 2.

Association between (A) ΔRT_cM1-cPMd and UEFM, (B) ΔRT_cM1-cPMd and FA_Asymmetry, (C) ΔRT_cM1-cPMd and AMT, using Spearman's rank correlation coefficient (ρ) for continuous variables and the Mann-Whitney U test for dichotomous data. * p < 0.05

Based on its significant relationship with ΔRT_cM1-cPMd, both UEFM and FA_Asymmetry were included in the CART analysis. AMT was excluded because 7 out of 24 participants were MEP−. CART analysis generated a decision tree with UEFM as the primary decision point (criterion score: 42/66) (Figure 3). Among participants with UEFM > 42, 11 out of 12 (91.67%) showed greater improvement with cM1 inhibition rTMS. In contrast, among participants with UEFM ≤ 42, 10 out of 12 (83.33%) improved more with cPMd facilitation rTMS. Secondary decision points based on UEFM and FA_Asymmetry further refined the classification. All nine participants with UEFM > 48 improved more with cM1 inhibition rTMS (100% accuracy). Similarly, all six participants with UEFM ≤ 42 and FA_Asymmetry > 0.216 showed greater improvement with cPMd facilitation rTMS (100% accuracy). For participants in the mid-range of severity, UEFM and FA_Asymmetry achieved 66.67% accuracy in distinguishing responses. Specifically, two out of three participants with UEFM scores between 43 and 48 (inclusive) improved more with cM1 inhibition rTMS, while four out of six participants with UEFM ≤ 42 but FA_Asymmetry ≤ 0.216 showed greater improvement with cPMd facilitation rTMS (66.67% accuracy) (Figure 3).

Figure 3.

The decisional algorithm defined by Classification and Regression Tree to stratify subjects to responders to contralesional primary motor cortex (cM1) inhibition with repetitive transcranial magnetic stimulation (rTMS) and contralesional dorsal premotor cortex (cPMd) facilitation rTMS (UEFM: Upper Extremity Fugl-Meyer; DTI: Diffusion Tensor Imaging; FA: fractional anisotropy; PLIC: posterior limb of the internal capsule)

Three participants could not be classified by CART analysis. This included one participant (subject # 10) with a UEFM score of 48, who improved more with cPMd facilitation rTMS rather than cM1 inhibition rTMS. Additionally, two participants (subjects #17 and #20) with UEFM ≤ 42 showed greater improvement with cM1 inhibition rTMS instead of cPMd facilitation rTMS (Table 1). All three non-classified participants had cortical or mixed lesions. Among them, one (subject #20) lacked DTI assessment of CST integrity, while two (subjects #17 and #20) were MEP− indicating absence of TMS response.

Internal validation yielded a C statistic of 0.87 (95% CI, 0.86, 0.89), indicating strong prediction performance of the overall model. A two-way ANOVA revealed a significant group * condition interaction (F = 3.908, p = 0.048, partial η² = 0.151) (Figure 4A). Post-hoc analysis showed less impaired participants tended to exhibit greater RT improvements with cM1 inhibition rTMS than more impaired participants (p = 0.065). More impaired participants demonstrated significantly greater RT gains with cPMd facilitation rTMS compared to sham rTMS (p = 0.003) and also tended to improve more with cPMd facilitation than cM1 inhibition rTMS (p = 0.054). Neurophysiological responses differed between subgroups: less impaired participants showed greater ipsilesional MEP gains following cM1 inhibition rTMS (U = 2.65, p = 0.007) (Figure 4B, 4C). More impaired MEP− participants exhibited greater IHI reduction after cPMd facilitation compared to cM1 inhibition rTMS (Z= −2.20, p= 0.028) or sham rTMS (Z= −2.20, p= 0.028) (Figure 4D, 4E, 4F).

Figure 4.

Changes following inhibitory repetitive transcranial magnetic stimulation (rTMS) over contralesional primary motor cortex (cM1), facilitatory rTMS over contralesional dorsal premotor cortex (cPMd), and sham in less and more severe subjects with stroke, which categorized with 42 of Upper Extremity Fugl-Meyer score. A) Improvement from pre-test to post-test) in reaching time (RT) in ms, B) Increase in MEP, C) A patient with UEFM of 57 demonstrates 295.5% MVC increases in ipsilesional corticospinal excitability (motor evoked potential, MEP, at 120% of active motor threshold) following cM1 inhibition rTMS, D) Reduction in iSP in more-severe patients with no MEP, E) Ipsilateral silent period (iSP) of a patient with UEFM of 31 is 47% at the baseline, F) The patient with UEFM of 31 shows reduction in iSP (from 47% to 37%) following cPMd facilitation rTMS (*p < 0.05).

4. Discussion

This study sought to identify severity-based criteria for differentiating patients who achieve greater paretic UE motor gains with inhibitory contralesional rTMS from those who benefit more from facilitatory cPMd rTMS, addressing the ongoing challenge of identifying an optimal measure of structural reserve for targeted brain stimulation treatments. Our findings led to the development of the START algorithm, a preliminary classification framework based on UE motor impairment (UEFM) and residual anatomic CST integrity (FA_Asymmetry). Patients with UEFM > 42 (91.67%) showed greater improvement with cM1 inhibition rTMS, associated with enhanced ipsilesional excitability, while those with UEFM ≤ 42 (83.33%) benefitted more from cPMd facilitation rTMS, likely based on IHI reduction, particularly among MEP− participants. More conservative thresholds (UEFM > 48 or FA_Asymmetry > 0.216) improved classification accuracy, but the UEFM 42 threshold alone provided a strong predictive value (83.33%- 91.67% accuracy). While our findings support the bimodal theory [7, 47], the START algorithm remains preliminary. Future studies should validate the START algorithm, examine long-term clinical outcomes in larger, more heterogeneous stroke populations, and establish sensitivity and specificity metrics to refine its predictive capabilities.

It is important to juxtapose the findings of this study with those of earlier studies. Notably, the UEFM 42 criterion aligns with our previous study (UEFM 43), which investigated IHI in a different sample and muscle (first dorsal interosseus or FDI rather than EDC). In that study, participants with less impairment (UEFM > 43) exhibited stronger IHI associated with greater motor impairment [10]. Similarly, the present study found that inhibiting cM1 led to greater motor gains in less impaired participants (UEFM > 42), reinforcing the idea that stronger inhibitory effects from the contralesional motor cortices may be a relevant target for UE motor improvement in chronic stroke survivors with moderate to mild impairment. Other studies have reported comparable UEFM thresholds for predicting positive responses to cM1 inhibition rTMS in moderately to mildly impaired stroke survivors [55, 56]. Additionally, 11 out of 12 less impaired participants (UEFM > 42) in this study had subcortical injuries and exhibited ipsilesional potentiation following cM1 inhibition rTMS. These physiological findings further support the notion that subcortical injuries (i.e., where the motor cortex is intact) may enhance the effects of cM1 inhibition rTMS by permitting greater disinhibition of residual ipsilesional motor output. This, in turn, facilitates broader propagation of stimulation effects to other nodes of the motor network, leading to a more favorable response [56-60].

The effects observed in more impaired participants (UEFM ≤ 42) further support previous findings on the relationship between IHI and motor impairment [10]. In our previous study, more impaired participants (UEFM < 43) exhibited reduced IHI associated with greater impairment and showed a further reduction in IHI alongside clinically meaningful UE motor gains [10]. Similarly, another study reported a comparable relationship between IHI and motor impairment in stroke survivors with more severe impairment (UEFM < 30) [9]. In addition, our findings indicate that in more impaired participants (UEFM ≤ 42), particularly those without evidence of residual motor physiology (MEP−), cPMd facilitation rTMS led to a further reduction in IHI compared to cM1 inhibition rTMS or sham rTMS. Our findings build on early case-series findings suggesting that cPMd facilitation rTMS may produce greater reductions in IHI and higher UE motor gains than ipsilesional M1 facilitation rTMS in severely impaired participants (n= 4, mean UEFM 27) [16]. Taken together, these preliminary findings suggest that cPMd facilitation rTMS may help reduce IHI, potentially enhancing positive contributions from the intact contralesional motor cortex in more severely impaired patients [7, 10]. The effects of cPMd facilitation rTMS on rehabilitation outcomes remain an active area of investigation [61-63].

We observed that the clinical motor impairment criterion (UEFM 42) demonstrated strong internal validity, which was further confirmed through inferential testing. Incorporating an additional assessment of residual CST integrity (FA_Asymmetry > 0.216) may improve the accuracy of the clinical criterion (UEFM ≤42) in identifying candidates for cPMd facilitation rTMS, particularly those with damage beyond the point-of-no-return [37]. Several lines of evidence support the utility of CST lesion load in stratifying patients with a higher likelihood of response [24, 64]. This study highlights the importance of this measure, especially for those with moderate to severe impairment (UEFM ≤ 42). However, in cases where CST integrity assessments are not readily available, a more conservative clinical impairment criterion (such as UEFM < 27- 30) may serve as a reasonable alternative [9, 16].

We must acknowledge several potential limitations of this study. First, the small size and exploratory nature of our study led to stratification into smaller subgroups. Balancing the need for extensive baseline assessments (TMS, MRI) and familiarization with the RT task against the burden of multiple crossover visits and the requirement for an effective washout interval (≥1 week) ultimately constrained the overall sample size. Second, to ensure a uniform collection of motor outcomes across a widely impaired sample, we adopted a proximal reaching motor task rather than a dexterity-based task. We also could not include kinematic testing due to a lack of equipment. While the reaching motor task is not specific to the muscle tested using TMS (EDC), reaching forward to a target button may still engage EDC as a synergist to maintain distal extension. TMS of proximal muscles can pose reproducibility concerns due to small and variable MEPs [65, 66], whereas the EDC has demonstrated high reliability with TMS in stroke populations [14]. Third, we observed that more impaired participants exhibited modest paretic RT motor gains with cPMd facilitation, though these gains remain noteworthy compared to other rTMS conditions (Figure 4A). Prior studies, including ours, have reported fatigue-related effects on movement and physiology in severely impaired participants [67, 68]. The potential of cPMd facilitation rTMS to mitigate such fatigue-related motor deficits warrants further investigation. Finally, due to time constraints, we could not examine ipsilateral brainstem-mediated pathways.

5. Conclusions

In conclusion, our study introduces the START algorithm, a preliminary classification framework designed to stratify patients for either contralesional M1 inhibition or cPMd facilitation rTMS. We identify clinical motor impairment (UEFM) and residual white matter integrity (FA asymmetry) as key severity criteria for predicting response, providing an optimal measure of the structural reserve for targeted contralesional treatments [37]. Specifically, patients with UEFM > 42 (91.67%) exhibit greater UE motor improvement with cM1 inhibition rTMS, associated with enhanced ipsilesional excitability, while patients with UEFM ≤ 42 (83.33%) benefit more from cPMd facilitation rTMS, likely based on IHI reduction, particularly among MEP− participants. More conservative thresholds (UEFM > 48 or FA_Asymmetry > 0.216) improved classification accuracy, but the UEFM 42 threshold alone provides a strong predictive value (83.33% - 91.67% accuracy). Our findings align with the bimodal theory, suggesting that in individuals with milder motor impairment, excessive IHI from the intact contralesional motor cortices is a viable therapeutic target to enhance motor gains. Conversely, in those with more severe ipsilesional damage, the contralesional motor cortices, especially the cPMd, may serve as a critical compensatory resource for greater motor gains. While promising, the START algorithm requires further clinical validation in larger, more heterogeneous samples, including investigations into its long-term impact on motor outcomes. Future studies should explore its utility in optimizing patient stratification to develop more targeted and effective neuromodulation treatments in efficacy trials.