Naked eye 3D-based automatic control system for classified exposure treatment of phobia

Abstract

The invention relates to the cross technical field of biomedical engineering and psychotherapy, in particular to an automatic control system for classified exposure therapy of phobia based on naked eye 3D. The system comprises an electroencephalogram signal feature extraction module, a heart rate variability feature extraction module, a skin electric response feature extraction module, an eye movement feature extraction module, a physiological load state feature extraction module, an exposure therapy process feature extraction module, a first feature fusion module, a second feature fusion module and a stimulation parameter adaptive control module. The method combines multi-mode biological signal real-time analysis and naked eye 3D stimulation parameter self-adaptive adjustment, and is suitable for clinical psychological treatment mechanisms and psychological health intervention scenes.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of biomedical engineering and psychotherapy, specifically to an automatic control system for graded exposure therapy for phobias based on naked-eye 3D. Background Technology

[0002] Exposure therapy, as a first-line treatment for anxiety disorders (such as social anxiety and specific phobias), works by gradually exposing patients to fear stimuli, allowing them to adapt and build anxiety tolerance. However, existing technologies have the following significant drawbacks in practical applications: The assessment is highly subjective: Current clinical assessments largely rely on subjective rating tools such as the Hamilton Anxiety Rating Scale. These scales are usually based on patient self-reports or physician subjective observations, which have a time lag and cannot reflect the patient's current physiological stress state in real time. For example, when faced with high-intensity stimuli, patients may have distorted scores due to anxiety, ultimately resulting in a mismatch between the intensity of the stimulus and the actual level of anxiety, affecting the treatment effect.

[0003] Fixed stimulation parameters: In naked-eye 3D exposure therapy, key parameters of the stimulation content (such as parallax, frame rate, etc.) are often adjusted using uniform rules. This "one-size-fits-all" approach ignores the significant differences in physiological tolerance between individuals and can easily lead to different negative effects: patients with low tolerance may experience severe dizziness, nausea and other discomfort; while for patients with high tolerance, the stimulation intensity may be insufficient and the expected therapeutic effect may not be achieved.

[0004] Deficiencies in multimodal signal processing: Traditional EEG signal analysis typically uses full-lead data, resulting in significant feature redundancy, increasing computational complexity and impacting analysis efficiency; Heart rate variability analysis is often based on a static baseline set before treatment, but during actual treatment, factors such as changes in ambient temperature and micro-movements of the patient's limbs can cause baseline drift. Continuing to use a static baseline may lead to misjudgment of the patient's true physiological state, thus affecting the rational adjustment of treatment parameters; Existing skin conductance response analysis relies too heavily on transient responses while neglecting the important indicator of baseline excitability. This one-sided analysis method can easily lead to missed or misjudged responses even when the patient exhibits anxiety under low baseline excitability conditions due to insignificant transient response signals.

[0005] In summary, existing technologies have significant shortcomings in achieving a closed loop of real-time physiological monitoring, precise stimulation adjustment, and personalized treatment. There is an urgent need to develop an auxiliary treatment system that can take into account objectivity, scenario adaptability, and individual differences to improve the clinical efficacy and safety of exposure therapy. Summary of the Invention

[0006] To address the shortcomings of existing methods and the needs of practical applications, and in order to solve the aforementioned problems, this invention provides an automatic control system for graded exposure therapy for phobias based on naked-eye 3D, characterized by comprising: The system includes an EEG signal feature extraction module, a heart rate variability feature extraction module, a skin conductance response feature extraction module, an eye movement feature extraction module, a physiological load state feature extraction module, an exposure therapy process feature extraction module, a first feature fusion module, a second feature fusion module, and a stimulation parameter adaptive control module. The first feature fusion module fuses features extracted from the EEG signal feature extraction module, the heart rate variability feature extraction module, and the exposure therapy process feature extraction module to obtain a first feature. The second feature fusion module fuses features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module to obtain a second feature. The stimulation parameter adaptive control module integrates the features extracted from the physiological load state feature extraction module with the first and second features to achieve automatic control of phobia graded exposure therapy.

[0007] This invention provides a precise and automated solution for graded exposure therapy for phobias through modular design and graded feature fusion. Based on multimodal physiological signals such as EEG, heart rate variability, skin conductance, and eye movement, it combines two levels of feature fusion (first fusion of central physiological and process features, and second fusion of peripheral emotional and behavioral features) to extract key information layer by layer, improving the accuracy of patient anxiety and physiological state assessment. By linking physiological load features with the two-level fusion features and using an adaptive stimulus parameter control module, it achieves dynamic and personalized adjustment of the intensity of therapeutic stimuli, avoiding discomfort or insufficient efficacy caused by "one-size-fits-all" parameter settings, and meeting the core requirement of "gradual tolerance" in graded exposure. The fully automated control reduces human intervention, improves treatment efficiency and stability, provides intelligent technical support for clinical treatment of phobias, and promotes the transformation of exposure therapy from experience-based to data-driven precision medicine.

[0008] Optionally, features are extracted through the EEG signal feature extraction module, including the following steps: The raw EEG signal is denoised; based on the processing results, short-time Fourier transform analysis is performed, and then the energy integral method is used to calculate the relative energy and quantify the energy proportion characteristics of different frequency bands.

[0009] This invention is based on the patient's prefrontal cortex, limbic system, and other brain regions. Wave, By monitoring abnormal fluctuations in frequency bands such as wavelet, the neural oscillation patterns related to the generation, maintenance, and fluctuation of anxiety can be accurately located. The abnormal neural circuit coupling state between the amygdala and the prefrontal cortex can be deeply analyzed, thereby obtaining an EEG feature map reflecting the mechanism of anxiety generation. This provides a precise central nervous system theoretical basis and quantitative indicators for the dynamic adjustment of personalized electrical stimulation parameters based on the principle of neural plasticity.

[0010] Optionally, features are extracted through the heart rate variability feature extraction module, including the following steps: The heart rate intervals were purified; based on the purified heart rate interval sequence, time-domain and frequency-domain indicators were analyzed to obtain a heart rate variability feature vector. This invention quantifies patient anxiety levels from the perspective of autonomic nervous function using a low-frequency / high-frequency power ratio indicator, combined with electroencephalogram (EEG) characteristics, to achieve a multi-dimensional and objective assessment of anxiety states.

[0011] Optionally, features are extracted using the skin conductance feature extraction module, including the following steps: The baseline of the skin conductance response (SCR) is dynamically calibrated using a sliding window mechanism. Based on the SCR signal after baseline calibration, anxiety-related transient SCR events are detected using the slope threshold method, and a three-dimensional emotional arousal feature vector is constructed. By monitoring the immediate anxiety response of patients after exposure to naked-eye 3D stimulation, and using the sudden increase in SCR as a quantitative indicator, objective evidence based on a rapid emotional feedback mechanism is provided for the real-time dynamic adjustment of stimulation parameters.

[0012] Optionally, features are extracted through the eye-tracking feature extraction module, including the following steps: The field of view was filtered to obtain effective fixation sequences. The dispersion of fixation points in the effective fixation sequences was calculated using the standard deviation ellipse method. Eye movement feature vectors were constructed using principal component analysis, combining fixation dispersion, average pupil diameter, and average blink frequency. Patients' behavioral response patterns to naked-eye 3D stimuli were quantitatively analyzed from the perspective of visual attention, including typical anxiety manifestations such as increased fixation dispersion and pupil diameter dilation. A multimodal data cross-validation system was constructed to achieve complementary validation of physiological signals and behavioral data.

[0013] Optionally, features are extracted through the physiological load state feature extraction module, including the following steps: Based on historical physiological tolerance data, safe threshold ranges for key parameters of naked-eye 3D stimulation were constructed. Based on these safe threshold ranges, physiological load values ​​were calculated using a physiological load value model. Combining these physiological load values ​​with disparity magnitude and stimulation frame rate, a stimulation load feature vector was obtained. This quantitative analysis of the load exerted on patients by naked-eye 3D visual stimulation using physiological indicators provides empirical evidence based on the stimulation source for subsequent parameter optimization decisions.

[0014] Optionally, features are extracted through the exposure therapy process feature extraction module, including the following steps: One-hot encoding technology is used to encode the stage features of each treatment phase; an adaptation quantification model is constructed and used to evaluate stage adaptation; the one-hot encoding of each stage, treatment duration, and adaptation score are fused to obtain a treatment process adaptation feature vector. By accurately anchoring the stage coordinates of the treatment process, the personalized adaptation dimensions of patients are deeply analyzed, thereby constructing a dynamic weighting system that integrates multiple features. In the high-intensity treatment phase, the stimulation load is used as the core observation variable to achieve precise and contextualized weight allocation.

[0015] Optionally, by fusing features extracted from the EEG signal feature extraction module, the heart rate variability feature extraction module, and the exposure therapy process feature extraction module through the first feature fusion module to obtain a first feature, the following steps are included: Based on stage information, the weights of EEG signal features and heart rate variability features are finely allocated. Based on the weighted feature results, the fit score is used as a dynamic threshold for feature selection to establish a correlation mechanism between data quality and analytical accuracy, thus obtaining the first feature. This invention integrates EEG signal features and heart rate variability features to construct a physiological-process feature fusion model dynamically coupled with the treatment process, providing a quantitative basis for precise intervention decisions.

[0016] Optionally, the second feature is obtained by fusing features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module through the second feature fusion module, including the following steps: Based on the Pearson correlation coefficient algorithm, the feature correlation between features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module is calculated. A sliding window midpoint filtering algorithm is used to smooth the amplitude and dispersion of the skin conductance response. Combining the feature correlation and smoothing results, a second feature is obtained. This invention constructs a patient state consistency verification model based on the emotional arousal level represented by skin conductance response and visual attention data obtained from eye movement tracking by calculating the correlation coefficient between the two. A high correlation coefficient can serve as a quantitative basis for the authenticity of the patient's current state, while a low correlation coefficient suggests possible external interference or faking behavior, thus providing an objective indicator for assessing the authenticity of the patient's state in clinical decision-making.

[0017] Optionally, the stimulation parameter adaptive control module integrates the features extracted by the physiological load state feature extraction module with the first feature and the second feature to complete the automatic control of phobia graded exposure treatment, including the following steps: determining the suitability level of the current stimulus through fuzzy calculation; and adjusting the naked-eye 3D stimulation parameters according to the determination result. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an automatic control system for graded exposure therapy of phobia based on naked-eye 3D, provided in an embodiment of the present invention. Detailed Implementation

[0019] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0020] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0021] Please see Figure 1 To address the aforementioned problems, this invention provides an automatic control system for graded exposure therapy for phobias based on naked-eye 3D, such as... Figure 1 As shown, in one embodiment, the system includes: The system includes modules for extracting features from electroencephalogram (EEG) signals, heart rate variability, skin conductance response, eye movement, physiological stress, exposure therapy process, first feature fusion, second feature fusion, and adaptive control of stimulation parameters. The first feature fusion module is used to fuse features extracted from the EEG signal feature extraction module, the heart rate variability feature extraction module, and the exposure therapy process feature extraction module to obtain a first feature; The second feature fusion module is used to fuse the features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module to obtain a second feature; The stimulus parameter adaptive control module is used to integrate the features extracted by the physiological load state feature extraction module with the first feature and the second feature to complete the automatic control of phobia graded exposure therapy.

[0022] In this embodiment, the extraction of features through the EEG signal feature extraction module includes the following steps: S11. Noise reduction processing is performed on the raw EEG signal.

[0023] Real-time acquisition of raw 16-lead electroencephalogram (EEG) signals, sampled at 250Hz, including... Wave (8-13Hz, relaxation related) Wave (14-30Hz, anxiety-related), Wave (31-50Hz, cognitive processing related) frequency band information.

[0024] Based on the international 10-20 system electrode positioning standard, the corresponding lead data of the prefrontal F3 / F4 (core area of ​​executive function) and temporal T3 / T4 (key projection sites of the amygdala) are accurately located. To address the subtle somatic motion interference caused by patient emotional fluctuations during exposure therapy, an adaptive Kalman filter algorithm is employed. By constructing a dynamic state-space model, the filter gain parameters are adjusted in real time to improve the suppression of motion artifacts.

[0025] Simultaneously, by combining a band-stop filter bank to eliminate 50Hz power frequency interference and notch filtering technology to remove electromyographic noise, a high-quality noise-reduced EEG signal is finally obtained.

[0026] S12. Based on the processing results, short-time Fourier transform analysis is performed, and then the energy integration method is used to calculate the relative energy and quantify the energy proportion characteristics of different frequency bands.

[0027] Short-time Fourier transform (STFT) analysis was performed on the high-quality denoised EEG signal. A parameter configuration of 2-second sliding window and 50% overlap was selected to achieve high-resolution signal characterization in the time-frequency two-dimensional space.

[0028] against (8-13Hz) (14-30Hz) In the 30-100Hz frequency band, the relative energy is calculated using the energy integration method. Energy ratio as a core anxiety assessment parameter ,satisfy: .

[0029] Meanwhile, previously frontal lobe F3 / F4 leads Wave absolute energy value serves as a key biomarker for quantifying cognitive regulation function. Ultimately, the two types of features mentioned above are integrated into a 2×1 dimensional feature vector, namely the EEG anxiety-cognitive feature vector, providing a standardized data foundation for subsequent real-time monitoring of anxiety states and objective evaluation of intervention effects.

[0030] In this embodiment, the heart rate variability feature extraction module extracts features, including the following steps: S21, Purify heart rate interval.

[0031] The pulse signal (sampling rate 100Hz) acquired by the photoplethysmography (PPG) sensor is converted into raw heart rate interval (RR interval) data through peak detection.

[0032] During exposure therapy, even minor patient movements (such as involuntary limb tremors or changes in respiratory rhythm) can easily trigger abnormal fluctuations in RR interval data. To accurately filter out such noise, a sliding window interpolation method is used. Outlier removal algorithm. In its implementation, a sliding window length of 5 seconds is set, and then the difference between adjacent RR intervals within the window is calculated based on statistical... The principle is to identify and remove values ​​that exceed the mean ± 3 standard deviations from the mean, ultimately extracting smooth and reliable cleaned RR sequences from the original RR sequences. .

[0033] S22. Based on the purified heart rate interval sequence, analyze the time domain index and frequency domain index to obtain the heart rate variability feature vector.

[0034] By analyzing the purified RR sequence Through in-depth analysis, two key time-domain parameters were extracted. One is the standard deviation (SDNN), which is calculated... The standard deviation of all adjacent RR interval differences can comprehensively reflect the overall activity level of the autonomic nervous system, and its numerical changes can intuitively reflect the body's autonomic nervous system regulation ability under different physiological and psychological states. The second is the root mean square (RMSSD) of the differences between adjacent RR intervals. This parameter focuses on characterizing the short-term fluctuations of the RR intervals. Because it is extremely sensitive to the regulatory effect of the parasympathetic nervous system, it is often used to assess the immediate activity of the parasympathetic nervous system.

[0035] To further analyze the frequency characteristics of heart rate variability, Power spectral density (PSD) analysis was performed on the sequences. An autoregressive model was used to divide the power spectrum into two key frequency bands: the low-frequency band (LF, 0.04-0.15 Hz), primarily controlled by the sympathetic nervous system, reflecting neural regulation under stress and alertness; and the high-frequency band (HF, 0.15-0.4 Hz), closely related to the parasympathetic nervous system, reflecting physiological characteristics under relaxation and recovery. The power ratio (LF / HF) of the low-frequency band was calculated to quantify the balance between the sympathetic and parasympathetic nervous systems. Clinical studies have shown that this ratio significantly increases during periods of anxiety.

[0036] Furthermore, to construct a high-dimensional and comprehensive physiological feature vector, the three core indicators SDNN, RMSSD, and LF / HF are deeply integrated to generate a heart rate variability feature vector with a dimension of 3×1. This vector systematically integrates key information from the time and frequency domains of heart rate variability, which can accurately characterize the overall functional state of the autonomic nervous system and provide an important data foundation for subsequent physiological signal analysis and psychological state assessment.

[0037] The skin conductance feature extraction module extracts features, including the following steps: S31. The baseline of skin conductance response is dynamically calibrated through a sliding window mechanism.

[0038] Raw fingertip GSR data (sampling rate 50Hz) collected by a skin conductance sensor includes skin conductance level (SCL, baseline arousal) and skin conductance response (SCR, transient emotional fluctuation).

[0039] To accommodate the 10-30 minute treatment duration requirements of exposure therapy, a sliding window mechanism was employed, with a calibration window cycle of 5 minutes. Within each window, the mean value of Galvanic Skin Response (GSR) was calculated. with standard deviation , construct a dynamic baseline model.

[0040] When real-time monitoring The value exceeds When the data falls within the specified range, a baseline drift phenomenon is detected. Then, a linear interpolation algorithm is automatically activated to correct the drifted data to the current window baseline level, thereby ensuring the accuracy and stability of subsequent data analysis.

[0041] S32. Based on the skin conductance response signal after baseline calibration, detect anxiety-related transient skin conductance response events according to the slope threshold method, and construct a three-dimensional emotional arousal feature vector.

[0042] For GSR signals after baseline calibration, the slope threshold method was used to detect anxiety-related transient skin conductance response (SCR) events. The specific criteria were: an SCR event was defined as a signal slope exceeding 0.5 and a duration greater than 0.5 seconds. During the detection process, two key parameters of the SCR event were automatically recorded: peak amplitude to reflect the intensity of the anxiety response, and peak frequency to reflect the frequency of anxiety occurrences, thus providing quantitative support for subsequent emotion analysis.

[0043] By fusing the calibrated stable skin conductance level, the peak amplitude and frequency of SCR events, a three-dimensional emotional arousal feature vector is constructed. This vector can comprehensively characterize an individual's emotional arousal state within a specific time period, providing a standardized data foundation for emotion recognition and psychological state assessment based on GSR signals.

[0044] In another embodiment, the feature extraction via the eye-tracking feature extraction module includes the following steps: S41. Filter the field of view to obtain an effective fixation sequence.

[0045] Raw eye-tracking data (sampling rate 60Hz) collected by an eye tracker includes fixation point coordinates (x,y), pupil diameter (PD), and blink frequency (BF).

[0046] Based on visual cognition theory, the effective perception area of ​​the human eye on a naked-eye 3D display screen is concentrated in the center of the screen. Taking a 27-inch, 1920×1080 resolution display screen as an example, the effective field of view for stimuli is defined by calculating and delineating 80% of the area in the center of the screen (coordinate range [192,1728]×[108,972]). The selection of this area has been verified by multiple pre-experiments to ensure that after removing interference data outside the field of view, more than 95% of the effective gaze behavior can be retained, and finally the effective gaze sequence is obtained.

[0047] S42. The dispersion of fixation points in the effective fixation sequence is calculated using the standard deviation ellipse method.

[0048] The dispersion of fixation points in an effective fixation sequence is calculated using the standard deviation ellipse method. This involves calculating the average Euclidean distance from all fixation points to the geometric center and combining this with a time weighting coefficient (higher weight for closer time points) to quantify fixation stability.

[0049] S43. Combine fixation dispersion, average pupil diameter, and average blink frequency to construct an eye movement feature vector using principal component analysis.

[0050] Using an iris recognition algorithm, the system captures the dynamic changes of the pupil in the effective field of view in real time and calculates the average pupil diameter; at the same time, it detects the eyelid opening and closing state and calculates the blinking frequency by using the inter-frame difference method.

[0051] Three features—fixation dispersion, average pupil diameter, and average blink frequency—were normalized and mapped to the [0,1] interval. Principal component analysis was then used to effectively eliminate redundant information between features, ultimately constructing a 3×1 eye-movement feature vector to comprehensively and accurately represent eye-movement visual attention states. This eye-movement feature vector can be used as standardized input to directly integrate with machine learning models, enabling real-time monitoring and quantitative assessment of anxiety.

[0052] In this embodiment, the feature extraction through the physiological load state feature extraction module includes the following steps: S51. Based on historical physiological tolerance data, construct a safe threshold range for key parameters of naked-eye 3D stimulation.

[0053] Real-time parameters of naked-eye 3D stimulation include parallax magnitude (-5°~+5°, negative values ​​are cross-parallax, positive values ​​are non-cross-parallax), stimulation frame rate (24~60fps), and stimulation content complexity (1~5 levels, based on the number of targets and motion speed).

[0054] By systematically mining and deeply statistically analyzing the historical physiological tolerance data of multiple patients undergoing exposure therapy, safe threshold ranges for each key parameter were constructed.

[0055] Furthermore, the Min-Max normalization method is used to normalize parameters such as disparity, frame rate, and complexity, resulting in the following values: The standardized values ​​are distributed in the range of 0 to 1. The closer the value is to 1, the closer the current parameter is to the patient's physiological tolerance limit, and the higher the intensity of the corresponding treatment stimulus.

[0056] S52. Based on the safety threshold range, calculate the physiological load value using the physiological load value model, and combine the physiological load value with the disparity size and the stimulus frame rate to obtain the stimulus load feature vector.

[0057] Based on the random forest regression algorithm, the actual impact of each parameter on physiological load is learned from massive historical data, and then the weight coefficients of the weighted summation model are determined.

[0058] In this embodiment, the parallax parameter, because it directly affects stereoscopic vision perception and has a significant impact on physiological load, is assigned a weight of 0.4; the frame rate parameter, because it affects the frequency of visual information input, is assigned a weight of 0.3; the complexity parameter, which comprehensively reflects the complexity of screen elements and the difficulty of processing, is also assigned a weight of 0.3, thereby accurately calculating the physiological load value of the current stimulus. This indicator can intuitively characterize the patient's physiological stress level during treatment.

[0059] Furthermore, physiological load values ​​were calculated using a physiological load value model. ,satisfy: After standardization , Parameters, and the calculated physiological load values By integrating these elements, a 3×1 stimulus load feature vector is constructed, achieving a high degree of integration of core information such as parallax, frame rate, and physiological load. This provides a standardized data foundation for subsequent treatment plan optimization, patient status assessment, and risk warning, helping medical and research personnel to quickly obtain key information and make scientific decisions.

[0060] In this embodiment, features are extracted through the exposure therapy process feature extraction module, including the following steps: S61. Use unique thermal coding technology to encode the stage features of the treatment stage.

[0061] Real-time progress data of exposure therapy, including the current treatment stage (Stage: 1 = relaxation, 2 = low-intensity stimulation, 3 = high-intensity stimulation) and the duration of the current stage.

[0062] One-hot encoding is used to encode the stage features of the treatment. Specifically, if the treatment is divided into 3 stages, Stage 2 is encoded as [0, 1, 0]. This encoding method intuitively reflects the current stage. Different stages correspond to different treatment target intensities. For example, Stage 3 requires maintaining a high stimulation intensity, but it must be strictly controlled within a safe threshold to avoid adverse reactions due to overstimulation. This encoding method allows subsequent models to more accurately identify and handle the treatment needs of different stages.

[0063] S62. Construct an adaptability quantification model and use the adaptability quantification model to evaluate the adaptability of the stage.

[0064] First, historical physiological data during the treatment phase were statistically processed, using electroencephalogram (EEG) and heart rate variability (HRV) as the observation objects, and the EEG values ​​were calculated separately. Energy ratio mean and the average power ratio of HRV low-frequency band to high-frequency band By statistically modeling multiple historical datasets, normal response intervals corresponding to each treatment stage are constructed. For example, the normal response interval corresponding to Stage=2 in treatment is... The normal range is [1.2, 1.8].

[0065] Furthermore, an adaptability quantification model is constructed, and the adaptability of each stage is evaluated using the adaptability quantification model, satisfying the following: in, Indicates the fit score. express The midpoint of the normal range. express The half-width of the normal range. The fit score ranges from [0,1], with the value closer to 1 indicating a higher degree of fit between the current treatment stage and the patient's physiological response.

[0066] S63, fusion phase unique heat encoding, treatment duration and fit score, to obtain the therapy process fit feature vector.

[0067] By fusing features from stage-specific unique heat encoding, treatment duration, and fit score, a treatment process fit feature vector is constructed. This vector has a 5×1 dimension, where the unique heat encoding occupies 3 dimensions to represent the treatment stage; the treatment duration occupies 1 dimension to record the duration of the current treatment; and the fit score occupies 1 dimension to reflect the degree of fit between the current treatment stage and the patient's physiological response. This feature integration method provides a comprehensive and structured data foundation for subsequent analyses such as treatment plan optimization and efficacy prediction.

[0068] Further, by fusing features extracted from the EEG signal feature extraction module, the heart rate variability feature extraction module, and the exposure therapy process feature extraction module through the first feature fusion module, a first feature is obtained, including the following steps: S71. Based on the stage information, perform refined weight allocation on the characteristics of EEG signals and heart rate variability.

[0069] Fine-grained weight allocation is performed based on stage information to meet the monitoring needs under different physiological and psychological states: Stage 1 (Relaxation Period): During this stage, the parasympathetic nervous system dominates, and the physiological state tends to stabilize. Therefore, the focus is on the HRV (Heart Rate Variability, reflecting parasympathetic activity) indicator. The weights of the EEG anxiety-cognitive feature vector and the heart rate variability feature vector are set to 0.3:0.7. By reducing the weight of the EEG, the capture of central nervous signals is weakened, and the representation of the autonomic nervous system regulation state by the HRV data is strengthened. Stage 2 (Low-intensity stimulation period): At this time, the body is in a state of moderate stress, and it is necessary to take into account both the central nervous system response and the autonomic nervous system regulation. Therefore, a balanced weight EEG anxiety-cognitive feature vector: heart rate variability feature vector = 0.5:0.5 is adopted to ensure that the cortical excitability changes captured by EEG and the cardiovascular regulatory capacity fed back by HRV are analyzed to the same extent. Stage 3 (High-intensity stimulation period): As external stimuli intensify, the central nervous system's regulatory role on emotions and stress becomes more prominent. Therefore, the focus is on EEG (reflecting central anxiety levels) indicators, with the weight adjusted to EEG anxiety-cognitive feature vector: heart rate variability feature vector = 0.7:0.3. Cortical electrical activity signals are amplified first to capture potential anxiety or stress responses.

[0070] S72. Based on the weighted feature results, the fit score is used as a dynamic threshold for feature selection to establish a correlation mechanism between data quality and analysis accuracy, and the first feature is obtained.

[0071] Based on the weighted feature results, a domain-knowledge-based weighted fusion strategy is employed to deeply integrate EEG features and HRV features. The original physiological signals are transformed from the time domain to the feature space, generating weighted features with stage adaptability. Subsequently, the two weighted features are combined sequentially to construct a preliminary fusion feature vector. This vector forms a 5×1 feature matrix structure in the spatial dimension, where five elements correspond to the fusion feature values ​​within different time windows. This standardized data preprocessing not only eliminates the dimensional differences of the original signals but also enhances the representational ability of key physiological indicators, providing a unified and standardized input data foundation for subsequent machine learning-based state recognition models.

[0072] Furthermore, by using the fit score as a dynamic threshold for feature selection, a correlation mechanism between data quality and analytical accuracy is established: when When the fit is poor, it indicates that the current data does not match the stage features well enough, and features with low correlation in the initial fusion feature vector need to be removed. For example, in the high-intensity stimulation period of Stage=3, if the SDNN (standard deviation of normal sinus interval) in the HRV index has a weak correlation with central anxiety, it will be removed from the feature vector. After screening, the final first feature is generated, and its dimension is dynamically adjusted in the range of 3~5×1 to ensure that the retained features can accurately reflect the physiological characteristics of the current stage and reduce the interference of redundant data on the analysis results.

[0073] In another embodiment, the step of fusing features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module to obtain the second feature through the second feature fusion module includes the following steps: S81. Calculate the feature correlation between the features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module according to the Pearson correlation coefficient algorithm.

[0074] Pearson correlation coefficient was used as a quantitative tool to perform pairwise analysis on skin conductance response amplitude and gaze dispersion. The Pearson correlation coefficient ranges from -1 to 1, with a value closer to 1 indicating a stronger linear correlation between the features. Taking experimental data under anxious emotional states as an example, when the correlation coefficient R between skin conductance response amplitude and gaze dispersion is greater than 0.6, a significant positive correlation can be determined between the two. This correlation coefficient R is used as a feature association index to characterize the association between emotional state and attention allocation.

[0075] S82. The sliding window mean value filtering algorithm is used to smooth the amplitude and dispersion of the skin conductance response. The second feature is obtained by combining the feature correlation and smoothing results.

[0076] The amplitude of the skin conductance response (GSR) and gaze dispersion were smoothed using a sliding window median filtering algorithm. By setting a sliding window with a fixed length of 3 seconds, the data points within the window were sorted by numerical value, and the median value was used to replace the center data point of the window. This effectively suppressed abnormal fluctuations caused by environmental interference (such as sudden noise causing a sudden rise in GSR signal) or equipment errors.

[0077] The smoothed skin conductance response amplitude and dispersion after outlier suppression, along with the feature correlation index R, are deeply fused. Using vector concatenation, the three key indicators are integrated into a new feature vector as the second feature with a dimension of 3×1. This achieves a multi-dimensional unified representation of physiological signals, attention features, and correlations, providing more representative input data for subsequent machine learning model training.

[0078] Furthermore, through the stimulus parameter adaptive control module, the features extracted by the physiological load state feature extraction module, along with the first feature and the second feature, are combined to complete the automatic control of phobia graded exposure treatment, including the following steps: S91. Determine the suitability level of the current stimulus through fuzzy calculation.

[0079] The stimulus load feature vector, the first feature, and the second feature are input into a fuzzy logic-based computational model. This model maps continuous feature values ​​to a linguistic variable space through a pre-defined membership function, and then outputs the suitability level of the current stimulus. The specific level classification is as follows: Level 1: Too Strong: When physiological signals fluctuate significantly (e.g., a sudden increase in heart rate, or a strong skin conductance response), and the EEG signal exhibits high-frequency anxiety characteristics, and the eye movement pattern shows avoidance gaze, the stimulus intensity is determined to exceed the patient's appropriate range. This situation may cause discomfort to the patient, and the stimulus parameters need to be adjusted promptly.

[0080] Level=2: Adaptation: Data from each module shows a stable physiological state, EEG characteristics reflect moderate activation, and eye movement behavior exhibits a natural exploration mode, indicating that the current stimulation intensity is in the optimal treatment range and can effectively promote the treatment effect. Maintaining the current stimulation parameters is sufficient.

[0081] Level=3: Shortcomings: If there are no significant changes in physiological indicators, low activity of EEG signals, and eye movement data showing distraction, it means that the stimulation intensity has not reached the effective treatment threshold. The stimulation intensity needs to be dynamically adjusted through algorithms to improve the effectiveness of treatment intervention.

[0082] S92. Adjust the naked-eye 3D stimulation parameters based on the judgment results.

[0083] Handling overly strong conditions: When the adaptability level is determined to be Level=1 (overly strong), a two-parameter linkage adjustment strategy is executed: Reduce the parallax parameter of naked-eye 3D display to alleviate visual fatigue; simultaneously reduce the frame rate to decrease the density of visual information input.

[0084] Insufficient status handling: If the adaptability level is Level=3 (insufficient), activate the enhanced parameter adjustment mechanism: Improve the parallax parameters of naked-eye 3D displays to enhance stereoscopic visual effects; increase the complexity of the image and introduce more visually stimulating elements.

[0085] Adaptation status handling: When Level=2 (adaptation) is determined, a fine-tuning strategy is executed: Maintain the stability of core display parameters (parallax, complexity); Based on the gaze stability index, the frame rate is dynamically fine-tuned (within ±1 FPS). If gaze dispersion is detected to exceed the threshold, the frame rate is reduced to decrease visual interference; otherwise, it is moderately increased to maintain visual smoothness.

[0086] Furthermore, a reinforcement learning framework based on temporal difference algorithm is constructed, using changes in patient physiological indicators after parameter adjustment as the core feedback signal. The specific reward and punishment mechanism is as follows: If the adjusted first feature vector contains A 5% decrease in energy ratio grants a +1 bonus; conversely, an increase in energy ratio incurs a -1 penalty.

[0087] The weights of the fuzzy rule base are dynamically updated using a Deep Q-Network (DQN) model, and an experience replay mechanism is employed to mitigate data correlation issues. During the optimization process, the weight update formula for each rule is as follows: in, Indicates the updated weight. Indicates the weight before the update. Indicates the learning rate. Indicates reward and punishment values. This represents the timing difference error.

[0088] After multiple rounds of iterative learning, the optimized combination of naked-eye 3D stimulation parameters is output, including updated parallax parameters, frame rate parameters, and image complexity parameters, to achieve personalized dynamic stimulation control.

[0089] In summary, this invention provides a precise and automated solution for graded exposure therapy for phobias through modular design and graded feature fusion. Based on multimodal physiological signals from EEG, heart rate variability, skin conductance, and eye movement, it combines two levels of feature fusion (first, central physiological and process features; second, peripheral emotional and behavioral features) to extract key information layer by layer, improving the accuracy of patient anxiety and physiological state assessment. By linking physiological load features with the two-level fusion features and using an adaptive stimulus parameter control module, it achieves dynamic and personalized adjustment of the intensity of therapeutic stimuli, avoiding discomfort or insufficient efficacy caused by one-size-fits-all parameter settings, and aligning with the core requirement of gradually increasing tolerance in graded exposure. The fully automated control process reduces human intervention, improves treatment efficiency and stability, provides intelligent technical support for clinical treatment of phobias, and promotes the transformation of exposure therapy from experience-based to data-driven precision medicine.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. An automatic control system for graded exposure therapy of phobias based on naked-eye 3D, characterized in that, include: The system includes modules for extracting features from electroencephalogram (EEG) signals, heart rate variability, skin conductance response, eye movement, physiological stress, exposure therapy process, first feature fusion, second feature fusion, and adaptive control of stimulation parameters. The first feature fusion module is used to fuse features extracted from the EEG signal feature extraction module, the heart rate variability feature extraction module, and the exposure therapy process feature extraction module to obtain a first feature; The second feature fusion module is used to fuse the features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module to obtain a second feature; The stimulus parameter adaptive control module is used to integrate the features extracted by the physiological load state feature extraction module with the first feature and the second feature to complete the automatic control of phobia graded exposure therapy.

2. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The extraction of features through the EEG signal feature extraction module includes the following steps: Noise reduction processing is performed on the raw EEG signals; Based on the processing results, short-time Fourier transform analysis is performed, and then the energy integration method is used to calculate the relative energy, quantifying the energy proportion characteristics of different frequency bands.

3. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The heart rate variability feature extraction module extracts features, including the following steps: Purify the heart rate interval; Based on the purified heart rate interval sequence, time-domain and frequency-domain indices were analyzed to obtain the heart rate variability feature vector.

4. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The skin conductance feature extraction module extracts features, including the following steps: The baseline of skin conductance response is dynamically calibrated using a sliding window mechanism; Based on the baseline-calibrated skin conductance response signal, anxiety-related transient skin conductance response events are detected using the slope threshold method, and a three-dimensional emotional arousal feature vector is constructed.

5. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The eye-tracking feature extraction module extracts features, including the following steps: Filter the field of view to obtain an effective fixation sequence; The dispersion of fixation points in the effective fixation sequence was calculated using the standard deviation ellipse method. By combining fixation dispersion, average pupil diameter, and average blink frequency, an eye movement feature vector is constructed using principal component analysis.

6. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The physiological load state feature extraction module extracts features, including the following steps: Based on historical physiological tolerance data, a safe threshold range for key parameters of naked-eye 3D stimulation was constructed. Based on the safety threshold range, the physiological load value is calculated using the physiological load value model. Combining the physiological load value with the disparity magnitude and the stimulus frame rate, a stimulus load feature vector is obtained.

7. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, Features are extracted using the exposure therapy process feature extraction module, including the following steps: The unique thermal coding technique is used to encode the stage features of the treatment phase; Construct an adaptability quantification model and use the adaptability quantification model to evaluate the adaptability of the stage; The therapy process adaptation feature vector is obtained by combining the unique heat encoding of the fusion phase, treatment duration, and adaptability score.

8. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The first feature fusion module fuses features extracted from the EEG signal feature extraction module, the heart rate variability feature extraction module, and the exposure therapy process feature extraction module to obtain a first feature, including the following steps: Based on the stage information, the weights of EEG signal features and heart rate variability features are finely allocated; Based on the weighted feature results, the fit score is used as a dynamic threshold for feature selection to establish a correlation mechanism between data quality and analysis accuracy, and the first feature is obtained.

9. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The second feature fusion module fuses features extracted from the skin conductance feature extraction module and the eye movement feature extraction module to obtain a second feature, including the following steps: Based on the Pearson correlation coefficient algorithm, the feature correlation between the features extracted from the skin conductance response feature extraction module and the eye movement feature extraction module is calculated. A sliding window midpoint filtering algorithm is used to smooth the amplitude and dispersion of the skin conductance response. The second feature is obtained by combining the feature correlation and the smoothing result.

10. The automatic control system for graded exposure therapy of phobia based on naked-eye 3D according to claim 1, characterized in that, The stimulation parameter adaptive control module, by integrating the features extracted by the physiological load state feature extraction module with the first and second features, completes the automatic control of phobia graded exposure therapy, including the following steps: The suitability level of the current stimulus is determined by fuzzy logic. Adjust the naked-eye 3D stimulation parameters based on the assessment results.

Images