The goal of this clinical trial is to test whether closed-loop fNIRS-BCI neurofeedback(NF) targeting the left dorsolateral prefrontal cortex(DLPFC) can reduce cardiac autonomic arousal, indexed by baseline-corrected heart rate(HR) under cold-induced pain stress, in adults with stable coronary heart disease(CHD) and comorbid anxiety.

The main questions it aims to answer are:

Does real left DLPFC neurofeedback, compared with sham neurofeedback, lead to a greater reduction in baseline-corrected HR during the cold-induced pain stimulation window?

Does real neurofeedback produce stronger volitional upregulation of left DLPFC activation and higher inter-hemispheric synchronisation between left and right DLPFC than sham neurofeedback?

Are changes in baseline-corrected HR statistically associated with, and partly mediated by, changes in left DLPFC activation or DLPFC inter-hemispheric synchronisation?

What adverse events(AEs) occur during adaptive training and the formal experimental session, and do AE rates differ between the two groups?

Researchers will compare a real neurofeedback group with a sham neurofeedback group to determine whether targeting the left DLPFC via closed-loop fNIRS-BCI yields superior modulation of cardiac autonomic responses and prefrontal activation patterns in CHD patients with anxiety.

Participants will:

undergo cardiac and psychiatric screening to confirm stable CHD, DSM-5 anxiety disorder, and other eligibility criteria;

attend three adaptive training sessions(days 1-3) with fNIRS-BCI neurofeedback targeting the left DLPFC, combined with slow-wave auditory stimulation and mild cold-water exposure, while ECG is recorded;

on day 4, complete one formal experimental session consisting of 15 blocks of cold-induced pain stimulation and slow-wave auditory stimulation, with simultaneous fNIRS and ECG recording, receiving either real or sham left DLPFC neurofeedback according to randomisation, and continuous monitoring for adverse events.

This exploratory clinical trial is grounded in the emerging field of psycho-cardiology, which highlights the close interaction between mental health and cardiovascular outcomes in patients with coronary heart disease(CHD). Converging epidemiological data indicate that anxiety is common in CHD and is associated with higher rates of in-hospital complications, rehospitalisation, major adverse cardiovascular events, and mortality. Experimental and clinical studies further suggest that heightened sympathetic arousal and disturbed cardiac autonomic regulation may form a physiological link between anxiety and adverse cardiac prognosis in this population.

At the neural level, the central autonomic network(CAN)-including the insula, cingulate cortex, prefrontal cortex, amygdala, hypothalamus, and brainstem nuclei-provides a structural and functional substrate for brain-heart coupling. Within this network, the dorsolateral prefrontal cortex(DLPFC) plays a key role in executive control, conflict monitoring, and cognitive reappraisal, and interacts with interoceptive and autonomic control circuits. Prior work supports functional lateralisation: the right DLPFC is more strongly associated with threat monitoring and sympathetic mobilisation, whereas the left DLPFC is more involved in top-down cognitive control and reappraisal. Under stress, robust recruitment of left DLPFC activity is thought to dampen excessive interoceptive drive and hyperarousal, potentially facilitating engagement of parasympathetic pathways and supporting emotional regulation.

On this mechanistic background, the present trial focuses on patients with stable CHD and comorbid anxiety disorders, a group characterised by more pronounced autonomic imbalance and substantial pharmacological and comorbidity-related confounding. The central hypothesis is that closed-loop neurofeedback(NF) targeting the left DLPFC can, under experimentally induced stress, strengthen prefrontal control and thereby attenuate cardiac autonomic arousal, indexed by heart rate(HR), relative to a sham control condition. The study further postulates that inter-hemispheric synchronisation between left and right DLPFC will increase under real NF, reflecting a more coordinated prefrontal response pattern, and that changes in HR will be associated with, and partially mediated by, changes in left DLPFC activation and DLPFC inter-hemispheric coupling.

Study design and population The study is designed as a prospective, randomised, sham-controlled, parallel-group exploratory clinical trial. Adults with stable CHD and comorbid anxiety disorders are screened using predefined cardiac and psychiatric criteria. Key features include confirmed CHD(stress testing, prior myocardial infarction, or angiographic stenosis), DSM-5 anxiety disorder, right-handedness, and resting HR within a prespecified range; individuals with major arrhythmias, unstable angina, advanced heart failure, severe valvular disease, high-risk blood pressure profiles, or other unstable medical or psychiatric conditions are excluded. All participants must be free of psychotropic medications for at least one month prior to enrolment.

After providing informed consent and completing baseline cardiovascular and psychiatric assessments, eligible participants are randomly assigned in a 1:1 ratio to a real-neurofeedback group or a sham-neurofeedback group. Randomisation is stratified and concealed according to local procedures. Participants and outcome assessors are blinded to group allocation, whereas the operator running the neurofeedback software is aware of assignment but does not participate in outcome evaluation.

Neurofeedback intervention and experimental paradigm The intervention uses a functional near-infrared spectroscopy-based brain-computer interface(fNIRS-BCI) to deliver closed-loop NF targeting the left DLPFC, combined with simultaneous electrocardiography(ECG). The full protocol comprises an adaptive training phase(days 1-3) followed by a single formal experimental session(day 4).

During adaptive training, participants receive standardised instructions on self-regulation strategies aimed at enhancing left DLPFC activation. They are encouraged to experiment with approaches such as attending to a 1 Hz slow-wave auditory rhythm, emotional calming, mood stabilisation, attentional reorientation, and positive imagery. Each training session consists of repeated 20 s rest-40 s stimulation blocks, with mild cold-water exposure(2-4°C) during the stimulation periods. Visual feedback is presented as an "energy bar" whose height and colour map, in sign-inverted form, to a real-time statistical measure of left DLPFC haemodynamic activation; lower bar height corresponds to stronger activation. The adaptive phase familiarises participants with the interface and tasks but is not included in the efficacy analyses.

On day 4, participants complete a single formal session consisting of 15 blocks, each 60 s in duration(20 s rest + 40 s stimulation). Throughout the stimulation periods, two cues are presented concurrently:

a 1 Hz sinusoidally amplitude-modulated auditory tone at approximately 60 dB serving as a rhythmic attentional anchor and NF training cue; and a cold-pain stressor, delivered by having participants press the palmar surface of both hands against the outer wall of a 0.5 L ice-water bottle(0-2°C) during the first 20 s of each stimulation period, without gripping or withdrawing their hands.

This block structure corresponds to a 0.0167 Hz(1/60 Hz) cycle, which falls within the very-low-frequency range relevant for slow autonomic and vascular responses. The cold-pain component creates a high-stress context, in which the ability of participants to engage left DLPFC-mediated top-down control and modulate cardiac responses can be tested.

Real versus sham neurofeedback In both groups, fNIRS probes are placed symmetrically over the bilateral prefrontal cortex, with an optode montage covering the left DLPFC as the target region. Three-lead ECG is recorded simultaneously. The NeuroMind-NIRS software performs online preprocessing of left DLPFC oxygenated haemoglobin(HbO) signals and applies a sliding-window general linear model(GLM) to estimate a task-related test statistic t in real time. For feedback display, a sign-inverted variable s = -t is used so that a lower on-screen bar indicates higher underlying DLPFC activation.

The real-neurofeedback group uses a stringent threshold in the display domain (T = -3.3, corresponding to an internal t ≥ +3.3, approximately p≈0.001). Under this setting, visual feedback is strongly and directionally coupled to true left DLPFC activation, allowing participants' volitional regulation efforts to produce a clear and interpretable effect on the feedback display, and thereby establishing a meaningful closed loop between cortical activity and perceived performance.

The sham-neurofeedback group uses a near-baseline threshold(T = -0.1, corresponding to t ≈ +0.1, approximately p≈0.99), which markedly weakens the effective coupling between feedback and instantaneous neural state while maintaining an interface visually indistinguishable from the real NF condition. In this group, the feedback behaves in a way that is largely insensitive to genuine changes in left DLPFC activation, making it difficult for participants to achieve true neurophysiological control despite similar task demands and exposure.

Across both groups, the auditory stimulus, cold-pain protocol, random noise embedding within the acoustic stream, and lack of in-session verbal guidance on day 4 are identical. Participants are free to withdraw at any time, and they are advised to avoid caffeine, alcohol, and smoking and to maintain adequate sleep before the formal session.

Physiological recordings and offline analysis fNIRS and ECG signals are recorded continuously during the experimental session. fNIRS data are preprocessed offline in MATLAB using the Homer2 toolbox. Standard steps include conversion of raw intensity to optical density, channel-wise detection and correction of motion artefacts(using predefined thresholds and cubic spline interpolation), and band-pass filtering(0.01-0.08 Hz) to remove slow drift and high-frequency physiological noise. HbO concentration changes are then derived using the modified Beer-Lambert law, and signals are baseline-corrected using the pre-stimulus interval. Subsequent analyses focus on HbO because of its higher signal-to-noise ratio compared with deoxyhaemoglobin.

For left DLPFC activation, preprocessed HbO time series from channels over the left DLPFC are standardised and entered into a GLM with a task regressor(boxcar convolved with a canonical haemodynamic response function and its derivatives), a physiological covariate derived from downsampled ECG, and an intercept term. For each participant, task-related β coefficients from the left DLPFC channels are summarised and carried forward into group-level linear mixed-effects models.

To characterise inter-hemispheric synchronisation(IHS), the study uses wavelet transform coherence(WTC) to estimate time-frequency coupling between homologous left-right DLPFC HbO channels. For each participant, WTC values are computed, averaged within a prespecified low-frequency band(0.01-0.08 Hz) and relevant task time windows, and transformed using the Fisher z transformation. These z-scores serve as quantitative indices of DLPFC inter-hemispheric coupling for subsequent group comparisons and mediation analyses.

HR is derived from ECG and baseline-corrected relative to the pre-stimulus resting period. The primary endpoint is the between-group difference in baseline-corrected HR during the stimulation window on day 4(real vs sham NF).

Outcomes and analytic framework The primary endpoint is the difference between real and sham neurofeedback groups in baseline-corrected HR during the cold-pain task window on the formal experimental day.

Key secondary endpoints are:

the between-group difference in left DLPFC activation, indexed by task-related HbO β coefficients; and the between-group difference in DLPFC inter-hemispheric synchronisation, indexed by Fisher z-transformed WTC measures.

Exploratory mediation analyses are planned to test whether the effect of group assignment on baseline-corrected HR is indirectly associated with changes in left DLPFC activation and with changes in DLPFC inter-hemispheric synchronisation. In these models, group is treated as the exposure, cortical metrics as mediators, and baseline-corrected HR as the outcome; covariates mirror those included in the primary analysis. Non-parametric bootstrap methods are used to estimate indirect effects and their confidence intervals.

The primary and secondary endpoints are analysed using linear mixed-effects models with group as a fixed effect, pre-specified baseline-imbalanced covariates entered as additional fixed effects, and participant identifier included as a random intercept to account for repeated measures where relevant. Parameters are estimated by restricted maximum likelihood, and statistical significance is evaluated using F tests with Satterthwaite approximations to the degrees of freedom. Effect estimates are reported with 95% confidence intervals and p values.

Safety endpoints consist of the incidence and profile of adverse events(AEs) from randomisation through completion of the experimental session, with particular attention to symptoms related to bradycardia or hypotension, pronounced sympathetic activation, cold-pain intolerance, skin irritation at sensor sites, and sound-related discomfort. AEs and serious AEs are documented and managed according to predefined procedures, with appropriate reporting to the principal investigator and ethics oversight bodies.

Safety analyses are conducted in the intention-to-treat population(all randomised participants). Analyses of the primary, secondary, and exploratory efficacy endpoints are restricted to participants who successfully complete all 15 stimulation blocks on the formal experimental day; data from sessions terminated early due to adverse or unexpected events and judged non-restartable by the clinician are excluded from efficacy analyses but retained for safety analysis.

Sample size and feasibility Sample size planning is informed by an earlier pilot study using a similar paradigm targeting the right DLPFC, which suggested a large effect size for between-group differences in baseline-adjusted HR. Based on those data and assuming a two-sided α of 0.05 and 95% power, approximately 48 participants(24 per group) would be required to detect the anticipated effect. To allow for attrition and incomplete sessions, the present trial plans to enrol a larger number of participants, with the final sample size reflecting both statistical considerations and operational feasibility.

Overall, this trial is designed to provide preliminary clinical-grade evidence on whether closed-loop fNIRS-BCI neurofeedback targeting the left DLPFC can modulate cardiac autonomic responses and prefrontal activation patterns in CHD patients with anxiety, and to explore the mechanistic role of cortical activation and inter-hemispheric synchronisation within a brain-heart regulatory framework.

Inclusion Criteria:

• Age >18 years, any sex.

• Right-handed, with resting heart rate between 60 and 100 beats per minute.

• Confirmed diagnosis of CHD, defined as at least one of the following:

  (i) positive stress test; (ii) documented myocardial infarction (MI) with electrocardiographic changes and concurrent elevation of creatine kinase MB isoenzyme or troponin; (iii) angiographically confirmed coronary atherosclerosis with ≥50% stenosis in at least one coronary artery.

• Diagnosis of an anxiety disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5).

• Hamilton Anxiety Rating Scale (HAMA) score ≥16 and 17-item Hamilton Depression Rating Scale (HAMD-17) score ≤17.

Exclusion Criteria:

• Acute unstable angina.
• Severe congestive heart failure (New York Heart Association [NYHA] class IV).
• Valvular heart disease.
• History of atrial fibrillation.
• Unstable blood pressure, defined as systolic blood pressure >180 mmHg or <90 mmHg.
• Pregnancy.
• History of unstable medical conditions, including cerebrovascular disease, dementia, hyperthyroidism, pulmonary disease, or malignancy. These are assessed through medical history, electronic health records, physical examination, and ECG findings.
• High risk of suicide or homicide.
• Presence of other psychiatric disorders, including psychotic disorders, bipolar disorder, or active substance use disorders.
• Use of psychotropic medication within 1 month prior to enrolment, to avoid potential interference with haemodynamic measurements.

Available at the time of publication

Participants and all clinical assessors (cardiology and psychiatry staff conducting eligibility assessment and rating scales) are masked to group allocation. Outcome assessors and statisticians performing the primary and secondary analyses are also blinded and receive only coded group labels. The neurofeedback operator is aware of allocation solely to configure the real vs sham feedback thresholds and has no role in clinical evaluation, endpoint assessment, or data analysis. The user interface and experimental procedures are identical across groups, minimising the risk of unblinding.