Integrated Neurorehabilitation Including EEG-Neurofeedback for Unilateral Spatial Neglect After Right Thalamo-Mesencephalic Intracerebral Hemorrhage: A Case Report

1. Introduction

Unilateral spatial neglect (USN) is a disabling neuropsychological syndrome commonly observed after stroke, characterized by a reduced ability to detect, orient toward, explore, or respond to stimuli located in the space contralateral to the brain lesion. This condition cannot be explained solely by primary sensory or motor deficits and may involve perceptual, attentional, motor-intentional, representational, and behavioral components. From a clinical perspective, USN is particularly relevant because it negatively affects functional recovery, autonomy in activities of daily living, and rehabilitation outcomes after stroke [1]. With a prevalence reaching up to 50% after a right-hemisphere stroke, USN remains a primary driver of long-term disability and poor functional recovery [2].

Being most frequently associated with right-hemisphere lesions, USN underlies cortical fronto-parietal and temporo-parietal network disruptions, affecting spatial orienting and attentional control. However, neglect should not be considered a purely cortical syndrome. Increasing evidence supports a network-based model in which cortical and subcortical structures jointly contribute to spatial attention, arousal, behavioral regulation, and interhemispheric balance. Thus, the USN clinical phenotype arises from disrupted connectivity between fronto-parietal attentional systems and vigilance networks [2] and its clinical recovery relies heavily on restoring balanced interhemispheric dynamics [3].

In this framework, the thalamus has a relevant role not only as a relay station, but also as an active node within cortico-subcortical circuits supporting attention and spatial exploration. Recent evidence suggests that USN can emerge following isolated right thalamic lesions, likely due to the aforementioned functional disruption of broader network nodes [4]. Within these subcortical circuits, lesions involving hubs like the mediodorsal nucleus can significantly impair goal-directed attention and cognitive flexibility, suggesting that disruption of talamo-cortical loops can compromise executive control [5].

Rehabilitation of USN remains challenging. Several approaches have been proposed, including visual scanning training, prism adaptation, limb activation, optokinetic stimulation, non-invasive brain stimulation, virtual reality-based interventions, and compensatory strategies [1]. Nevertheless, treatment response is variable, and patients with complex post-stroke presentations often require multidisciplinary rehabilitation programs integrating motor, cognitive, language, and functional interventions. This is particularly relevant in real-world neurorehabilitation settings, where clinical improvement usually reflects the interaction of multiple therapeutic components rather than the isolated effect of a single technique. Current strategies often combine top-down and bottom-up approaches, such as visual scanning, compensatory strategy training, and technology-assisted interventions, to promote environmental exploration and functional adaptation [2].

In recent years, EEG-based neurofeedback has been proposed as a potential adjunctive approach for post-stroke neurorehabilitation. Neurofeedback provides patients with real-time information about their own brain activity, allowing them to progressively learn to modulate specific neurophysiological patterns through implicit or explicit self-regulation strategies. Concerning spatial neglect, preliminary studies have focused particularly on alpha-band activity, given its relationship with attentional control, cortical excitability, and interhemispheric dynamics. Ros et al. [6] reported that modulation of alpha-rhythm dynamics was associated with improvement in USN symptoms, while Saj et al. [7] discussed the potential of real-time neuromodulation to address network-level dysfunction in spatial neglect. However, current evidence remains preliminary, and further clinical reports are needed.

While preliminary studies have primarily targeted posterior parietal networks to modulate alpha rhythms [6] or reduce interhemispheric imbalance via brain–computer interface (BCI)-based approaches, which translate brain activity patterns into control signals or feedback to support rehabilitation [3,8], central electrode sites (e.g., Cz) may represent an alternative target for supporting broader attentional stability and sensorimotor regulation. In a multimodal rehabilitation context, a central electrode placement may contribute to the optimization of cortical vigilance and sensorimotor regulation rather than aiming at direct influence of deep subcortical structures.

This case report describes the clinical, neuropsychological, and functional evolution of a patient with USN following right thalamic hemorrhagic stroke who underwent an intensive multidisciplinary rehabilitation program including physiotherapy, speech and language therapy, conventional cognitive stimulation, and adjunctive EEG-NF. The primary objective was to document the feasibility, tolerability, and clinical evolution of the patient throughout treatment, without drawing causal conclusions regarding the isolated therapeutic impact of the neurofeedback protocol. A secondary exploratory objective was to characterize pre–post changes in resting-state spectral power and sensor-level functional connectivity, assessed using the weighted Phase Lag Index, as potential neurophysiological correlates of the patient’s clinical evolution within the integrated rehabilitation program. These electrophysiological analyses were descriptive and were not intended to test a treatment-specific neurophysiological mechanism.

2. Materials and Methods

2.1. Case Presentation

On 18 March 2026, a 70-year-old man was admitted to the emergency department following the acute onset of neurological symptoms consistent with a cerebrovascular event.

Symptoms developed abruptly during lunch with his family: the patient presented with sudden weakness in his left hand, rapidly followed by loss of trunk control with left-sided deviation and deviation of the corner of the mouth. The regional emergency system was activated, the stroke pathway was triggered, and the patient was transferred to the Stroke Unit.

He had completed 11 years of education and was retired. His vascular risk profile was notable: untreated hypertension, hypercholesterolaemia, hyperhomocysteinemia due to folate deficiency, and hyperuricaemia. He was a current smoker (approximately 10 cigarettes per day) and reported moderate alcohol consumption (one glass of wine with meals). A family history of stroke was present (father deceased from stroke). He reported no allergies and was not taking any medication.

On admission, the patient was alert, cooperative and able to answer questions. Neurological examinations revealed left homonymous lateral hemianopia, moderate dysarthria, left facio-brachio-crural hemiplegia and left hemianesthesia.

An emergency non-contrast cranial computed tomography scan (CT) showed an intraparenchymal haematoma in the right thalamus with compression of the third ventricle. Serial imaging confirmed haemorrhagic stability, while contrast-enhanced CT on 19 March better delineated perifocal oedema around the right thalamic-mesencephalic haemorrhagic focus, without significant interval change.

The patient was closely monitored throughout his stay in the Stroke Unit. Brain Magnetic Resonance Imaging (MRI) on 23 March (Figure 1) confirmed a right thalamomesencephalic haemorrhagic focus with perifocal oedema extending to the superior cerebellar peduncle, pons and right optic tract. The haematoma had opened into the ventricular system, with haemorrhagic traces visible particularly in the left occipital horn. Furthermore, a less recent haemorrhagic focus was identified in the right temporo-basal region adjacent to the lateral ventricle, with a haemosiderin rim and minimal peripheral enhancement. Leukoaraiosis of the periventricular white matter was noted, and Susceptibility Weighted Imaging (SWI) sequences revealed punctate haemosiderin foci in the bilateral parietal subcortical regions and median pons, also raising the possibility of capillary telangiectasia. Mild cortical atrophy was reflected in the widening of cortical sulci and basal cisterns.

By the time of discharge on 26 March, the patient had clinically stabilised, though meaningful deficits persisted: moderate left hemiparesis, left hemihypoesthesia and gaze deviation consistent with a right third cranial nerve deficit. Accordingly, intensive post-acute inpatient rehabilitation was deemed necessary, and the patient was transferred to the Neuromotor Rehabilitation Unit of the IRCCS Maugeri Institute in Bari to begin a structured programme aimed at motor recovery. At admission to inpatient neurorehabilitation, the patient was alert but drowsy, cooperative, and partially oriented. Reduced awareness of his motor and functional deficits, increased distractibility, and limited spontaneous engagement were observed. Neuromotor examination showed left facio-brachio-crural hemiparesis, with minimal voluntary activation of the left upper limb and more preserved recruitment in the left lower limb. Muscle strength ranged from 0 to 2 on the left upper limb and from 2 to 3 on the left lower limb, whereas right-sided strength was relatively preserved. Left-sided tactile, nociceptive, and proprioceptive hypoesthesia was present. Additional findings included a left central facial deficit, right eyelid ptosis, limitation of upward gaze, abnormalities of saccadic movements and convergence, left homonymous hemianopia, and mild dysarthria.

Head and trunk control and sitting balance were severely impaired. Standing and gait were not initially achievable, and the Timed Up and Go test was not executable. The patient required complete assistance with transfers, personal hygiene, postural transitions, and basic activities of daily living. Baseline functional scores were 17/91 on the FIM motor subscale, 19/35 on the FIM cognitive subscale, 36/126 on the FIM total score, and 5/100 on the Modified Barthel Index. Dysphagia predominantly affecting the oral phase was also present.

At the baseline neuropsychological assessment on 23 April 2026, the patient appeared drowsy but arousable, cooperative, and partially oriented in space and time. Psychomotor slowing, reduced sustained attention, and limited awareness of motor and functional deficits were evident. Speech was mildly dysarthric but intelligible, and verbal comprehension was sufficiently preserved for assessment. Severe left visuospatial neglect involved personal, peripersonal, and extrapersonal space and was characterized by markedly asymmetric exploration, systematic omission of left-sided stimuli, impaired reading and visuoconstructive performance, and the need for continuous external facilitation during testing and functional activities.The patient remained hospitalized until discharge on 8 June 2026. The main clinical and rehabilitation time points are summarized in Figure 2.

2.2. Multidisciplinary Rehabilitation Program

The study was conducted at the Neurorehabilitation Unit of ICS Maugeri IRCCS, Bari, Italy. The patient was admitted for intensive post-acute neurorehabilitation following a right thalamic hemorrhagic stroke and presented with clinically significant USN, together with associated motor, functional, and cognitive difficulties.

Before treatment initiation, the patient underwent a multidimensional baseline evaluation, including neuromotor and functional assessment, neuropsychological testing with neglect-specific measures, and resting-state quantitative EEG (qEEG) recording. These assessments were used to characterize the patient’s initial clinical profile and to establish pre-intervention baseline across behavioral, cognitive, electrophysiological, and functional domains.

The patient received an integrated multidisciplinary rehabilitation program including physiotherapy, speech and language therapy, conventional cognitive stimulation, and EEG-based neurofeedback (EEG-NF). Physiotherapy focused on improving progressive mobilization, postural control, motor coordination, balance, endurance, and functional independence. Treatment included hemiplegia-oriented interventions targeting trunk control and gait recovery, with emphasis on left-sided hemiplegia and progressive restoration of weight-bearing symmetry. The program followed sequential phases: bed positioning and mobilization of the left hemibody, postural management and limb strengthening with cycle ergometer/Motomed use; transition to upright posture using standing devices with active-assisted upper limb exercises; and gait training with gradual reduction of assistance. Speech and language therapy addressed dysphagia rehabilitation and swallowing function, with progression toward a soft solid diet and recovery of independent feeding. Communication support was also provided. Treatment included electrical stimulation and specific swallowing techniques, together with caregiver training and supervision to ensure safety and generalization to daily life. Conventional cognitive stimulation targeted visual exploration, sustained attention, contralesional awareness, and compensatory strategies for USN. The program included specific training, computerized exercises, ecological group activities, and paper-and-pencil tasks involving complex scene analysis, aimed at promoting transfer to real-life functioning. EEG-NF was integrated as an adjunctive component of the multidisciplinary program to support cortical self-regulation, attentional stability, sensorimotor regulation, and functional engagement during recovery [7].

The frequency, duration, and main objectives of each rehabilitation component are summarized in Table 1, while the detailed EEG-NF parameters, including electrode placement, target frequency bands, feedback modalities, threshold adjustment, and safety monitoring, are reported in Section 2.2.1 and Table 2. At the end of the intervention period, the patient underwent repeated neuromotor, functional, neuropsychological, and qEEG assessments, to document pre–post evaluation clinical changes across neglect-related, cognitive, electrophysiological, and functional outcomes.

As part of the integrated multidisciplinary rehabilitation program, EEG-NF training was administered using the ProComp5 system together with BioGraph Infiniti software (Thought Technology, Montreal, QC, Canada), in accordance with established EEG biofeedback procedures. Electrode placement followed the international 10–20 system. The active electrode was positioned at Cz, with reference and ground electrodes placed at left and right ears respectively. Electrode impedance was maintained below 5 kΩ throughout all sessions to ensure adequate signal quality and reliability.

The protocol was individualized according to the patient’s clinical presentation, baseline EEG findings, and rehabilitation goals. Training was delivered at Cz and simultaneously reinforced activity in the 12–15 Hz band while inhibiting activity in the 4–7.5 Hz and 22–26 Hz bands. Thus, all three frequency components were continuously monitored and incorporated into the feedback contingencies during each training session.

Each session lasted approximately 35 min and was administered 5 days a week for 3 weeks, for a total of 15 sessions (first session: 4 May 2026; last session: 22 May 2026). During training, the patient received real-time visual and auditory feedback through a gamified interface, designed to enhance engagement and reinforce the desired EEG patterns. Feedback thresholds were adaptively adjusted by the clinician during each session to maintain an appropriate level of difficulty and responsiveness. In general, threshold regulation aimed to sustain effective reinforcement of the target rhythm while limiting excessive activity in the inhibited frequency bands.

All sessions were supervised by trained personnel and monitored for fatigue, discomfort, artifacts, reduced engagement, and adverse events. The intervention was conceived as an adjunctive component of the broader multidisciplinary rehabilitation program, rather than as a stand-alone treatment.

Full neurofeedback parameters are summarized in Table 2.

2.3. Outcome Measures

Baseline clinical, neuropsychological, and functional assessments were conducted on 23 April 2026, and the baseline resting-state EEG was acquired on 24 April 2026. Post-treatment clinical, neuropsychological, functional, and resting-state EEG assessments were conducted on 25 May 2026. Outcomes were organized into five domains: treatment feasibility and tolerability; cognitive and affective functioning; neglect-specific and ecological functioning; neuromotor and functional independence; and exploratory electrophysiological outcomes.

The primary clinical outcomes were changes in neglect-specific performance, ecological neglect-related disability, functional independence, and EEG-NF feasibility and tolerability. Secondary outcomes included global cognitive functioning, executive-attentional performance, affective symptoms, gait feasibility, and exploratory resting-state qEEG measures.

2.3.1. Clinical, Neuropsychological, Neglect-Specific, and Functional Assessment

Global cognitive functioning was assessed using the Montreal Cognitive Assessment (MoCA). Executive-attentional functioning was examined using the Frontal Assessment Battery (FAB), Clock Drawing Test, Digit Span Forward and Backward, phonemic and semantic verbal fluency, the Italian short Stroop Test, and the Arrow Stroop Test. Affective symptoms were assessed using the anxiety and depression subscales of the Hospital Anxiety and Depression Scale (HADS).

Neglect-specific assessment included the spatial asymmetry measure of the Oxford Cognitive Screen, the Bells Cancellation Test and Letter Cancellation Test, including detected targets and Center of Cancellation, the Timed Neglect Test, the Baking Tray Task, star cancellation, Diller letter cancellation, article reading, and object search in extrapersonal space. Ecological manifestations of neglect during activities of daily living were assessed using the Catherine Bergego Scale.

Neuromotor status was characterized through clinical examination of muscle strength, voluntary motor control, somatosensory functioning, cranial nerves, postural control, balance, and gait. Functional independence was assessed using the Functional Independence Measure (FIM), including total, motor, and cognitive scores, and the Modified Barthel Index. Mobility and gait feasibility were assessed using the Timed Up and Go test whenever executable.

EEG-NF feasibility was evaluated according to the number of completed sessions relative to the 15 sessions planned, treatment adherence, and the occurrence of treatment-related interruptions. Tolerability was assessed through clinical monitoring during and immediately after each session for fatigue, discomfort, headache, dizziness, nausea, agitation, reduced engagement, or sustained worsening of neurological or cognitive symptoms. Any adverse event and its possible relationship with treatment were recorded.

2.3.2. EEG Acquisition and Preprocessing

Resting-state EEG was recorded at baseline and after completion of the rehabilitation program using a Mitsar-201 EEG system and 19 Ag/AgCl electrodes positioned according to the international 10–20 system. Recordings were obtained under both eyes-open (EO) and eyes-closed (EC) conditions for approximately 6.3 min per condition, at a sampling rate of 250 Hz. Electrode impedance was maintained below 5 kΩ whenever possible.

EEG preprocessing was performed in MATLAB R2025b, version 25.2 (MathWorks, Natick, MA, USA), using EEGLAB v2026.0.0. Only the 19 standard EEG channels were retained, and standard electrode locations were assigned using the EEGLAB standard_1005 coordinate file. Continuous data were band-pass filtered between 1 and 80 Hz, and 50-Hz line noise was attenuated using a 48–52 Hz notch filter.

After visual identification and removal of gross artefactual segments, data were re-referenced to the average reference. Extended Infomax independent component analysis was then performed with rank correction. Components were classified using ICLabel and visually reviewed before rejection. Components were removed only when their topography, time course, and classification were consistent with ocular, muscular, or other non-neural artefactual activity. No channels required removal or interpolation. Following component rejection, the data were again re-referenced to the average reference and retained for spectral and connectivity analyses.

2.4. Data Analysis

2.4.1. Clinical Outcome Interpretation

Baseline and post-treatment values were reported together with their absolute change, calculated consistently as post-treatment minus baseline. Given the single-case design, no group-level inferential statistics were performed.

Individual change was examined using the Reliable Change Index (RCI), according to Jacobson and Truax [14], whenever numerical baseline and post-treatment observations and suitable psychometric parameters were available:

RCI = (X_post − X_pre)/S_diff,

where S_diff represents the standard error of the difference. When a published 95% minimal detectable change was available, S_diff was obtained by dividing the minimal detectable change by 1.96. RCI values were oriented so that positive values indicated change in the clinically favorable direction. Values of at least +1.96 were classified as reliable improvement, whereas values of −1.96 or lower were classified as reliable deterioration.

Psychometric parameters were selected from the most closely matched available validation or reliability studies, prioritizing stroke-specific evidence whenever available. Estimates derived from the same instrument and an appropriate clinical population were considered directly applicable or compatible. Estimates based on different versions, administration modalities, languages, or clinical populations were considered indirect. Proxy-based RCI calculations were retained only as exploratory sensitivity analyses and were not interpreted as confirmatory evidence. Complete psychometric parameters, applicability grades, calculations, and references are reported in Supplementary Table S1.

For the FAB, raw scores were converted to age- and education-adjusted scores solely for normative classification according to Aiello et al. [15]. Because age and education were unchanged between assessments, this adjustment did not affect the magnitude of the raw pre–post difference. An RCI was not calculated because an appropriate test–retest reliability coefficient was unavailable.

RCI findings were integrated with published clinical interpretability criteria. Changes in MoCA, FIM, and Modified Barthel Index scores were compared with available minimal clinically important difference or minimal detectable change thresholds [9,10,11,12,23]. Center of Cancellation values were compared with the pathological cut-offs of 0.081 for the Bells Cancellation Test and 0.083 for the Letter Cancellation Test [13]. HADS and Catherine Bergego Scale scores were interpreted according to their corresponding clinical ranges.

2.4.2. Exploratory qEEG Analysis

Spectral and connectivity analyses were performed in MATLAB using custom scripts applied to the final EEGLAB-processed datasets. Before spectral estimation, each continuous EEG channel was linearly detrended. Power spectral density was estimated using Welch’s method with 2-s symmetric Hann windows, 50% overlap, and a 1024-point fast Fourier transform. Within each segment, the mean was removed before windowing, and the resulting one-sided power spectral density estimates were averaged across segments. To provide a single and consistent spectral metric, absolute power was selected for the analyses reported in the manuscript.

Absolute band power was calculated by trapezoidal integration of the power spectral density within the following non-overlapping frequency bands: delta (1–4 Hz), theta (4–7.5 Hz), alpha (8–12 Hz), sensorimotor rhythm (SMR; 12–15 Hz), beta-1 (15–20 Hz), beta-2 (20–29 Hz), and low-gamma (29–45 Hz). SMR corresponded to the EEG-NF reward band, whereas beta-2 included the 22–26 Hz inhibitory range used during training.

For each band and recording condition, absolute change was calculated as post-treatment minus baseline, and percentage change was calculated relative to the baseline value. EO and EC conditions were analyzed separately. Frequency-resolved power spectral density was summarized across all 19 electrodes and separately at Cz, the active EEG-NF training site.

To reduce overemphasis on single-electrode variations, band-power values were also summarized across five scalp regions: frontal (Fp1, Fp2, F7, F3, Fz, F4, and F8), central (C3, Cz, and C4), temporal (T7, T8, P7, and P8), parietal (P3, Pz, and P4), and occipital (O1 and O2). Because electrodes within a single participant cannot be considered statistically independent observations, no inferential statistics were performed on electrode-level or regional values. Analyses focused on the direction, magnitude, and spatial distribution of post-treatment changes.

A sensitivity analysis was performed to determine whether spectral findings were driven by residual high-amplitude segments. Each final recording was divided into non-overlapping 2-s epochs. For each recording, a robust z-score was calculated from epoch-level log-transformed broadband power as the difference between each value and the median, divided by 1.4826 times the median absolute deviation. Epochs with a robust z-score greater than 3.5 were classified as high-amplitude outliers and excluded. Spectral estimates were then recalculated using the retained epochs and summarized using median values.

As a secondary exploratory analysis, sensor-level functional connectivity was estimated using the weighted Phase Lag Index (wPLI), selected to reduce the influence of zero-lag synchronization potentially associated with volume conduction. Connectivity estimation used 2-s Hann-windowed segments, 50% overlap, and a 1024-point fast Fourier transform. Each channel was mean-centered within each segment before windowing. For each channel pair and frequency band, the imaginary component of the cross-spectral density was pooled across segments and frequency bins within the band. wPLI was calculated as the absolute mean of the imaginary cross-spectrum divided by the mean of its absolute value.

Regional connectivity values were obtained by averaging pairwise wPLI estimates across all channel pairs within and between the five predefined scalp regions. Connectivity was summarized separately for EO and EC conditions in the theta, alpha, SMR, beta-1, beta-2, and low-gamma bands. Delta-band connectivity was not emphasized because of its greater sensitivity to slow artifacts, vigilance fluctuations, and residual non-stationary activity.

All electrophysiological analyses were descriptive and were not used to infer a treatment-specific neurophysiological mechanism.

3. Results

Results are presented according to the outcome domains described in the Methods. Baseline and post-treatment values, post-treatment minus baseline changes, RCI values, and clinical interpretability criteria are summarized in Table 3. To avoid duplication, the text reports the principal findings without restating all values included in the table. Of the 94 variables included in the multidimensional assessment, an RCI could be calculated for 80 outcomes with comparable numerical baseline and post-treatment observations and suitable psychometric parameters. Fourteen outcomes were not suitable for RCI calculation because an observation was absent, qualitative, categorical, or non-executable, or because an appropriate psychometric parameter was unavailable.

3.1. EEG-Neurofeedback Feasibility and Tolerability

The patient completed all 15 planned EEG-NF sessions, corresponding to an adherence rate of 100%. No session was interrupted or discontinued because of treatment intolerance.

No adverse events were observed or reported during or immediately after training. Specifically, no clinically relevant fatigue, discomfort, headache, dizziness, nausea, agitation, or sustained worsening of neurological or cognitive symptoms was recorded. Feedback thresholds were adjusted by the clinician when required to maintain adequate engagement and reinforcement.

3.2. Cognitive and Affective Outcomes

The MoCA total score increased from 20/30 to 24/30. The change reached the published MCID but did not exceed the 5.1-point MDC95, and the RCI remained below the reliable-change threshold.

The FAB raw score increased from 8 to 12. After adjustment for age and education according to Aiello et al. [15], the adjusted score increased from 8.24 at baseline, corresponding to Equivalent Score 0 and the abnormal range, to 12.24 after treatment, corresponding to Equivalent Score 1 and the borderline range. An RCI was not calculated for the FAB because an appropriate test–retest reliability coefficient was unavailable.

The Clock Drawing Test score increased from 5 to 9 (see Figure 3). Phonemic verbal fluency increased from 12 to 20 and semantic verbal fluency from 19 to 25; neither change exceeded the reliable-change threshold. Digit Span Forward remained unchanged at 4, whereas Digit Span Backward decreased from 3 to 2 without reliable deterioration.

Stroop completion time decreased from 49 to 36 s, and the Stroop error-interference index changed from −3 to −1. Neither outcome met the confirmatory reliable-change threshold.

On the Arrow Stroop Test, total responses increased from 53 to 103 and the number of errors decreased from 10 to 5, corresponding to a reduction in error percentage from 18.9% to 4.9%. Congruent-condition reaction time increased from 0.798 to 1.863 s, whereas incongruent-condition reaction time decreased from 1.118 to 0.935 s.

HADS-Anxiety decreased from 4 to 3, whereas HADS-Depression remained unchanged at 1. Both scores remained within the non-clinical range.

3.3. Neglect-Specific and Ecological Outcomes

On the Bells Cancellation Test, detected targets increased from 5 to 17 and the Center of Cancellation decreased from 0.910 to 0.200. The post-treatment Center of Cancellation remained above the pathological cut-off of 0.081.

On the Letter Cancellation Test, detected targets increased from 28 to 50 and the Center of Cancellation decreased from 0.522 to 0.070, crossing below the pathological cut-off of 0.083. Oxford Cognitive Screen spatial asymmetry decreased from 17 to 11.

On the Timed Neglect Test, left-sided target detection increased from 0 to 4 and the Neglect Index decreased from 1.44 to 0.26. Star-cancellation performance increased from 6 to 36, Diller letter-cancellation omissions decreased from 84 to 35, and article-reading performance increased from 3 to 9. Improvements were also recorded on the Baking Tray Task and during object search in left extrapersonal space.

The Catherine Bergego Scale decreased from 17 to 5, corresponding to a change from the moderate to the mild neglect range.

3.4. Neuromotor and Functional Outcomes

The FIM total score increased from 36 to 58, reaching the published post-stroke MCID and the reliable-change threshold. The FIM cognitive score increased from 19 to 32, exceeding both its MCID and reliable-change threshold. The FIM motor score increased from 17 to 26 but did not reach the published 17-point MCID or the reliable-change threshold.

The Modified Barthel Index increased from 5 to 21. The change exceeded the adopted MCID equivalent and the published MDC95 of 15.4 points and met the reliable-change threshold.

The Timed Up and Go test, which was not executable at baseline, became executable after treatment, with a completion time of 19.72 s.

3.5. Exploratory qEEG Outcomes

All four resting-state EEG recordings were successfully processed and included in the exploratory analyses. The retained signal duration was approximately 380 s for the baseline EO, post-treatment EO, and baseline EC recordings, and approximately 376 s for the post-treatment EC recording.

Absolute band-power changes were calculated as post-treatment minus baseline and differed according to resting-state condition, frequency band, and scalp region.

During the EO condition, absolute delta power increased across all regions, with changes ranging from 3.4% in the occipital region to 46.9% in the central region. Absolute theta power also increased across all regions, ranging from 20.1% in the parietal region to 66.5% in the central region. Alpha-power changes were heterogeneous, ranging from a 10.3% reduction in the frontal region to a 17.3% increase in the occipital region. Absolute SMR power increased across all regions, with changes ranging from 6.6% frontally to 68.5% occipitally. Absolute beta-1, beta-2, and low-gamma power also increased across regions. The largest beta-2 and low-gamma percentage increases were observed in the central region, reaching 264.4% and 472.3%, respectively.

During the EC condition, absolute delta power increased across all regions, with changes ranging from 132.6% in the temporal region to 167.3% in the frontal region. Absolute theta power increased by 29.5% to 75.0%, and absolute alpha power increased by 3.6% to 27.7% across regions. SMR and higher-frequency changes were region-dependent. Temporal SMR, beta-1, and beta-2 power decreased by 19.3%, 34.0%, and 35.2%, respectively, whereas these frequency bands increased in the frontal, central, parietal, and occipital regions. Absolute low-gamma power increased across all regions, ranging from 1.0% temporally to 368.4% centrally.

Frequency-resolved whole-scalp and Cz power spectral density curves for the EO and EC conditions are shown in Figure 4. A summary of regional absolute-power changes is reported in Supplementary Table S2, while complete absolute-power topographic maps for the EO and EC conditions across all predefined frequency bands are provided in Supplementary Figure S1.

The sensitivity analysis excluded 3/190 epochs from the baseline EO recording, 2/190 epochs from the post-treatment EO recording, 1/190 epochs from the baseline EC recording, and 1/188 epochs from the post-treatment EC recording. Median-based spectral estimates calculated after exclusion of high-amplitude epochs reproduced the principal regional patterns observed in the primary analysis.

Sensor-level connectivity changes also differed across frequency bands and recording conditions. During the EO condition, theta within-frontal wPLI decreased from 0.270 to 0.131 (ΔwPLI = −0.139; −51.5%), and theta temporal–occipital wPLI decreased from 0.276 to 0.166 (ΔwPLI = −0.110; −39.9%). Conversely, beta-2 within-frontal wPLI increased from 0.070 to 0.195 (ΔwPLI = +0.125; +179.7%).

Figure 4. Frequency-resolved resting-state EEG power spectral density before and after rehabilitation. (A) Whole-scalp mean spectrum during the eyes-open condition; (B) whole-scalp mean spectrum during the eyes-closed condition; (C) Cz spectrum during the eyes-open condition; and (D) Cz spectrum during the eyes-closed condition. Curves represent baseline and post-treatment power spectral density from 1 to 45 Hz. Insets provide an enlarged view of the 4–20 Hz range to improve visualization of alpha- and SMR-range activity. Inset y-axis limits were scaled independently within each panel and should not be compared across panels. Cz was the active EEG-neurofeedback training site. Spectra are presented descriptively as exploratory electrophysiological correlates of the patient’s clinical evolution.

Figure 4. Frequency-resolved resting-state EEG power spectral density before and after rehabilitation. (A) Whole-scalp mean spectrum during the eyes-open condition; (B) whole-scalp mean spectrum during the eyes-closed condition; (C) Cz spectrum during the eyes-open condition; and (D) Cz spectrum during the eyes-closed condition. Curves represent baseline and post-treatment power spectral density from 1 to 45 Hz. Insets provide an enlarged view of the 4–20 Hz range to improve visualization of alpha- and SMR-range activity. Inset y-axis limits were scaled independently within each panel and should not be compared across panels. Cz was the active EEG-neurofeedback training site. Spectra are presented descriptively as exploratory electrophysiological correlates of the patient’s clinical evolution.

During the EC condition, low-gamma central–temporal wPLI increased from 0.054 to 0.272 (ΔwPLI = +0.219; +406.3%), and beta-2 within-frontal wPLI increased from 0.040 to 0.203 (ΔwPLI = +0.163; +409.2%). Alpha frontal–occipital wPLI decreased from 0.314 to 0.188 (ΔwPLI = −0.126; −40.3%). Selected regional connectivity changes are reported in Supplementary Table S3, while complete regional wPLI heatmaps for the EO and EC conditions across the analyzed frequency bands are provided in Supplementary Figure S2.

4. Discussion

This case report describes the clinical, neuropsychological, functional, and exploratory electrophysiological evolution of a 70-year-old patient with visuospatial neglect following a right thalamo-mesencephalic hemorrhagic stroke who underwent an intensive multidisciplinary rehabilitation program including adjunctive EEG-NF. After treatment, convergent improvements were observed in neglect-specific performance, ecological neglect-related disability, selected executive-attentional measures, and functional independence. Exploratory qEEG analyses also identified condition- and region-dependent changes in absolute spectral power and sensor-level functional connectivity. The patient completed all 15 EEG-NF sessions without adverse events. However, because neurofeedback was administered concurrently with physiotherapy, speech and language therapy, and conventional cognitive stimulation, none of the observed clinical or electrophysiological changes can be attributed specifically to EEG-NF.

The present case is clinically relevant because it illustrates that USN is not exclusively associated with cortical injury. Although neglect most commonly follows right fronto-parietal and temporo-parietal lesions, subcortical structures contribute to spatial attention through distributed cortico-subcortical networks involved in arousal, attentional selection, executive control, and behavioral orienting [1,2,3,4,5]. In the present patient, the right thalamo-mesencephalic hemorrhage may therefore have disrupted thalamo-cortical and attentional network dynamics, contributing to the severe left-sided neglect observed at baseline. Nevertheless, the involvement of specific thalamic nuclei and their relationship with the clinical phenotype cannot be determined from the available structural imaging or from a single case [4].

The clinical outcome pattern indicated selective rather than generalized recovery. Neglect-specific measures showed the most consistent changes, including reduced spatial asymmetry, increased contralesional target detection, improved cancellation performance, and reduced ecological disability on the Catherine Bergego Scale. The Letter Cancellation Center of Cancellation crossed below the published pathological cut-off, whereas the Bells Cancellation Center of Cancellation improved substantially but remained within the pathological range [13]. These findings indicate a clinically meaningful reduction, but not a complete resolution, of spatial neglect.

Cognitive changes were also selective. The MoCA improvement reached the published MCID but remained below the MDC95 [11,12]. FAB performance transitioned from the abnormal to the borderline normative range according to the updated Italian data [15]. Functional gains were particularly evident in FIM total, FIM cognitive, and Modified Barthel Index scores, which met the corresponding clinical or reliable-change criteria [9,10,23]. The more limited change in the FIM motor score suggests that gains in attentional engagement, spatial exploration, and cognitive-functional participation may have contributed to greater autonomy despite persistent motor impairment. Affective scores remained within the non-clinical range at both assessments, leaving little scope for measurable improvement.

The rationale for including EEG-NF was its potential to support self-regulation, attentional readiness, and sensorimotor engagement within the broader rehabilitation program. The protocol reinforced SMR activity while inhibiting theta and high-beta activity at Cz. The choice of Cz should be understood as a protocol-level target for SMR up-training and theta/high-beta inhibition rather than as an attempt to directly modulate the thalamic lesion or other deep subcortical structures. Previous preliminary studies have suggested that neurofeedback, BCI-based interventions, and real-time neuromodulation may be relevant to spatial neglect through modulation of oscillatory activity and attentional-network balance [3,6,7,8]. Nevertheless, differences in electrode placement, frequency targets, lesion characteristics, and rehabilitation context prevent direct comparison between the present protocol and previous neglect-specific interventions.

The absolute-power analyses did not reveal a simple normalization of a single frequency band or a uniform neurofeedback-like effect. During the EO condition, delta, theta, SMR, beta-1, beta-2, and low-gamma power generally increased, whereas alpha changes differed across regions. During the EC condition, delta, theta, and alpha power increased across regions, while SMR and higher-frequency changes were spatially heterogeneous, including reductions over the temporal region. The particularly large percentage changes observed for some beta-2 and low-gamma measures should be considered alongside their absolute values because several corresponding baseline values were small. Overall, the findings are more consistent with a broad and condition-dependent redistribution of resting-state spectral activity than with selective modulation of the trained bands.

The sensitivity analysis indicated that the principal spectral patterns were not driven by a small number of high-amplitude epochs. Nevertheless, resting-state EEG remains influenced by vigilance, fatigue, arousal, and other state-dependent factors that may vary across assessments in post-stroke patients. The spectral findings should therefore be regarded as exploratory correlates of the patient’s evolution rather than as biomarkers of treatment efficacy. This cautious interpretation is consistent with previous observations that unilateral thalamic and striato-thalamic hemorrhages may be accompanied by bilateral EEG abnormalities, reflecting disruption of distributed thalamo-cortical and cortico-thalamic networks [26].

The wPLI findings were similarly heterogeneous. During EO recording, theta connectivity decreased within frontal regions and between temporal and occipital regions, whereas beta-2 within-frontal connectivity increased. During EC recording, alpha frontal–occipital connectivity decreased, while beta-2 within-frontal and low-gamma central–temporal connectivity increased. The divergence between absolute band power and phase-based connectivity indicates that local oscillatory amplitude and large-scale synchronization did not necessarily change in parallel. Given the single-case design, the limited electrode density, and the sensor-level nature of the analyses, these findings cannot be interpreted as evidence of a specific reorganization mechanism or a causal pathway between EEG-NF and clinical recovery.

From an integrated rehabilitation perspective, EEG-NF may have provided a framework for practicing cortical self-regulation and attentional readiness, while cognitive stimulation reinforced visuospatial exploration and compensatory strategies and physiotherapy provided repeated sensorimotor input through posture, balance, weight-bearing, and gait-related practice [1,2,7]. The observed clinical evolution may therefore reflect interactions among complementary top-down and bottom-up rehabilitation components delivered during an intensive post-acute recovery period, rather than the isolated effect of any single intervention.

Limitations

The main limitation is the uncontrolled single-case design. Physiotherapy, speech and language therapy, conventional cognitive stimulation, and EEG-NF were delivered concurrently, preventing separation of their individual contributions. Spontaneous recovery during the post-acute phase, repeated-testing effects, changes in arousal or engagement, and the cumulative intensity of rehabilitation may all have contributed to the observed improvements. This limitation is especially relevant because spontaneous and standard-care-related recovery is common during the early and subacute phases after stroke [27,28]. Additional limitations include the absence of long-term follow-up, the pragmatic rather than standardized assessment of EEG-NF tolerability, and the indirect psychometric basis of several exploratory RCI estimates. Proxy-based RCIs were therefore not considered confirmatory evidence [14]. Moreover, EEG was acquired at only two time points, precluding formal EEG–behavior association analyses and characterization of within-treatment trajectories. The 19-channel montage and sensor-level wPLI analysis did not permit source-level localization, and higher-frequency findings remain especially sensitive to small baseline values and residual physiological noise. Accordingly, the qEEG results should be considered descriptive and hypothesis-generating.

5. Conclusions

This case documents the feasibility of integrating 15 sessions of Cz-based EEG-NF into an intensive multidisciplinary inpatient rehabilitation program for a patient with visuospatial neglect following right thalamo-mesencephalic hemorrhagic stroke. Treatment was completed with full adherence and without adverse events. Clinically meaningful improvements were observed in neglect-related performance, ecological disability, and functional independence, alongside heterogeneous changes in resting-state spectral power and sensor-level connectivity.

Because of the uncontrolled single-case design, concurrent rehabilitation interventions, and absence of follow-up, these findings do not establish a specific therapeutic or neurophysiological effect of EEG-NF. Controlled longitudinal studies using standardized neglect outcomes, predefined EEG measures, appropriate comparison conditions, and longer-term assessments are required to determine whether EEG-NF provides benefits beyond spontaneous recovery and multidisciplinary standard care.