Feasibility Randomized Controlled Trial of Real-Time fMRI Neurofeedback for Reading Rehabilitation in Aphasia

Abstract

Background:

Reading impairments are common in stroke-induced aphasia and limit participation in functional and leisure activities. Traditional rehabilitation strategies show limited generalization, underscoring the need for novel interventions targeting residual neural networks.

Method:

This feasibility randomized controlled trial (RCT) evaluated real-time functional magnetic resonance imaging (fMRI) neurofeedback (NFB) intervention for post-stroke reading deficits. Subacute left-hemisphere stroke survivors and healthy controls (HCs) completed three weekly fMRI NFB and 10 out-of-scanner practice sessions. Stroke participants were randomized to contingent NFB (based on left supramarginal gyrus [SMG] activity, N=4) or noncontingent NFB (shuffled feedback from another participant, N=3). HCs (N=4) received contingent NFB and served as a normative reference. Primary outcomes were changes from baseline to post-intervention (~3 weeks) in task-based brain activity (motor imagery/word/nonword reading > baseline), resting state connectivity, and reading aloud. Reading comprehension was a secondary outcome. Group × session effects were tested using repeated-measures analyses and planned contrasts.

Results:

Task fMRI revealed training-related activation increases in left SMG (z=4.7, cluster-corrected p=0.05) and broader reading network in the contingent NFB group, particularly during nonword reading. Activation increases in noncontingent stroke group and HCs were more widespread and less reading-specific. Resting-state fMRI revealed greater integration among motor, auditory, and language networks in the contingent groups, with more disorganized patterns in the noncontingent group (permutation p=0.01; Δr=−0.1−0.1). No changes were observed in reading aloud.

A significant group × session interaction was found for Reading Comprehension Battery for Aphasia, 2^nd ed. (RCBA-2), (F(2,8)=8.00, p<0.05, η²=0.67). The contingent NFB stroke group improved more than HCs, (MD in RCBA-2 change =9.75, 95% CI=1.99–17.51, t(6)=3.07, p<0.05) and noncontingent NFB stroke group (RCBA-2 change=11.42, 95% CI=1.12–21.71, t(5)=2.85, p<0.05).

Conclusions:

These findings support the feasibility of targeting residual reading network during early recovery using fMRI NFB. Confirmation of these preliminary effects awaits completion of the ongoing RCT.

Registration:

ClinicalTrials.gov, NCT# NCT04875936

Graphical Abstract

1. Introduction

Among 795,000 new stroke survivors annually in the US and 13.7 million worldwide, 20% experience reading impairments due to aphasia¹. For 61% of individuals with aphasia, language and reading deficits persist beyond one year after stroke and negatively impact life quality². We explore whether real-time functional magnetic resonance imaging (rtfMRI) neurofeedback (NFB) can promote adaptive neural plasticity and alleviate post-stroke language and reading impairments.

One major challenge in reading rehabilitation is the limited generalization beyond trained materials. Biologically inspired interventions, such as rtfMRI NFB may address this limitation by directly promoting post-stroke neuroplasticity. By receiving feedback about brain activity, participants can learn mental strategies to regulate brain states³. When practiced repeatedly, self-regulation efforts can drive learning-induced cognitive and motor improvements.

FMRI NFB has shown promise for neurorehabilitation across several neurological and psychiatric conditions^4–7. In stroke, small studies report benefits for motor^8–10 and visuospatial function¹¹. Only two studies have explored its use in aphasia rehabilitation, with mixed results^12,13. Sreedharan et al.¹³ successfully trained 4 individuals with nonfluent aphasia to upregulate connectivity between Broca’s and Wernicke’s areas by performing covert language tasks. However, a follow-up study found no significant language improvements in the NFB group compared to usual care¹². Thus, while rtfMRI NFB shows potential in aphasia rehabilitation by inducing neural changes, its behavioral impact remains unclear.

Building on the emerging but limited evidence of fMRI NFB in aphasia, we aimed to develop a biologically informed intervention for stroke-induced reading deficits. Central to this aim was identifying modifiable neural mechanisms underlying impaired reading. Neuroimaging studies suggest that stroke-related dysfunction extends beyond structural lesions to persistent hypoperfusion and hypoactivity in perilesional and functionally connected areas^14–16, associated with poorer language outcomes. In our longitudinal study of 31 left-hemisphere stroke survivors¹⁷, both subacute and chronic cerebral blood flow (CBF), was reduced in intact parts of the left reading network relative to the right-hemisphere. Reduced subacute CBF predicted poorer chronic reading recovery. Although the study lacked a control group; a subsequent investigation of 43 chronic aphasia participants and 25 healthy controls confirmed reduced left parietal and temporal CBF associated with reduced residual language ability in the aphasia group¹⁶. Thus, reduced cerebral perfusion and brain activity in the intact parts of the left language network represent a modifiable target for aphasia rehabilitation.

Building on evidence linking left hemisphere hypoperfusion to persistent language deficits, we propose that restoring cerebral perfusion and neural activity in the intact left reading network components may enhance recovery. Supporting this view, Hillis¹⁸ showed that pharmacologically induced reperfusion of left perisylvian language cortex following stroke transiently improved naming and comprehension. Saur et al.¹⁹ and Fridriksson²⁰ similarly reported that chronic aphasia recovery was associated with re-emergence of left-hemisphere brain activity. Consistent with these findings, our perfusion study¹⁷, showed that higher subacute perfusion in intact ipsilesional reading network predicted better chronic phonology competence, independent of lesion size.

To facilitate left-lateralized activation of the reading network, we employed motor imagery, a mental simulation of movement without motor output, which can be performed even by stroke patients with severe hand impairments. Lateralized motor imagery preferentially activates the hemisphere contralateral to the imagined movement (e.g., left hemisphere for right hand imagery)²¹, an asymmetry that can be further enhanced through fMRI NFB. This approach offers a novel means to modulate activity in structurally intact but hypoperfused areas. We, therefore, tested whether right-hand motor imagery could increase perfusion and activation in left-hemisphere regions that are important for reading.

The most common reading deficit for left-hemisphere stroke patients is difficulty with orthography-phonology conversion ^1,17. This process, essential for reading and decoding unfamiliar words, typically activates the left inferior frontal (IFG) and supramarginal gyri (SMG), which are part of the dorsal language pathway^22–25. We selected the left SMG as our target region because of its established role in both phonological processing and motor-related functions. For example, the left SMG was implicated in planning finger positioning for tool use²⁶, and kinesthetic motor imagery of finger movements²⁷. Owing to SMG’s dual role in motor planning and reading-related processes, we expected that right hand motor imagery could be an effective mental strategy to help increase its activation, potentially facilitating reading recovery.

This paper reports feasibility data from an ongoing randomized controlled trial (RCT) evaluating rtfMRI NFB combined with motor imagery for post-stroke reading rehabilitation. The aims were to assess protocol viability, including participant engagement and outcome measure suitability, and to identify preliminary behavioral and neural responses to training. These feasibility findings provide preliminary insight into protocol implementation and outcome measures, support continued evaluation within the ongoing RCT, and will be interpreted in the context of the trial’s full results.

2. Materials and Methods

2.1. Data Availability Statement

In accordance with a pre-established data dissemination plan, de-identified data and documentation will be made available on request under a data-sharing agreement requiring use for research purposes only, appropriate data security, and destruction or return of the data after analyses are completed.

2.2. Participants

Stroke participants were inpatients at an acute rehabilitation hospital and were screened using electronic medical records for documentation of left-hemisphere stroke and aphasia. Age-matched healthy controls (HCs) were recruited from the community. Planned eligibility criteria for stroke participants included: right-handedness, fluency and literacy in English prior to stroke, no contraindications to MRI, first-ever left-hemisphere stroke with reading deficits defined by performance on subsets of the Reading Comprehension Battery for Aphasia 2^nd ed. (RCBA-2)²⁸; and no history of developmental dyslexia, other neurological conditions, or vision/hearing problems uncorrectable with lenses or reported as severe. Although stroke participants were not screened for ideomotor apraxia, none had lesions in the inferior parietal lobe, a region associated with apraxia²⁹. All participants provided written informed consent. Aphasia-friendly forms were used alongside standard consent forms for stroke participants. The study was approved by the Kessler Foundation Institutional Review Board (Protocol: R-1147–21).

2.3. Materials and Procedure

2.2.1. Study Design

This feasibility study represents the initial phase of an ongoing preregistered randomized controlled trial (ClinicalTrials.gov, NCT# NCT04875936). The planned recruitment includes 14 participants per group (stroke and HCs) assigned to contingent or noncontingent NFB. This manuscript adheres to the CONSORT 2010 extension for randomized pilot and feasibility trials³⁰ and guidelines for the design and reporting of neurofeedback interventions³¹. Four masked stroke participants and four HCs received contingent feedback, while three stroke participants received noncontingent feedback (Fig. S1). Contingent NFB reflected participants’ own brain activity; whereas noncontingent NFB, designed to control for task engagement, was derived from another participant’s data randomized across sessions and trials. (See supplemental materials for details, including randomization). HCs served as a normative reference for task-related activation and NFB engagement. Feasibility analyses focused on acceptability and initial responsiveness to the NFB intervention across these preliminary groups.

2.2.2. Screening

All participants completed screening, demographic and handedness questionnaires. Stroke participants were screened for reading deficits using Paragraphs test (subtests VIII, IX) of the (RCBA-2)²⁸ (See supplemental materials).

2.2.3. Baseline and Outcome Assessments

Baseline testing included comprehensive aphasia assessment using the Western Aphasia Battery-Revised (WAB-R)³², reading comprehension assessment using the full RCBA-2, picture naming using the Boston Naming Test – short version³³, and two-alternative forced-choice (2AFC) tests of semantics, phonology, and orthography (See supplemental materials). Participants also completed a computerized reading aloud task with 120 monosyllabic words and 60 length-matched pronounceable nonwords. Tasks were presented on a 15-inch monitor using OpenSesame v. 3.3.11.

We assessed unilateral gross manual dexterity using the Box and Blocks test³⁴, and grip strength using an electronic dynamometer (Jamar Smart Hand Dynamometer, Patterson Medical) to assess for potential motor effects of imagery. All instruments were re-administered at outcome.

Acceptability was assessed via self-report ratings of motivation, commitment, and session difficulty on 7-point Likert scales (motivation/commitment: 1=very low to 7=very high; difficulty: 1=very easy to 7=very hard).

2.2.4. Neurofeedback Training

Participants were trained in right-hand motor imagery (e.g., finger tapping, typing, playing piano, gesturing, tracing text), emphasizing large amplitude movements from a first-person perspective and associated sensory experience without overt movement. All participants practiced the task before fMRI NFB training to ensure successful in scanner performance.

After motor imagery (MI) practice, participants underwent MRI scanning, including structural and functional scans (Fig.1A). FMRI NFB training consisted of 3 sessions each with two 15-minute runs of 9 trials. Trials comprised fixation, motor imagery, reading aloud, feedback, and rest, with timing locked to the TR (1.5 s) (Fig. 1B). During reading blocks, participants read aloud either words or nonwords, randomized across runs and sessions. Each word/nonword was presented for 1 second followed by 1–3 second jitter. Voice responses were collected via an MRI compatible fiberoptic microphone (FOMRI III+, Optoacoustics). Each session included unique lists of 72 words and 72 nonwords, matched across training days on linguistic properties.

Figure 1.

Study design and fMRI neurofeedback (NFB) trial structure. (A) Timeline: Stroke participants (n = 7) and healthy controls (n = 4) completed three weekly fMRI NFB training sessions (T1–T3). Transfer runs were acquired at T1 and T3 to assess pre- vs. post-training effects without feedback. Structural imaging (FLAIR, T1-weighted) supported lesion delineation and image co-registration. (B) Trial structure: Each trial began with fixation, followed by a regulation block using mental imagery. This was followed by reading aloud 8 words/nonwords each presented for 3 s and receiving feedback (numeric score with smiley face), then a rest period.

Online analysis and stimulus presentation were implemented using OpenNFT (opennft.org). Participants received intermittent feedback based on percent signal change in the target ROI (scaled 0–100). NFB signal for each trial was computed as activation during MI and reading relative to baseline while controlling for head motion. Higher values paired with a smiling face indicated greater ROI recruitment. Stimuli were back-projected and viewed via a mirror mounted on the head coil.

2.2.5. Pre- and Post-Training Runs

Pre- and post-training (transfer) runs were conducted before and after NFB training using the same task structure but without feedback and with four trials each (Fig.1A). These scans assessed transfer of training-related changes to a situation without feedback. Reading stimuli consisted of 32 additional matched words and nonwords, identical across pre- and post-training runs.

2.2.6. Homework

All participants completed 10 supervised homework sessions that mirrored the structure and timing of in-scanner NFB sessions, presented using E-prime 3.0. Participants were encouraged to practice the imagery strategy that produced the strongest feedback during in-scanner training. Homework sessions 1–5 followed the first NFB session, and sessions 6–10 followed the second NFB session, each on separate days.

2.2.7. ROI mask selection

Left SMG was selected as the target ROI for its central role in letter-to-sound conversion. A separate pilot study confirmed that SMG is reliably activated by both reading and motor imagery (See supplemental materials, “ROI Selection”, Fig.S2).

2.2.8. Neuroimaging

Participants underwent 3 MRI scans (Fig. 1A, T1-T3) on a 3.0T Siemens Skyra scanner using a 20-channel head coil. Structural Neuroimaging. T1-weighted structural scans were acquired at each session to support registration and ROI localization (TR=2100 ms, TE=3.43 ms, TI=900ms, 176 sagittal slices, 1mm³ isotropic voxels). T2-weighted Fluid Attenuated Inversion Recovery (FLAIR) scans (TR=9000 ms, TE=91 ms, TI=2500ms, 50 slices, 1×1×3mm³ voxels) was acquired at T1 and T3 to aid lesion segmentation and detect new pathology. FMRI data were acquired using rapid simultaneous multi-slice echo-planar imaging (TR=1.5s, TE=30ms, 44 slices, gap = .5mm, 2mm³ isotropic voxels). Functional runs included resting state (326 TRs), pre-training (262 TRs), training (588 TRs) and post-training (262 TRs) scans. During resting state scans, participants fixated on a central cross and let their mind wander.

2.2.9. Lesion Mapping

Lesions were segmented using combined manual and intensity-based voxel selection in FSLeyes (v.1.13.0) (Fig. S3), with cost-function masking applied to avoid warping during registration.

2.2.10. Analysis

Behavioral Data:

Group differences over time were assessed using repeated-measures analysis comparing testing session (baseline vs. outcome) across HCs and stroke participants in the contingent and noncontingent conditions for behavioral outcomes. Group was modeled as a between-subjects factor and session as a within-subjects factor; significant interactions were followed by planned between-group comparisons. Motivation, commitment, and perceived session difficulty were assessed on each of the three NFB training days using one-sample t-tests against a reference value of 1 (very low/very easy). Pre- to post-training changes in NFB signal were evaluated using repeated-measures analyses with session as a within-subjects factor and group as a between-subjects factor.

Task fMRI:

FMRI data processing and analysis were performed using FEAT³⁵. Eight runs per participant (6 training, 2 transfer) were skull-stripped, motion corrected, smoothed (6mm FWHM), and high-pass filtered (100s cutoff), then co-registered to individual T1-weighted images and the MNI152 2-mm template. Trial events (baseline fixation, motor imagery, word reading, nonword reading, and NFB display) were modeled as separate regressors, with standard motion parameters and their temporal derivatives included as regressors of no interest. For each run, we modeled the contrasts between the block of interest (motor imagery, reading words, and reading nonwords) and baseline. Whole-brain group-level analyses were performed separately for stroke participants and HCs and for training and transfer runs. Training effects were assessed by contrasting consecutive runs, and transfer effects by comparing post- to pre-training runs. Mixed-effects analyses were performed using FLAME stages 1 and 2 (FMRIB's Local Analysis of Mixed Effects). Statistical maps were cluster-corrected (z > 1.96, p < 0.05).

Resting State fMRI:

Dynamic resting state functional connectivity (dFC) was compared for rs-fMRI scans acquired before and after NFB training. Each scan was skull-stripped, motion corrected, smoothed (5-mm FWHM), and co-registered to the MNI152 2-mm template. Motion artifacts were further reduced using ICA-AROMA³⁶. We regressed signals from CSF and white matter, performed linear detrending and applied a Gaussian high-pass temporal filter (100s FWHM). Mean timeseries were extracted from 333 ROIs in the Gordon et al. ³⁷ parcellation, excluding lesioned voxels.

DFC matrices were computed using Pearson correlation with a sliding window of 60 TR (120s) and a 6-TR (12s) step, yielding 44 matrices per participant per session. Each matrix was vectorized as geodesic distances to cluster centroids and used to train a LASSO-regularized logistic regression classifier with 5-fold cross-validation to identify connectivity changes distinguishing scans acquired before versus after NFB training. Statistical significance was assessed using permutation testing with shuffled feature labels. Connections with weights exceeding the 95th percentile of permutation values were retained, with the top 1% reported in the whole brain analysis. Changes in dFC from day 1 to day 3 are expressed as differences in Pearson r.

3. Results

Participants:

A total of 11 participants were enrolled: 7 individuals with subacute left-hemisphere stroke (mean age=61.1±15.3, 1 woman, 3 Caucasian, 4 African American, 1 Hispanic, mean days post stroke =30±17.6, WAB-R Aphasia Quotient (AQ)=80.43, SD=24.20) and 4 age-matched HCs (mean age=63.5±5.1, 1 woman and 3 men, 3 Caucasian, 1 African American). Stroke participants had a mean lesion volume of 28.89 ml (SD = 35.60 ml) and exhibited reading deficits on the RCBA-2. WAB-R aphasia subtypes included anomic (n=6) and Broca’s aphasia (n=1) (Table S1). Groups were not matched on education due to sample size.

Behavioral Data:

A significant group-by-session interaction was observed for the RCBA-2, F(2,8)=8.00, p<0.05, η²=0.67. Stroke participants receiving contingent NFB improved by ~10 points, whereas noncontingent NFB stroke participants showed no improvement; HCs performed near perfect levels at both time points. Planned comparisons of RCBA-2 change scores showed greater improvement in the contingent NFB stroke group relative to HCs, mean group difference in RCBA-2 change =9.75, 95% CI=1.99–17.51, t(6)=3.07, p<0.05 and noncontingent NFB stroke group, mean group difference in RCBA-2 change=11.42, 95% CI=1.12–21.71, t(5)=2.85, p<0.05, with no differences between HC and the noncontingent NFB stroke group, mean group difference in RCBA-2 change= 1.67, 95% CI=-2.32–5.66, t(5)=1.07, p=.33. No other significant group × session interactions were observed (see Table 1 and Table S2).

Table 1.

Behavioral Baseline and Outcome Assessments, Including Means and Standard Deviations, P-values for Planed Comparisons, and Effect Sizes.

Test Name	Testing Session	Participant Group			P-value (Group × Session Interaction)	Effect Size (η2)
		HC (n=4)	Stroke – Contingent NFB (n=4)	Stroke – Noncontingent NFB (n=3)
RCBA (Max=100)	Baseline	98.50 (1.73)	72.5 (16.42)	86.33 (13.43)	0.01*	0.67
	Outcome	98.50 (1.00)	82.25 (17.78)	84.67 (11.68)
WAB (Max=100)	Baseline	99.45 (0.50)	73.38 (31.11)	89.83 (8.54)	0.23	0.28
	Outcome	98.95 (1.11)	78.85 (27.87)	88.83(12.85)
Reading Aloud Words (Max=100%)	Baseline	99.00 (1.41)	76.25 (38.88)	91.33 (8.96)	0.21	0.32
	Outcome	99.50 (0.58)	79.75 (35.21)	90.67 (5.86)
Reading Aloud Nonwords (Max=100%)	Baseline	92.25 (8.50)	54.75 (36.04)	56.67 (20.21)	0.27	0.28
	Outcome	90.25 (9.18)	66.50 (41.23)	60.00 (17.44)
Semantics 2AFC (Max=100%)	Baseline	93.25 (2.36)	86.75 (4.27)	88.67 (5.86)	0.49	0.16
	Outcome	94.50 (4.93)	87.25 (4.11)	84.33 (13.32)
Phonology 2AFC (Max=100%)	Baseline	98.00 (0)	67.00 (20.70)	54.00 (25.87)	0.13	0.40
	Outcome	97.25 (2.99)	82.75 (14.91)	66.67 (20.40)
Orthography 2AFC (Max=100%)	Baseline	97.50 (1.67)	93.33 (4.71)	92.00 (4.01)	0.45	0.18
	Outcome	97.92 (1.60)	96.25 (2.85)	90.33 (3.06)
Manual Dexterity BBT Right Hand	Baseline	57.5 (20.60)	24.25 (17.63)	4.23 (7.32)	0.39	0.21
	Outcome	56.00 (14.54)	29.88 (19.25)	7.33 (12.70)
Grip Force Right Hand	Baseline	29.72 (9.10)	11.85 (8.74)	7.03 (12.18)	0.77	0.06
	Outcome	34.23 (7.22)	12.78 (9.65)	10.10 (17.49)

Note: RCBA – 2AFC – 2 Alternative Forced Choice; BBT – Box and Blocks Test, Reading Comprehension Battery for Aphasia 2^nd ed., WAB – Western Aphasia Battery - Revised. P-values are from group × session repeated-measures analyses (baseline vs. outcome), reported for the interaction.

^*Significant at p<0.05, uncorrected. Effect sizes estimated using eta squared (η²) should be interpreted as follows: Small effect: η²=0.01; Medium effect: η²=0.06; Large effect: η²=0.14.

Across participants, self-rated commitment, motivation, and session difficulty were significantly higher than 1 for all sessions (all p’s < 0.005, Bonferroni-corrected, Table S3). Commitment and motivation were rated as high and the rating increased from Session 1 to Session 3. Difficulty was rated as somewhat easy to neutral across sessions. No significant group differences were found in the NFB signal magnitude from pre- to post-training across groups (F (2,8) = 0.91, p = 0.44, η² = 0.19, Fig. S4).

Task fMRI Learning Runs:

Across training days stroke participants receiving contingent NFB showed increased activation for motor imagery>baseline on day 2 vs. day 1 and day 3 vs. day 1 in the left basal ganglia, central operculum, insula, and left inferior frontal gyrus (IFG) (Table 2). For nonword reading > baseline, activation increased on day 3 vs. day 2 in the left inferior and middle temporal, precuneal, and occipital fusiform cortex, including the targeted left SMG ROI (Fig. 2).

Table 2.

Significant Clusters of Increased Task fMRI Activation from Day 1 to Day 3 of NFB Training

Cluster Index	Cluster Location	Local Maxima (z-value)	MNI Coordinates
			x	y	z
Day 2 > Day 1 Stroke Participants Contingent NFB Condition: Motor Imagery > Baseline
1	Left Caudate Nucleus	4.46	−12	10	12
2	Left Putamen	3.19	−32	−4	0
3	Left Central Opercular Cortex; Left Inferior Frontal Gyrus, pars opercularis	3.05	−52	4	−2
4	Left Insular Cortex	3.04	−34	−4	−4
Day 3 > Day 1 Stroke Participants Contingent NFB Condition: Motor Imagery > Baseline
1	Left Insular Cortex; Frontal Operculum Cortex	4.46	−34	18	4
2	Left Inferior Frontal Gyrus, pars opercularis	4.22	−52	10	0
Day 3 > Day 2 Stroke Participants Contingent NFB Condition: Nonword Reading > Baseline
1	Left Inferior Temporal Gyrus, temporooccipital part	5.47	−40	−56	−2
2	Left Precuneus Cortex	5.01	−30	−56	8
3	Left Occipital Fusiform Gyrus	4.78	−26	−86	−6
4	Left Middle Temporal Gyrus, temporooccipital part, Left Supramarginal Gyrus	4.7	−46	−54	−2
Day 3 > Day 2 Stroke Participants Noncontingent NFB Condition: Motor Imagery > Baseline
1	Left Cingulate Gyrus, posterior division/ Bilateral Precuneus	3.13	−8	−42	6
Day 3 > Day 1 Stroke Participants Noncontingent NFB Condition Word Reading > Baseline
1	Lef Frontal Pole	4.63	−12	64	10
2	Subcallosal Cortex, Right Caudate Nucleus	4.26	2	6	14
Day 3 > Day 2 Stroke Participants Noncontingent NFB Condition Word Reading > Baseline
1	Bilateral Brainstem	3.68	12	−26	−24
2	Left Cingulate Gyrus, anterior division/Frontal Pole	4.22	−16	34	18
Day 2 > Day 1 Stroke Participants Noncontingent NFB Condition Nonword Reading > Baseline
1	Left Postcentral Gyrus	3.73	−56	−30	54
2	Cerebellum (Left V, Vermis VI)	3.73	4	−70	−10
3	Right Lateral Occipital Cortex	3.90	56	−70	2
Day 3 > Day 1 Stroke Participants Noncontingent NFB Condition Nonword Reading > Baseline
1	Subcallosal Cortex/ Bilateral Pallidum, Caudate Nucleus	5.12	2	4	−2
2	Cerebellum (Right V, Vermis VI)	3.62	2	−68	−10
Day 3 > Day 2 Stroke Participants Noncontingent NFB Condition Nonword Reading > Baseline
1	Left Inferior Longitudinal Fasciculus	3.70	−26	−52	28
2	Brainstem	4.04	0	−20	−22
3	Left Frontal Pole	5.46	−26	38	10
Day 3 > Day 2 Healthy Controls: Motor Imagery > Baseline
1	Right Frontal Pole	4.90	28	42	−8
2	Right Inferior Frontal Gyrus, pars triangularis	3.41	44	28	6
Day 3 > Day 2 Healthy Controls: Word Reading > Baseline
1	Left Thalamus/Anterior Thalamic Radiation/ Putamen/ Superior Longitudinal Fasciculus	5.06	−20	−14	18
2	Right Precentral Gyrus/ Superior Longitudinal Fasciculus	4.04	34	−4	30
3	Right Thalamus	3.6	8	0	4
Day 2 > Day 1 Healthy Controls: Nonword Reading > Baseline
5	Left Anterior Thalamic Radiation, Thalamus, Hippocampus	4.29	−36	−44	4
Day 3 > Day 1 Healthy Controls: Nonword Reading > Baseline
2	Left Thalamus	4.45	−12	−14	16

Note: Group-level effects were estimated using FLAME (FMRIB's Local Analysis of Mixed Effects) stages 1 and 2. Z (Gaussianised T/F) statistic images were thresholded using clusters determined by z>1.96 and a (corrected) cluster significance threshold of p=0.05⁶⁸.

Figure 2.

Brain activation increases among stroke participants during nonword reading relative to baseline on day 3 of NFB training compared to day 2. The target ROI used in NFB training is highlighted in green. Z (Gaussianised T/F) statistic images are displayed with a cluster threshold of z > 1.96, corrected to p = 0.05.

In the noncontingent NFB stroke group, imagery>baseline activation increased on day 3 vs. day 2 in the left cingulate gyrus and bilateral precuneus. For word reading>baseline, activation increased on day 3 vs. day 1 in the left frontal pole, subcallosal cortex and right caudate, and on day 3 vs. day 2 in the bilateral brainstem, frontal pole, and left anterior cingulate. Nonword reading>baseline showed increased activation in the cerebellum, left postcentral gyrus, and right lateral occipital cortex on day 2 vs. day 1; in the subcallosal cortex, bilateral basal ganglia and cerebellum on day 3 vs. day 1, and left frontal pole, inferior longitudinal fasciculus and brainstem on day 3 vs. day 2 (Table 2).

In HCs, motor imagery>baseline elicited increased activation in the right frontal pole and IFG on day 3 vs. day 2. Word reading>baseline showed greater activation on day 3 vs. day 2 in the bilateral thalamus, left putamen, and right precentral gyrus. For nonword reading> baseline, activation increased in the left thalamus and hippocampus on day 2 vs. day 1 of training, with additional increases in the left thalamus on day 3 vs. day 1 (Table 2).

Task fMRI Transfer Runs:

No significant activation increases were observed for the contingent NFB stroke group from pre- to post-training. In the noncontingent stroke group, motor imagery> baseline elicited increased activation in the right temporal-occipital fusiform cortex. Word reading>baseline showed increases in the right cerebellum, right precentral gyrus, and left corpus callosum, while nonword reading>baseline activated the left thalamus, anterior thalamic radiation, corticospinal tract, inferior lateral occipital and parietal cortices (including SMG), and bilateral central opercular cortex. In HCs, motor imagery>baseline increased activation in the right frontal pole from pre- to post-training. For word reading>baseline, activity increased in bilateral frontal and subcallosal cortex, and the left pallidum, and for nonword reading>baseline activation increased in the right medial frontal cortex, basal ganglia, left frontal pole and posterior parietal cortex including the SMG. (Table 3).

Table 3.

Significant Clusters of Increased Task fMRI Activation from Pre- to Post-Training (Transfer Runs).

Cluster Index	Cluster Location	Local Maxima (z-value)	MNI Coordinates
			x	y	z
Post-Training > Pre-Training Stroke Participants Noncontingent NFB: Motor Imagery > Baseline
1	Right Temporal-Occipital Fusiform Cortex	3.95	38	−54	−16
Post-Training > Pre-Training Stroke Participants Noncontingent NFB: Word Reading > Baseline
1	Right Cerebellar Lobules I-IV	5.53	8	−40	−24
2	Left Cerebral White Matter/Corpus Callosum	4.47	−4	−10	22
3	Right Precentral Gyrus	3.84	40	−14	54
Post-Training > Pre-Training Stroke Participants Noncontingent NFB: Nonword Reading > Baseline
1	Left Thalamus/Anterior Thalamic Radiation	4.80	−12	−2	−12
2	Right Central Opercular Cortex/Planum Polare	4.78	60	−6	10
3	Left Corticospinal Tract/Anterior Thalamic Radiation	3.96	−8	−30	−22
4	Left Central Opercular Cortex/Supramarginal Gyrus	4.12	−58	−20	22
5	Left Inferior Lateral Occipital Cortex/Angular Gyrus	4.78	−40	−66	20
Post-Training > Pre-Training Healthy Controls: Motor Imagery > Baseline
1	Right Frontal Pole	4.44	30	70	6
Post-Training > Pre-Training Healthy Controls: Word Reading > Baseline
1	Left Frontal Pole	4.58	−16	56	38
2	Subcallosal Cortex	4.61	0	8	−10
3	Right Middle Frontal Gyrus	4.28	30	28	52
4	Left Pallidum	3.7	−10	0	−8
Post-Training > Pre-Training Healthy Controls: Nonword Reading > Baseline
1	Right Caudate Nucleus/Putamen/White Matter	4.73	16	8	10
2	Left Frontal Pole	5.91	−24	42	−20
3	Left Postcentral Gyrus/Supramarginal Gyrus	4.34	−10	−30	58
4	Right Juxtapositional Lobule (formerly Supplementary Motor Cortex)/ Superior Frontal Gyrus	4.43	4	−8	48

Note: Group-level effects were estimated using FLAME (FMRIB's Local Analysis of Mixed Effects) stages 1 and 2. Z (Gaussianised T/F) statistic images were thresholded using clusters determined by z>1.96 and a (corrected) cluster significance threshold of p=0.05⁶⁸.

Resting State fMRI:

Permutation testing confirmed significant dynamic connectivity changes (p=0.01), with Pearson correlation differences (Δr) ranging from −0.10 to 0.10. From day 1 to day 3, intrahemispheric dFC increased bilaterally between sensorimotor hand areas (SM hand) and auditory and cingulo-opercular networks. Interhemispheric dFC increased in SM hand networks and between left SM hand and right cingulo-opercular networks. This pattern was present in both HC and contingent NFB stroke participants but was less extensive in the stroke group (Fig. 3). The noncontingent NFB stroke group showed a less organized pattern, with intrahemispheric connectivity restricted to the left hemisphere and preserved SM hand interhemispheric connectivity but no distinct long-range increases on day 3 vs. day 1 (Fig. 3).

Figure 3.

Dynamic resting state functional connectivity across 333 Gordon atlas ROI, comparing brain connectivity following NFB training on day 3 to pretraining resting state scans on day 1. Top 1% of all significant connections are shown. Network labels are from Gordon et al ³⁷ parcellation.

Reductions in dFC also differed by group (Fig. 3). In HCs, connectivity decreased within visual cortex and between SM hand, fronto-parietal, cingulo-opercular, and default networks both intra- and interhemispherically. Contingent NFB stroke participants showed broader decreases across fronto-parietal, dorsal attention, default, and cingulo-parietal networks, without reductions in visual cortex connectivity. Noncontingent NFB stroke group exhibited widespread disorganized decreases spanning ventral attention, cingulo-opercular, visual dorsal attention, splenial temporal, salience, and default mode network.

4. Discussion

Reading deficits are common in aphasia and substantially limit daily functioning³⁸. We hypothesized that rtfMRI NFB could enhance post-stroke reading recovery by targeting the left SMG, a key region for orthography-phonology conversion^24,39, which is frequently impaired in aphasia^1,17,40. Consistent with this rationale, reduced CBF in the SMG correlates with diminished language abilities¹⁶.

This paper reports the methodology and preliminary feasibility data for an NFB-based reading intervention, focusing on participant engagement, outcome measure performance, and initial behavioral and neural responses to training.

4.1. Behavioral Results

Preliminary behavioral results indicate improved reading comprehension in stroke participants receiving contingent NFB, as shown by marked improvements on the RCBA-2 with a large effect size. The RCBA includes sentence and paragraph reading, suggesting generalization beyond single word reading. In contrast, the HC group performed with consistently high accuracy, suggesting potential ceiling effects across baseline and outcome assessments, while the noncontingent NFB stroke group showed no improvement, with a small nonsignificant decline in RCBA-2 scores. These differences were confirmed in planned comparisons allowing us to rule out spontaneous recovery as an explanation for improvement in this subacute stroke sample. These patterns will be further evaluated in the fully powered trial.

Although group-level increases in NFB signal magnitude were observed from pre- to post-training, these changes were not statistically significant and showed considerable interindividual variability (Fig. S4). Similar signal changes across groups, regardless of feedback contingency, prevent attributing these effects to the intervention. Thus, NFB signal magnitude during transfer runs does not provide strong evidence of training-related modulation in this small sample. Behavioral improvements in the contingent NFB stroke group therefore cannot be explained by volitional NFB control alone but may instead reflect the combined influence of contingent feedback and repeated task practice during homework sessions. These findings underscore the potential value of pairing neurofeedback with structured behavioral practice and highlight important considerations for future studies seeking to isolate the specific effects of feedback contingency.

Self-report measures of commitment, motivation, and perceived session difficulty supported intervention feasibility. Participants rated commitment and motivation as high across sessions, with slight increases over time, indicating sustained engagement. Session difficulty was rated as somewhat easy to neutral, with a slight increase in perceived difficulty by the third session, indicating that the tasks were challenging yet manageable.

4.2. Task fMRI Results

Task fMRI Learning Runs:

To evaluate training-related changes, we contrasted brain activity across training days. In stroke participants receiving contingent NFB, motor imagery>baseline elicited increased activation on day 2 vs. day 1 in the left basal ganglia, IFG, pars opercularis, central opercular cortex (extending into the auditory cortex), and insula, with similar increases in the insula and IFG on day 3 vs. day 1. Several of these regions are central to language processing. The left IFG pars opercularis plays a key role in orthography-phonology mapping, in a difficulty sensitive manner. For example, nonwords – being less familiar and more challenging to process – activate this area more than real words²⁴, and lesions in this region and the left insula are linked to difficulty reading nonwords⁴¹. Similar activation patterns have been reported during silent reading of motion verbs⁴², real and imagined stuttering⁴³, and auditory processing of meaningful sounds^44,45. The CBF in the frontal portion of this cluster has been linked to flexing and extending fingers of the right hand⁴⁶. Together, these findings suggest that contingent NFB training engaged overlapping language and motor-related regions during imagery potentially reflecting shared neural substrates for articulatory planning, language, and imagined movement, a connection that may emerge with writing expertise.

For contingent NFB stroke participants, we also observed increased activation during nonword reading>baseline for day 3 vs. day 2 in the left inferior and middle temporal gyri (ITG, MTG), precuneus, and occipital fusiform. The ITG and MTG link conceptual information with word forms, as shown by lesion symptom-mapping^47,48, while the fusiform cortex supports written word-form processing and letter-sound mapping^49,50. Increased precuneus activation may reflect its role in reading and attention, as gray matter volume in this area predicts performance in both domains⁵¹. Thus, as training progressed, contingent NFB stroke group showed increased recruitment of the left reading network including SMG, targeted by NFB training. Although the NFB signal reflected activation during both motor imagery and reading, training-related increases in SMG activation emerged only during nonword reading, suggesting that nonword processing may be a stronger driver of SMG modulation. While task-specific contribution to outcomes cannot be isolated, these findings highlight nonword reading as a potentially effective task for SMG-focused NFB.

In the noncontingent NFB stroke group, widespread task-related activation increases were observed across fMRI runs despite the absence of volitional control. Because participants viewed randomized feedback but completed the same motor imagery and reading homework as other groups these changes are unlikely to reflect feedback contingency and may instead arise from repeated task practice inside and outside the scanner. Motor imagery>baseline on day 3 vs. day 2 elicited greater activity in the left anterior cingulate gyrus and bilateral precuneus, regions implicated in internally guided action, self-referential processing and first-person imagery ^52,53. During word reading, activation increased in the left frontal pole, anterior cingulate, subcallosal cortex, and right caudate nucleus, associated with evaluation of internally generated information⁵⁴, goal directed actions⁵⁵, feedback learning⁵⁶, and effortful word retrieval and response suppression⁵⁷, possibly reflecting rising cognitive engagement with training. Nonword reading produced the most extensive activation increases, involving cerebellar, sensorimotor, and visual association networks. This widespread activation may reflect the growing cognitive and perceptual demands of nonword reading and greater reliance on multisensory integration. Overall, these findings highlight the influence of task practice on neural activation independent of feedback contingency.

For HCs, motor imagery>baseline on day 3 vs. day 2 elicited greater activation in the right IFG, pars triangularis, and right frontal pole, regions associated with evaluation and maintenance of internally generated information⁵⁴, consistent with the noncontingent NFB condition for the same contrast. During word reading, activation increased in the bilateral thalamus and right precentral gyrus on day 3 vs. day 2, regions linked to successful reading aloud ⁵⁰, and visuomotor imagery ⁵⁸, suggesting combined reading and motor imagery engagement. Nonword reading elicited increased activity in the left thalamus and hippocampus on day 2 vs. day 1, with additional thalamic increases on day 3 vs. day 1. The thalamus is implicated in attentional and phonological processing⁵⁹, while the hippocampus responds preferentially to novel pseudowords.⁶⁰ These findings suggest that NFB training in HCs engaged neural circuits supporting reading, cognitive control, and visuomotor imagery.

Task fMRI Transfer Runs:

Pre- and post-training fMRI runs were identical to the training runs but without feedback. We found no significant increases in brain activity in the contingent NFB stroke group, suggesting that training-related changes may have been insufficiently robust to persist without feedback, possibly due to limited training dose or sample size. By contrast, the noncontingent NFB stroke group showed widespread activation increases, including the right temporal-occipital fusiform cortex during motor imagery; the right cerebellum, precentral gyrus, and left corpus callosum during word reading; and the left thalamus, anterior thalamic radiation, corticospinal tract, inferior lateral occipital and parietal cortices (including SMG), and bilateral central opercular cortex during nonword reading. These findings suggest that repeated task exposure alone can modulate the engagement of sensory, motor, and phonological networks. The absence of consistent changes in the contingent condition highlights the need for larger well-powered studies and systematic dose-intensity investigations, to determine whether contingent NFB produces lasting neural changes.

In HCs, all of whom received contingent NFB, brain activity increased during motor imagery and word and nonword reading relative to baseline, largely outside of the canonical language network. Motor imagery showed activation increases in the right frontal pole, indicating evaluation and manipulation of internally generated information. Word reading activated the left frontal pole, pallidum, right middle frontal gyrus, and the subcallosal cortex, implicated in processing unfamiliar text⁶¹, task switching⁶², goal maintenance⁶³ and control of internal thought⁶⁴. Nonword reading increased activation in the left postcentral gyrus and the targeted SMG ROI. These results suggest that NFB training in HCs engaged diverse cognitive processes and selectively recruited phonological components during the demanding nonword task.

4.3. Resting State fMRI - Dynamic dFC Results

Following training both contingent NFB stroke group and HCs showed increased dFC of the SM hand network³⁷, which supports motor and sensory processing of hand movements and is commonly engaged during motor imagery.⁶⁵ Connectivity between the SM hand network and cingulo-opercular and auditory networks increased, suggesting that practicing hand motor imagery prior to reading strengthened functional coupling among motor, auditory and reading-related systems. In contrast, decreased intra- and interhemispheric connectivity was observed in fronto-parietal, dorsal attention, default, and SM hand networks, consistent with prior work linking imagined motor sequence learning to reduced default mode connectivity⁶⁶, possibly reflecting decreased introspection. HCs also showed decreased visual cortex connectivity aligning with evidence that imagery practice reduces interference from visual inputs and supports internally generated representations⁶⁷. This effect was absent in stroke survivors possibly reflecting greater difficulty with motor imagery. Although overall connectivity patterns were consistent across groups, stroke survivors had fewer connections, potentially indicating reduced network flexibility and learning, consistent with the absence of transfer effects in the contingent NFB stroke group.

In contrast, noncontingent NFB stroke group showed a less organized pattern of dFC changes. Intrahemispheric connectivity between SM hand, auditory, and cingulo-opercular networks was limited to the left hemisphere, and although interhemispheric SM hand connectivity was preserved, no distinct long-range connections emerged. This disorganized pattern, together with widespread dFC decreases across ventral attention, visual, default mode, and salience networks, may reflect reduced training specificity in the absence of a coherent NFB signal.

Thus, dFC results align with prior motor imagery studies and indicate that NFB modulated brain networks supporting motor imagery, language, and cognitive control. Group differences in dFC strength and organization suggest both reduced post-stroke plasticity and the importance of contingent feedback for coordinated network adaptation.

5. Conclusion

This study shows the feasibility of real-time fMRI NFB for post-stroke reading rehabilitation. Stroke participants receiving contingent NFB targeting the left SMG demonstrated improved reading comprehension, suggesting generalization beyond single-word processing. Neuroimaging revealed increased recruitment of phonological, language, and motor imagery regions, with nonword reading especially effective at modulating SMG. Resting-state analyses showed strengthened motor–auditory–language connectivity in contingent groups, whereas noncontingent participants exhibited diffuse changes, underscoring the importance of feedback contingency. Although transfer effects were not detected, these findings support NFB’s potential to enhance language network engagement during stroke recovery.

6. Limitations and Future Directions

To date, only five fMRI NFB stroke rehabilitation studies have been published, including two in aphasia with 2–8 stroke participants each. This study contributes to this emerging field by demonstrating the feasibility of fMRI NFB for post-stroke reading rehabilitation. However, findings should be interpreted cautiously given the small sample and interindividual variability. One participant had an isolated left cerebellar lesion; although cerebellar strokes are not classically associated with aphasia, most lesion-symptom mapping studies exclude infratentorial lesions, limiting inferences about cerebellar contributions to language. Considering evidence for cerebellar–cortical language interactions, this participant’s symptoms may reflect disrupted language network connectivity. Baseline differences in reading ability across groups also cannot be ruled out as an alternative explanation for observed effects in this small sample. Completion of the ongoing randomized controlled trial will be essential to determine whether these preliminary behavioral and neural effects are replicated in a larger sample. Beyond the current trial, additional studies incorporating sham NFB conditions, masking of participants and assessors, and standard-of-care or active control groups will be essential to establish whether fMRI NFB offers benefits beyond existing therapies, and whether deficit severity modulates responsiveness to NFB. To improve accessibility, future research should explore more portable and cost-effective NFB platforms (e.g EEG or fNIRS). Longer-term follow-up will be needed to assess durability of behavioral gains, and inclusion of writing and spelling assessments would clarify whether improvements in phonological processing generalize to written skills. Despite these limitations, this research marks an important preliminary step toward biologically informed reading interventions for aphasia.

Supplementary Material

Supplemental Methods

Supplemental Results

Tables S1–S3

Figures S1–S4

References 1–79

Abbreviations

2AFC

Two-alternative forced choice

AQ

Aphasia Quotient

BET

Brain Extraction Tool

CBF

Cerebral blood flow

CI

Confidence interval

CONSORT

Consolidated Standards of Reporting Trials

CSF

Cerebrospinal flui

dFC

Dynamic functional connectivity

EPI

Echo-planar imaging

FEAT

FMRI Expert Analysis Tool

FLAIR

Fluid-attenuated inversion recovery

FLAME

FMRIB’s Local Analysis of Mixed Effects

FLIRT

FMRIB’s Linear Image Registration Tool

fMRI

Functional magnetic resonance imaging

FNIRT

FMRIB’s Nonlinear Image Registration Tool

FOMRI

Fiber Optic Microphone (Optoacoustics)

FSL

FMRIB Software Library

FSLMEANTS

FSL tool for extracting mean time series

FWHM

Full width at half maximum

HC

Healthy control

ICA-AROMA

Independent Component Analysis–Automatic Removal Of Motion Artifacts

IFG

Inferior frontal gyrus

IRB

Institutional Review Board

ITG

Inferior temporal gyrus

LASSO

Least absolute shrinkage and selection operator

MCFLIRT

Motion Correction using FMRIB’s Linear Image Registration Tool

MD

Mean difference

MI

Motor imagery

MNI

Montreal Neurological Institute

MTG

Middle temporal gyrus

NFB

Neurofeedback

OpenNFT

Open-source neurofeedback training platform

RCBA-2

Reading Comprehension Battery for Aphasia, 2nd edition

ROI

Region of interest

rs-fMRI

Resting-state functional magnetic resonance imaging

rsFC

Resting-state functional connectivity

RCT

Randomized controlled trial

rtfMRI

Real-time functional magnetic resonance imaging

SD

Standard deviation

SM

Sensorimotor

SMG

Supramarginal gyrus

SPD

Symmetric positive definite

TE

Echo time

TI

Inversion time

TR

Repetition time

WAB-R

Western Aphasia Battery–Revised

WM

White matter