An emotional intervention system and device based on nostalgia therapy

Abstract

The application discloses an emotion intervention system and device based on nostalgia therapy, and belongs to the technical field of mental health intervention and intelligent medical equipment. The device comprises a stimulus presentation, a physiological signal collection, a behavior reaction identification, a neural response modeling, an intervention effect evaluation, and a central control and feedback adjustment subsystem. A closed-loop intervention architecture is formed through a real-time communication bus interconnection. An individualized life event image set is presented. Multi-modal physiological signals and behavior reactions are synchronously collected. A neural response model is constructed based on abnormal activation patterns of brain regions revealed by functional magnetic resonance imaging. The activation level of target brain regions is predicted. The intervention effect is evaluated by comprehensively evaluating multiple indicators. The central control system uses a hierarchical reinforcement learning algorithm. The stimulus sequence is dynamically adjusted according to the evaluation results and the neural response prediction value. The individualized and precise emotion intervention for subclinical depression is realized. The intervention effect and the neural regulation pertinence are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of mental health intervention and intelligent medical equipment, and in particular relates to an emotion intervention system and device based on nostalgia therapy. BACKGROUND

[0002] Older mental health problems have become an important issue in the field of public health. Among various types of mental disorders in the elderly, depressive symptoms, as a precursor of depression, have a high incidence and disease burden, and if not intervened in time, they are likely to progress to clinical depression, thereby significantly affecting the quality of life, social function and even physiological health of individuals. Therefore, early and effective emotional intervention for depressive symptoms in the subclinical stage of the elderly has become one of the key goals in the field of mental health and geriatric care. Under this background, non-drug psychological intervention methods have gradually attracted widespread attention from academia and the practical field due to their high safety, good compliance, and few adverse reactions.

[0003] Cognitive psychological intervention, as a type of intervention strategy that focuses on regulating thinking patterns and emotional processing, has shown good application prospects in improving emotional disorders in the elderly. Nostalgia therapy, as an important branch of cognitive psychological intervention, guides individuals to review their past life experiences, reconstruct the meaning of life, and promote self-identity, and has been widely practiced in community care institutions and long-term care environments. Its simple operation, low cost, and strong cultural adaptability make it particularly suitable for resource-limited elderly service scenarios, and it has been proven to effectively enhance self-esteem, alleviate feelings of loneliness, improve life satisfaction, and improve mild cognitive function and emotional state.

[0004] Therefore, how to organically integrate behavioral observation and neuroimaging evidence to build a multi-dimensional analysis model covering the whole process of "stimulus-response-brain mechanism-emotional change", and on this basis, develop an emotion intervention system with neuroscientific basis, quantifiable and verifiable, has become a key challenge and technical problem to be solved for technical personnel in the field. SUMMARY

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] According to the first aspect of the present application, an emotion intervention device based on nostalgia therapy is claimed, comprising a stimulus presentation subsystem, a physiological signal acquisition subsystem, a behavioral response recognition subsystem, a neural response modeling subsystem, an intervention effect evaluation subsystem, and a central control and feedback adjustment subsystem, each subsystem is interconnected through an industrial-grade real-time communication bus to form a closed-loop emotion intervention architecture;

[0007] The stimulation presentation subsystem is used to present individualized life event image sets to the subject, the image sets are divided into five subsets of childhood, adolescence, early adulthood, middle age and old age according to the life stage, and each image is marked with a timestamp, geographical coordinates, a character relationship label and a three-dimensional emotion model quantified emotional valence grade;

[0008] The physiological signal acquisition subsystem includes a scalp electroencephalogram acquisition unit, a heart rate variability monitoring unit, a skin conductance response detection unit and a respiratory frequency sensing unit, which are used to synchronously acquire multi-modal physiological signals and mark them with UTC timestamps;

[0009] The behavior response recognition subsystem includes an audio acquisition module and a video behavior coding module, which are used to capture the speech and facial behavior of the subject, and classify and recognize the behavior response through a deep learning model deployed in a behavior feature extraction server;

[0010] The neural response modeling subsystem is based on the abnormal activation pattern of the brain region related to autobiographical memory extraction in subclinical depression state revealed by functional magnetic resonance imaging, and constructs a repetitive transcranial magnetic stimulation (RTMS) of the target regulation target area including bilateral superior temporal gyrus, left inferior frontal gyrus, right insular lobe, left cingulate gyrus and bilateral cerebellar anterior lobe, and uses a dynamic causal model and a multilayer perception regression model to map real-time electroencephalogram features to RTMS expected activation level;

[0011] The intervention effect evaluation subsystem calculates a comprehensive intervention effect index (CIEI) according to the theta frequency band power change rate of electroencephalogram, the change of cognitive behavior exploration type behavior proportion and the decline rate of negative semantic feature weight; The central control and feedback adjustment subsystem runs a hierarchical reinforcement learning controller to dynamically generate an optimal stimulation sequence based on CIEI, RTMS predicted activation level and current stimulation parameters.

[0012] Further, the individualized life event image set is stored in a redundantly configured solid state storage array, encoded in a lossless compression format, and the emotional valence grade is determined by averaging the average values of independent evaluations according to a standardized scale, the evaluation results need to meet the consistency standard, the image carousel duration is set to a default value in a adjustable range, and blank screens are inserted before and after as baseline periods.

[0013] Further, the scalp electroencephalogram acquisition unit adopts a multi-lead wet electrode cap, the electrode material is silver-coated silver chloride, the impedance is controlled within a low impedance range, the sampling frequency is high frequency, the band pass filter covers the low frequency to high frequency range, the reference electrode is a bilateral mastoid average reference, the ground wire is placed at the forehead position, the data is transmitted through a wireless protocol, the end-to-end delay is low, and the physiological artifacts are removed by independent component analysis algorithm.

[0014] Further, the heart rate variability monitoring unit employs a chest strap type photoelectric sensor to acquire a pulse wave signal, with a medium-high frequency sampling rate, and the R-R interval sequence is subjected to abnormal interval detection, interpolation correction and detrending processing according to a standard procedure, and the ratio of low frequency power to high frequency power is used as an autonomic nervous balance index to participate in the regulation and decision-making.

[0015] Further, the skin conductance response detection unit fixes two disc electrodes between the fingers, applies a direct current bias voltage, and the sampling rate is medium frequency. After low-pass filtering, the original signal separates the skin conductance level and the skin conductance response. The skin conductance response peak amplitude exceeding the threshold value and the rise time being shorter than the set value define an effective emotional arousal event.

[0016] Further, the audio acquisition module is configured with a super-directivity microphone installed above both sides of the display unit, with the pickup angle controlled in a conical region. After noise reduction processing, the voice signal is input into a neural network-based speech recognition engine, and the transcribed text is sent into a semantic model to generate a semantic vector embedding for behavior classification.

[0017] Further, the video behavior coding module employs an infrared camera installed at a proper height and distance in front, and cooperates with a near-infrared fill light to realize clear imaging in dark environments. The video stream is input into a behavior feature extraction server equipped with a graphics processing unit, and a behavior classification model with a convolutional neural network as the backbone network is run. The input features include audio spectrum graph, facial action unit intensity, head posture angle change rate and semantic vector, and the output is a multi-class behavior probability distribution.

[0018] Further, the behavior occurrence determination rule is that the probability of a certain category continuously exceeds the threshold value for more than a preset time, and then the behavior event is recorded, as well as the duration and intensity level, wherein the intensity level is divided into multiple levels, and the model training data comes from real intervention videos independently coded by psychology experts and meeting consistency;

[0019] The neural response modeling subsystem acquires the subject's structural MRI image in the individualized calibration program, performs bias field correction, tissue segmentation and nonlinear registration to the standard space, and projects the target area RTMS mask to the individual anatomical structure for subsequent brain electrical source localization analysis.

[0020] Further, the multi-layer perception regression model input features include event-related potential amplitude, component amplitude and frequency band power, which are input into a network containing multiple fully connected hidden layers after standardization, with a nonlinear function as the activation function, and the output is the RTMS blood oxygen level dependent signal change rate prediction value. The model is deployed in an acceleration engine.

[0021] The central control and feedback adjustment subsystem runs a closed-loop control algorithm, which is essentially a hierarchical reinforcement learning controller. The upper policy network uses a reinforcement learning algorithm. The state space includes CIEI values, RTMS predicted activation mean, current image emotional valence mean, the number of presented images, and life stage position. The action space includes maintaining the current stage, switching to a higher positive valence subset, introducing prosocial theme images, inserting cognitive reappraisal guidance sentences, or terminating the conversation.

[0022] The central control and feedback adjustment subsystem also has a lower rule engine built in, which contains multiple deterministic control rules with higher priority than the upper policy output.

[0023] According to the second aspect of the present application, an emotion intervention system based on nostalgia therapy is claimed, comprising:

[0024] One or more processors;

[0025] A memory having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, so that the one or more processors are connected with the stimulus presentation subsystem, the physiological signal acquisition subsystem, the behavior reaction recognition subsystem, the neural response modeling subsystem, the intervention effect evaluation subsystem, and the central control and feedback adjustment subsystem of the emotion intervention device based on nostalgia therapy, and perform corresponding operations.

[0026] The present application discloses an emotion intervention system and device based on nostalgia therapy, belonging to the technical field of mental health intervention and intelligent medical equipment. The device includes stimulus presentation, physiological signal acquisition, behavior reaction recognition, neural response modeling, intervention effect evaluation, and central control and feedback adjustment subsystems. Through real-time communication bus interconnection, a closed-loop intervention architecture is formed. By presenting a personalized set of life event images, multi-modal physiological signals and behavior reactions are synchronously acquired. Based on the abnormal activation patterns of brain regions revealed by functional magnetic resonance imaging, a neural response model is constructed to predict the activation level of target brain regions. Intervention effects are evaluated by integrating multiple indicators. The central control system uses a hierarchical reinforcement learning algorithm to dynamically adjust the stimulus sequence based on evaluation results and neural response predictions, achieving personalized and precise emotion intervention for subclinical depression, and effectively improving intervention effectiveness and neural modulation relevance. BRIEF DESCRIPTION OF DRAWINGS

[0027] Fig. 1 The structure module diagram of the emotion intervention device based on nostalgia therapy claimed by the embodiments of the present application;

[0028] Fig. 2 The second structure module diagram of the emotion intervention device based on nostalgia therapy claimed by the embodiments of the present application;

[0029] Fig. 3 A third structural module diagram of an emotion intervention device based on nostalgia therapy claimed in the embodiments of the present application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0031] The terms "first", "second", "third" in the present application are only used for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise explicitly and specifically limited. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0032] In this document, reference to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. The appearances of the phrase "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily mutually exclusive of one another. A person of ordinary skill in the art will understand that the embodiments described herein can be combined with other embodiments.

[0033] Traditional reminiscence therapy is usually conducted in a group setting, relying on cues such as photos, objects, or music to evoke participants' autobiographical memories, and then to promote emotional expression and interpersonal interaction. In recent years, individualized reminiscence therapy has gradually emerged, emphasizing the customization of intervention content according to the individual's life history characteristics, and improving the pertinence and depth of therapy. Studies have shown that such interventions can significantly reduce the levels of depression and anxiety in the elderly in the short term, and behavioral observations have observed the reorganization of reminiscence functions and the optimization of cognitive emotional regulation strategies, such as the reduction of negative coping styles such as self-blame and catastrophic thinking, and the enhancement of adaptive regulation mechanisms such as positive reappraisal. Furthermore, the behavioral response patterns exhibited by participants during the intervention process, including descriptive statements, emotional exploration, cognitive reconstruction, and therapeutic insights, have also been found to be significantly associated with emotional improvement, suggesting that the intervention process itself contains important mechanism information, rather than just changes in outcome indicators. However, as related research deepens and clinical practice demands for intervention precision increase, the traditional research paradigm based on behavioral observation and scale assessment begins to show inherent limitations in explanatory power and objectivity. The reason is that changes in behavior can reflect trends in psychological state changes, but it is difficult to reveal the underlying neurobiological basis, especially to elucidate how specific stimulus tasks activate or regulate specific brain function networks to achieve emotional improvement. Although functional magnetic resonance imaging technology has been widely used in brain mechanism research on patients with depression, especially in the analysis of neural circuits related to autobiographical memory processing, most existing research has focused on patients with diagnosed major depression, and there is still a lack of systematic exploration of the brain function response characteristics of older people in the subclinical depression state, especially when they receive structured reminiscence interventions.

[0034] More importantly, current research on the mechanism of reminiscence therapy generally stays at the level of correlation analysis of behavioral and psychological variables, and has not established a complete causal chain from external stimulus input, internal neural activity response, to final emotional output. This break in the theoretical framework leads to two outstanding problems: first, it is difficult to determine whether emotional improvement is due to memory awakening itself or to the accompanying emotional processing and cognitive reappraisal process; second, it is difficult to identify which brain regions have abnormal functions that limit the effectiveness of reminiscence therapy in subclinical depression. In older women with depressive symptoms, the autobiographical memory network composed of the prefrontal-limb system-parahippocampal gyrus shows a widespread decrease in activation when viewing highly arousing old photos, involving key regions such as the bilateral superior temporal gyrus, left inferior frontal gyrus, right insular lobe, and cingulate gyrus, which are closely related to episodic memory retrieval, emotional regulation, and social cognition. This result suggests that even in a state of depression that does not meet the diagnostic criteria, an individual's neural function has already shifted measurably, and traditional reminiscence therapy, which relies solely on subjective guidance and ignores the targeted regulation of these underlying neural dysfunction, may limit the maximization of its intervention effectiveness.

[0035] According to the first embodiment of the present invention, referring to Figs. 1-3 This invention claims protection for an emotion intervention device based on nostalgia therapy, comprising a stimulus presentation subsystem, a physiological signal acquisition subsystem, a behavioral response recognition subsystem, a neural response modeling subsystem, an intervention effect evaluation subsystem, and a central control and feedback regulation subsystem. Each subsystem is interconnected through an industrial-grade real-time communication bus to form a closed-loop emotion intervention architecture.

[0036] The stimulus presentation subsystem is used to present a personalized set of life event images to the subject. The set of images is divided into five subsets according to life stages: childhood, adolescence, early adulthood, middle age, and old age. Each image is labeled with a timestamp, geographic coordinates, relationship tags, and emotional valence level quantified based on a three-dimensional emotion model.

[0037] The physiological signal acquisition subsystem includes a scalp EEG acquisition unit, a heart rate variability monitoring unit, a skin conductance response detection unit, and a respiratory rate sensing unit, which are used to simultaneously acquire multimodal physiological signals and add UTC timestamps.

[0038] The behavioral response recognition subsystem includes an audio acquisition module and a video behavior encoding module, which are used to capture the speech and facial behavior of the subjects and classify and recognize the behavioral responses through a deep learning model deployed on the behavioral feature extraction server.

[0039] The neural response modeling subsystem is based on the abnormal activation patterns of brain regions related to autobiographical memory retrieval in subclinical depression revealed by functional magnetic resonance imaging. It constructs a repetitive transcranial magnetic stimulation (RTMS) targeting the bilateral superior temporal gyrus, left inferior frontal gyrus, right insula, left cingulate gyrus, and bilateral anterior cerebellar lobes. It also uses a dynamic causal model and a multilayer perceptron regression model to map real-time EEG features to the expected activation level of the RTMS.

[0040] The intervention effect evaluation subsystem calculates the comprehensive intervention effect index (CIEI) based on the change rate of the theta band power of EEG, the change in the proportion of cognitive and exploratory behaviors, and the decrease rate of negative semantic feature weights. The central control and feedback regulation subsystem operates a hierarchical reinforcement learning controller to dynamically generate the optimal stimulus sequence based on the CIEI, RTMS predicted activation levels, and current stimulus parameters.

[0041] The embodiment provides an emotion intervention system and a special device thereof, which integrates neuroscience mechanism analysis, individualized stimulation content generation, multi-modal physiological signal monitoring and closed-loop feedback control. The system integrates the abnormal brain function network features of subclinical depression state of the elderly during the autobiographical memory extraction process revealed by functional magnetic resonance imaging, combines the structured coding analysis of the behavior response pattern in the intervention process, and constructs a precise nostalgia therapy implementation framework guided by neural plasticity regulation, so as to realize early, targeted and quantifiable intervention on the depression symptoms of the elderly group. The system includes a stimulation presentation subsystem, a physiological signal acquisition subsystem, a behavior response recognition subsystem, a neural response modeling subsystem, an intervention effect evaluation subsystem and a central control and feedback regulation subsystem, each subsystem is interconnected through an industrial real-time communication bus to form a closed-loop emotion intervention architecture with dynamic adaptability.

[0042] The stimulation presentation subsystem is arranged in the visual range of the subject, and the core component thereof is a high-resolution liquid crystal display unit, which is connected with an embedded graphics processing module through an HDMI 2.1 interface, supports 4K@60Hz video signal output, covers sRGB 100% in color gamut, and has a peak brightness of not less than 500 cd / m², so as to ensure that the color restoration degree and detail performance of the image meet the presentation requirements of the emotion arousal stimulation material. The stimulation material is a set of individualized life event images collected and digitized in advance, each image is labeled with accurate time stamp, geographic coordinate, character relationship label and emotional valence grade, wherein the emotional valence grade is quantitatively labeled according to a three-dimensional emotional model through a standardized emotion rating scale, the range is 1 to 9 points, the valence is less than 4 points for negative memory image, more than 7 points for positive memory image, and between 4 and 7 points for neutral memory image. The image set is divided into five life stage subsets of childhood (0-12 years old), adolescence (13-20 years old), early adulthood (21-40 years old), middle age (41-60 years old) and old age (61 years old and above) according to time sequence, each subset contains not less than 30 representative images with high arousal, the image resolution is not less than 1920x1080 pixels, and the JPEG 2000 lossless compression format is used to store in a solid state disk array. The graphics processing module runs a Linux real-time operating system, loads an image rendering engine based on OpenGL ES 3.0, can automatically cycle the images according to the preset time interval, and inserts a 500 millisecond gray blank screen as a baseline period before and after each image presentation to eliminate the visual residual effect;

[0043] The physiological signal acquisition subsystem is composed of a multi-channel bioelectricity sensor array, including a scalp EEG acquisition unit, a heart rate variability monitoring unit, a skin conductance response detection unit, and a respiratory frequency sensing unit. The scalp EEG acquisition unit uses a 64-lead wet electrode cap with an international 10-20 system standard layout. The electrode material is silver-plated silver chloride, with an impedance controlled below 5 kΩ. The sampling frequency is set to 1000 Hz, and the band-pass filter range is 0.1-100 Hz. The reference electrode is a bilateral mastoid average reference, and the ground wire is placed at the FPz position on the forehead. The data is transmitted to the central control unit through the wireless Bluetooth 5.2 protocol, with a transmission delay of less than 10 milliseconds. The heart rate variability monitoring unit obtains the pulse wave signal through a chest strap photoelectric sensor with a sampling rate of 250 Hz. The R-R interval sequence is used for time and frequency domain analysis, and the ratio of low-frequency power to high-frequency power is used as an indicator of autonomic nervous balance. The skin conductance response detection unit fixes two circular stainless steel electrodes between the index finger and middle finger of the non-dominant hand, applies a 0.5V direct current bias voltage, measures the skin conductance level and skin conductance response, and has a sampling rate of 32 Hz for evaluating the intensity of emotional arousal. The respiratory frequency sensing unit uses a piezoelectric respiratory belt tied to the subject's chest and abdomen junction. The output signal is amplified and filtered to extract the respiratory cycle, with a sampling rate of 100 Hz. All physiological signals are synchronized with UTC time stamps.

[0044] The behavioral response recognition subsystem includes a high-fidelity audio acquisition module and a video behavior coding module. The audio acquisition module is configured with two directional microphones installed on both sides of the display unit, with a sampling rate of 48 kHz, a quantization accuracy of 24 bits, and a signal-to-noise ratio of more than 90 dB, for capturing the subject's verbal response during image presentation. The video behavior coding module uses an infrared camera installed 1.5 meters in front, with a near-infrared fill light to achieve clear imaging in dark environments. The collected audio and video data streams are input in real time to the behavior feature extraction server, which deploys a behavior classification model based on deep learning. The backbone network of the model is ResNet-50 architecture, and the output layer is connected to five behavior response labels: descriptive statement, emotional exploration, cognitive behavior exploration, resistance behavior, and therapeutic change. The model training data comes from 120 hours of intervention session videos independently coded and highly consistent in previous research. The input features include speech spectrogram, facial action unit intensity, head posture angle change rate, and semantic vector embedding. The semantic vector is generated by the pre-trained language model BERT-base, and the input text is obtained by real-time transcription by the speech recognition module. The model output is the probability distribution of each behavior category in each second time window. When the probability of a certain category exceeds the threshold value of 0.7 for more than 3 seconds, it is determined that the behavior occurs, and the start and end time, duration, and intensity level are recorded.

[0045] The neural response modeling subsystem establishes an individualized brain function response prediction model based on the results of previous functional magnetic resonance studies. According to the research data, older women with depressive symptoms showed significant activation reduction in the bilateral superior temporal gyrus, left inferior frontal gyrus, right insular lobe, left cingulate gyrus, and bilateral cerebellar anterior lobe when watching highly arousing old photos. The t-test difference was statistically significant. The modeling subsystem defines the above brain regions as target regulation target areas and uses structural and functional data from the public neuroimaging database to construct a group prior template. For each new subject, the system first obtains their T1-weighted structural image, calculates the gray matter volume density map through voxel-based morphological analysis, and registers it to the MNI standard space. Then, a dynamic causal model is used to construct an effective connection network from the prefrontal cortex to the limbic system, with the initial parameters initialized by the group average estimate. During each intervention process, the system receives real-time EEG data from the EEG acquisition unit, selects relevant EEG components corresponding to RTMS functions: P300 amplitude as the episodic memory extraction indicator, N2 component as the conflict monitoring indicator, and theta band power as the prefrontal-hippocampus coupling strength indicator. These EEG-derived features are input into a pre-trained multilayer perceptron regression model, which takes the fMRI BOLD signal change rate as the output target. The hidden layer contains three fully connected layers, the activation function is LeakyReLU, the output layer is a linear function, the loss function is mean squared error, the training set comes from the synchronous acquisition of EEG-fMRI data, and the cross-validation R² reaches 0.73. The model outputs the predicted value of the expected activation level of each RTMS under the current stimulation condition in real time, which serves as a proxy indicator of the neural regulation effect.

[0046] The intervention effect evaluation subsystem integrates multi-dimensional quantitative indicators to dynamically evaluate the psychological state changes in the intervention process. This subsystem continuously receives raw data streams from the physiological signal acquisition subsystem and the behavior response recognition subsystem, and performs the following calculation process:

[0047] First, the degree of mood relief was represented by the relative power change rate of theta band (4-8 Hz) in the left prefrontal cortex (Fp1, F3, F7) of EEG, calculated as: (Post_theta - Pre_theta) / Pre_theta x 100%, where Pre_theta is the average power in the baseline period, and Post_theta is the average power in the 5-10 seconds window after stimulation. Negative changes represent mood improvement. Second, the degree of reminiscence function reorganization was quantified by the total proportion of two types of behaviors, "cognitive behavioral exploration" and "emotional exploration", in the behavioral response, calculated as: (Post_ratio - Pre_ratio) / Pre_ratio x 100%, where Pre_ratio is the average proportion of the first three intervention sessions, and Post_ratio is the proportion of the current session. Third, the degree of optimization of cognitive emotion regulation strategies was represented by the decline rate of the total TF-IDF weight of negative regulation methods such as self-blame, rumination, and catastrophic related speech semantic features, with a keyword library containing 27 entries such as "I shouldn't", "if only", "finished", "all my fault", etc., and a weight update period of every sentence. Finally, the system weighted and combined the above three indicators to form a comprehensive intervention effect index, with weight coefficients of 0.4, 0.35, and 0.25, respectively. Each 5% increase in CIEI was considered as an effective progress stage.

[0048] The central control and feedback regulation subsystem, as the core decision-making unit of the entire device, uses an industrial-grade embedded computer as its hardware platform, running a real-time operating system (RT-Linux) with a kernel scheduling period of 1 ms. This subsystem runs a closed-loop control algorithm, which is essentially a hierarchical reinforcement learning controller, consisting of an upper policy network and a lower execution network. The upper policy network uses the deep deterministic policy gradient algorithm, with a state space composed of CIEI, RTMS predicted activation level, current image emotional valence, number of presented images, and time series position. The action space is the selection strategy for the next stimulus image, including: maintaining the current life stage, switching to a higher positive valence subset, introducing prosocial theme images, inserting cognitive reappraisal guiding sentences, or terminating the current session. The lower execution network is a rule engine with 12 deterministic control rules, such as: "If the theta power change rate is less than -15% for two consecutive stimuli and the cognitive behavioral exploration proportion is less than 20%, select a positive image with valence ≥8 from the early adult stage subset for intensive stimulation"; "If the skin conductance response peak value exceeds the baseline by 2 standard deviations and there is a resistance behavior lasting more than 10 seconds, insert 30 seconds of soothing music and switch to a neutral image sequence". All rules are based on the behavioral-neuro-emotional correlation laws discovered in previous studies, and have clear clinical basis.

[0049] Further, the individualized life event image set is stored in a redundantly configured solid state storage array, encoded in a lossless compression format, and the emotional valence level is determined by averaging the values independently assessed according to a standardized scale, the assessment results need to meet the consistency standard, the image carousel duration is set to a default value in a adjustable range, and a blank screen is inserted before and after as a baseline period.

[0050] Further, the scalp electroencephalogram acquisition unit adopts a multi-lead wet electrode cap, the electrode material is silver-coated silver chloride, the impedance is controlled within a low impedance range, the sampling frequency is high frequency, the band-pass filter covers the low frequency to high frequency range, the reference electrode is a bilateral mastoid average reference, the ground wire is placed on the forehead, the data is transmitted through a wireless protocol, the end-to-end delay is low, and the independent component analysis algorithm is used to remove physiological artifacts.

[0051] Further, the heart rate variability monitoring unit uses a chest strap type photoelectric sensor to obtain pulse wave signals, the sampling rate is medium-high frequency, the R-R interval sequence is subjected to abnormal interval detection, interpolation correction and detrend processing through a standard process, and the ratio of low frequency power to high frequency power is used as an autonomic nervous balance index to participate in regulation and decision-making.

[0052] Further, the skin conductance response detection unit fixes two disc electrodes between the fingers, applies a direct current bias voltage, and the sampling rate is medium frequency. The original signal is separated into skin conductance level and skin conductance response after low-pass filtering. The skin conductance response peak amplitude exceeding the threshold value and the rise time shorter than the set value are defined as effective emotional arousal events.

[0053] Further, the audio acquisition module is configured with a super-directivity microphone installed on the top of both sides of the display unit, the pickup angle is controlled in a conical region, the voice signal is input into a voice recognition engine based on a neural network after noise reduction processing, and the transcribed text is sent into a semantic model to generate a semantic vector embedding for behavior classification.

[0054] Further, the video behavior encoding module uses an infrared camera installed at a proper height and distance in front, and cooperates with a near-infrared fill light to realize clear imaging in dark environment. The video stream is input into a behavior feature extraction server equipped with a graphics processing unit, and a behavior classification model with a convolutional neural network as the backbone network is run. The input features include audio spectrum graph, facial action unit intensity, head posture angle change rate and semantic vector, and the output is a multi-class behavior probability distribution.

[0055] Further, the behavior occurrence determination rule is that the probability of a certain category continuously exceeds the threshold value for more than a preset time, then the behavior event is recorded, and the duration and intensity level are also recorded, wherein the intensity level is divided into multiple levels, and the model training data comes from real intervention videos that are independently encoded and meet the consistency standard.

[0056] The neural response modeling subsystem acquires the subject structural MRI image in the individualized calibration procedure, performs bias field correction, tissue segmentation and nonlinear registration to standard space, back-projects the target adjustment target RTMS mask to the individual anatomy for subsequent electroencephalography source localization analysis.

[0057] Further, the multi-layer perceptron regression model input features include event-related potential amplitude, component amplitude and frequency band power, which are standardized and input into a network containing multiple fully connected hidden layers, with a nonlinear function as the activation function, and the output is the RTMS blood oxygen level dependent signal change rate prediction value, and the model is deployed in an acceleration engine.

[0058] The central control and feedback regulation subsystem runs a closed-loop regulation algorithm, which is essentially a hierarchical reinforcement learning controller. The upper policy network uses a reinforcement learning algorithm. The state space includes CIEI values, RTMS predicted activation mean, current image emotional valence mean, the number of presented images, and life stage position. The action space includes maintaining the current stage, switching to a higher positive valence subset, introducing prosocial theme images, inserting cognitive reappraisal guidance sentences, or terminating the conversation.

[0059] The central control and feedback regulation subsystem also has a lower rule engine built in, which contains multiple deterministic regulation rules with higher priority than the upper policy output.

[0060] According to the second embodiment of the present application, the present application claims to protect an emotion intervention system based on nostalgia therapy, comprising:

[0061] One or more processors;

[0062] A memory having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors are connected to and perform corresponding operations on the stimulus presentation subsystem, physiological signal acquisition subsystem, behavior response recognition subsystem, neural response modeling subsystem, intervention effect evaluation subsystem, and central control and feedback regulation subsystem of the emotion intervention device based on nostalgia therapy.

[0063] A specific embodiment and comparative example are provided below to illustrate the technical effects of the present application.

[0064] Embodiment:

[0065] A 72-year-old female subject S20231015 was selected, a retired teacher, with a GDS-15 score of 7, meeting the mild depressive symptom criteria. A 7-day calibration procedure was performed, and the system identified that her high-arousal images were mainly concentrated in childhood rural life scenes and early adulthood teaching award moments. It was also found that "disease hospitalization" images triggered a significant negative semantic response, with TF-IDF weight rising by 42%.

[0066] Based on this, an intervention plan was generated: interventions were conducted every Monday, Wednesday, and Friday at 9:30 AM, with each session consisting of six blocks. The first block consistently used positive childhood images, with subsequent blocks dynamically adjusted according to closed-loop regulation. After 12 interventions over 4 consecutive weeks, data showed that the power change rate decreased from -6.3% in the first session to -18.7% in the 12th session; the proportion of cognitive behavioral exploration increased from 18.5% initially to 39.2%; the frequency of negative semantic features decreased by 61%; and the CIEI cumulatively increased by 23.4 percentage points, spanning four effective progress stages. From the 13th session onwards, the system recommended reducing the intervention frequency to twice a week to maintain consolidation.

[0067] Comparative example:

[0068] Another 74-year-old female subject, C20231016, with similar basic characteristics, was selected but received traditional, non-personalized nostalgia therapy: a therapist manually selected a generic set of old photos without emotional labeling, weekly face-to-face conversations, no monitoring of physiological signals, and no closed-loop feedback. After 12 interventions, theta power was...

[0069] The rate of change improved only from -5.9% to -9.1%; the proportion of cognitive behavioral exploration increased from 17.8% to 26.3%; negative semantic features decreased by 28%; and the CIEI improved by 9.6 percentage points. These results are significantly lower than those of the previous example.

[0070] Experimental results show that the nostalgia-based emotion intervention system and device provided by this invention can significantly improve the accuracy and effectiveness of intervention for geriatric depressive symptoms through individualized stimulus content generation, multimodal physiological monitoring, and closed-loop feedback regulation mechanisms.

[0071] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0072] In addition, the various functional units in the embodiments of the present application can be integrated in one processing unit, or each can exist as an independent physical unit, or two or more than two of them can be integrated in one physical unit. The above-mentioned integrated unit can be implemented in the form of hardware, or in the form of a software functional unit. The above is only an embodiment of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation using the content of the specification and drawings of the present application, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

[0073] The specific embodiments of the application are described above, but it is only as an example, and the present application is not limited to the specific embodiments described above. Any equivalent modification or replacement of the present application for those skilled in the art is also within the scope of the present application, therefore, any equivalent transformation and modification, improvement, etc. made without departing from the spirit and principle range of the present application should be covered in the scope of the present application.

Claims

1. An emotion intervention device based on nostalgia therapy, characterized in that, It includes a stimulus presentation subsystem, a physiological signal acquisition subsystem, a behavioral response recognition subsystem, a neural response modeling subsystem, an intervention effect evaluation subsystem, and a central control and feedback regulation subsystem. Each subsystem is interconnected through an industrial-grade real-time communication bus to form a closed-loop emotion intervention architecture. The stimulus presentation subsystem is used to present a personalized set of life event images to the subject. The set of images is divided into five subsets according to life stages: childhood, adolescence, early adulthood, middle age, and old age. Each image is labeled with a timestamp, geographic coordinates, relationship tags, and emotional valence level quantified based on a three-dimensional emotion model. The physiological signal acquisition subsystem includes a scalp EEG acquisition unit, a heart rate variability monitoring unit, a skin conductance response detection unit, and a respiratory rate sensing unit, which are used to simultaneously acquire multimodal physiological signals and add UTC timestamps. The behavioral response recognition subsystem includes an audio acquisition module and a video behavior encoding module, which are used to capture the speech and facial behavior of the subjects and classify and recognize the behavioral responses through a deep learning model deployed on the behavioral feature extraction server. The neural response modeling subsystem is based on the abnormal activation patterns of brain regions related to autobiographical memory retrieval in subclinical depression revealed by functional magnetic resonance imaging. It constructs a repetitive transcranial magnetic stimulation (RTMS) targeting the bilateral superior temporal gyrus, left inferior frontal gyrus, right insula, left cingulate gyrus, and bilateral anterior cerebellar lobes. It also uses a dynamic causal model and a multilayer perceptron regression model to map real-time EEG features to the expected activation level of the RTMS. The intervention effect evaluation subsystem calculates the comprehensive intervention effect index (CIEI) based on the change rate of the theta band power of EEG, the change in the total proportion of cognitive behavioral exploration and emotional exploration, and the decrease rate of negative semantic feature weights; the central control and feedback regulation subsystem runs a hierarchical reinforcement learning controller to dynamically generate the optimal stimulus sequence based on the CIEI, RTMS predicted activation level, and current stimulus parameters. The behavioral response recognition subsystem classifies and identifies five categories of behavioral response labels: descriptive statements, emotional exploration, cognitive behavioral exploration, resistance behavior, and therapeutic change.

2. The emotion intervention device based on nostalgia therapy according to claim 1, characterized in that, The individualized life event image set is stored in a redundant solid-state storage array and encoded using a lossless compression format. The emotional valence level is determined by taking the average value after independent evaluation based on a standardized scale. The evaluation results must meet the consistency standard. The image carousel duration is set to a default value within an adjustable range, and blank screens are inserted before and after as the baseline period.

3. The emotion intervention device based on nostalgia therapy according to claim 1, characterized in that, The scalp EEG acquisition unit uses a multi-lead wet electrode cap, the electrode material is silver-plated silver chloride, the impedance is controlled within the low impedance range, the sampling frequency is high frequency, the bandpass filter covers the low to high frequency range, the reference electrode is the average reference of the bilateral mastoid process line, the ground wire is placed in the forehead, the data is transmitted through a wireless protocol, the end-to-end latency is low, and physiological artifacts are removed by an independent component analysis algorithm.

4. The emotion intervention device based on nostalgia therapy according to claim 1, characterized in that, The heart rate variability monitoring unit uses a chest-strap photoelectric sensor to acquire pulse wave signals with a sampling rate of medium to high frequency. The RR interval sequence undergoes abnormal interval detection, interpolation correction, and detrending processing according to a standard procedure. The ratio of low-frequency power to high-frequency power is used as an autonomic nervous system balance indicator to participate in regulatory decisions.

5. The emotion intervention device based on nostalgia therapy according to claim 1, characterized in that, The skin conductance response detection unit has two disc electrodes fixed between the fingers, a DC bias voltage applied, and a sampling rate of intermediate frequency. After the raw signal is low-pass filtered, the skin conductance level and skin conductance response are separated. A skin conductance response peak amplitude exceeding the threshold and rise time shorter than the set value are defined as a valid emotional arousal event.

6. The emotion intervention device based on nostalgia therapy according to claim 1, characterized in that, The audio acquisition module is equipped with hyperdirectional microphones, which are installed on both sides above the display unit. The pickup angle is controlled in a conical area. After noise reduction processing, the speech signal is input to a neural network-based speech recognition engine. The transcribed text is sent to a semantic model to generate semantic vector embeddings for behavior classification.

7. The emotion intervention device based on nostalgia therapy according to claim 1, characterized in that, The video behavior encoding module uses an infrared camera mounted at an appropriate height and distance directly in front of the camera, and is equipped with a near-infrared fill light to achieve clear imaging in low-light environments. The video stream is input to a behavior feature extraction server equipped with a graphics processor, and a behavior classification model with a convolutional neural network as the backbone network is run. The input features include audio spectrograms, facial action unit intensity, head pose angle change rate, and semantic vectors, and the output is a probability distribution of multiple behavior types.

8. The emotion intervention device based on nostalgia therapy according to claim 7, characterized in that, The rule for determining the occurrence of a behavior is that if the probability of a certain category continuously exceeds a threshold for a preset time, the behavior event is recorded, along with its duration and intensity level. The intensity level is divided into multiple levels. The model training data comes from independently encoded and consistent real intervention videos. The neural response modeling subsystem acquires structural MRI images of the subject in the individualized calibration procedure, performs bias field correction, tissue segmentation and nonlinear registration to the standard space, and back-projects the target area RTMS mask to the individual anatomical structure for subsequent brain power localization analysis.

9. The emotion intervention device based on nostalgia therapy according to claim 8, characterized in that, The input features of the multilayer perceptron regression model include event-related potential amplitude, component amplitude and frequency band power. After standardization, the input is a network with multiple fully connected hidden layers. The activation function is a non-linear function. The output is the predicted value of the RTMS blood oxygen level dependent signal change rate. The model is deployed in the acceleration engine. The central control and feedback regulation subsystem operates a closed-loop regulation algorithm, which is essentially a hierarchical reinforcement learning controller. The upper-level policy network adopts a reinforcement learning algorithm. The state space includes CIEI value, RTMS predicted activation mean, current image emotional valence mean, number of presented images and life stage position. The action space includes maintaining the current stage, switching to a higher positive valence subset, introducing prosocial theme images, inserting cognitive re-evaluation guidance statements, or terminating the conversation. The central control and feedback regulation subsystem also has a built-in lower-level rule engine containing multiple deterministic regulation rules, which have a higher priority than the upper-level strategy output.

10. An emotion intervention system based on nostalgia therapy, characterized in that, include: One or more processors; A memory storing one or more programs, which, when executed by one or more processors, cause the one or more processors to connect to and perform corresponding operations with the stimulus presentation subsystem, physiological signal acquisition subsystem, behavioral response recognition subsystem, neural response modeling subsystem, intervention effect evaluation subsystem, and central control and feedback regulation subsystem of the nostalgia-based emotional intervention device according to any one of claims 1 to 9.

Images