VR grading exposure treatment system for anxiety disorder and trauma and stress related disorder
The VR graded exposure therapy system, which integrates wearable skin conductance sensors, AI edge computing nodes, and graded exposure scenario management modules, solves the problems of physiological feedback delay and rigid exposure process in virtual reality exposure systems. It enables personalized and dynamic treatment of anxiety disorders, trauma and stress-related disorders, and improves the safety and effectiveness of treatment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-07
- Publication Date
- 2026-03-13
AI Technical Summary
Existing virtual reality exposure systems suffer from problems such as delayed physiological feedback, rigid exposure processes, and lack of psychological intervention during the treatment of anxiety disorders, trauma, and stress-related disorders, leading to inaccurate treatment, discontinuous experience, and high patient dropout rates.
By employing a wearable skin conductance sensor module, an AI edge computing node, a graded exposure scene management module, a scene parameter adjuster, and a virtual reality head-mounted display rendering unit, a VR graded exposure treatment system is constructed that integrates real-time physiological monitoring, personalized graded exposure management, and embedded adaptive training. Through millisecond-level visual smooth transition and personalized graded exposure scene management, a safe and effective treatment closed loop is achieved.
Significantly improves treatment safety and intervention timeliness, provides personalized graded exposure and dynamic adaptation, constructs a highly reliable and low-latency hardware closed loop, and enhances the standardization of treatment and patients' self-management capabilities.
Smart Images
Figure CN121648422A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the interdisciplinary fields of medicine, virtual reality and artificial intelligence, and in particular to a VR graded exposure treatment system for anxiety disorders and trauma and stress-related disorders. Background Technology
[0002] Virtual Reality (VR) exposure therapy, as an important technical approach for psychological intervention in Post-traumatic Stress Disorder (PTSD), has gained widespread attention and application in the field of clinical psychiatry in recent years. This therapy guides patients to gradually access and reconstruct stimuli associated with their traumatic memories by constructing a controllable and repeatable immersive virtual environment, thereby achieving emotional desensitization and cognitive remodeling. Compared to traditional face-to-face exposure therapy, virtual reality technology significantly improves the standardization, contextual reconstruction capabilities, and engagement of treatment, especially suitable for scenarios that are difficult to safely reproduce or precisely control in reality (such as battlefields, disaster sites, specific social situations, or school environments). However, despite its clear theoretical advantages, existing virtual reality exposure systems still face severe technical and clinical logic bottlenecks in actual clinical deployment, especially in the treatment of dynamic and personalized adaptive disorders, seriously limiting the accuracy, safety, and long-term patient compliance of treatment.
[0003] Current mainstream virtual reality exposure systems generally adopt an architecture based on a pre-set static scene library. Their core logic relies on therapists manually setting exposure sequences and switching timings. While this framework provides a structured treatment process, it generally lacks two key capabilities: first, the ability to perceive and intelligently respond to the patient's real-time physiological state at the millisecond level; and second, a standardized, tiered exposure and embedded psychological intervention process deeply integrated with cognitive behavioral therapy. When patients are exposed to highly arousing stimuli (such as explosions, crowds, or specific social interactions), their skin conductivity often spikes dramatically within hundreds of milliseconds, creating a momentary stress peak. Because the systems in these technologies lack a real-time, adaptive closed-loop control mechanism for physiological signals and scene parameters, such strong stimuli often persist for too long, easily triggering panic-induced interruptions. A deeper problem lies in the fact that the regulatory logic of these systems (such as delayed switching based on heart rate) is not only slow to respond but also abrupt in its adjustment methods (such as direct scene jumps), leading to visual abrupt changes that may cause secondary psychological discomfort and disrupt the immersion and continuity of treatment.
[0004] Furthermore, even though some cutting-edge research attempts to incorporate artificial intelligence for prediction or dynamic adjustment, their solutions still have significant limitations. On the one hand, most systems focus on a single, static "safety threshold" determination, ignoring the central role of anxiety hierarchy theory in treatment. They cannot dynamically manage the stepped exposure sequence from low to high anxiety based on individual differences and treatment progress. For example, in treating school phobia, they cannot intelligently adapt and navigate between levels such as "exterior view of the school gate," "empty classroom," and "occupied classroom." On the other hand, when patients experience strong stress reactions, related technologies generally can only perform passive operations such as "stopping exposure" or "switching to a safe scene," lacking the ability to actively embed empirical psychological adjustment tools such as mindfulness breathing and relaxation training into the treatment process, thus missing the valuable treatment opportunity to transform "exposure interruption" into "emotion regulation skill learning."
[0005] In summary, the relevant technological systems suffer from a complex set of deficiencies when addressing generalized anxiety and adjustment disorders: a lack of real-time physiological feedback, a static exposure process, and a limited range of intervention methods. This makes it difficult for the system to achieve precise coupling with the patient's dynamic psychological state and to provide a complete "exposure-response-regulation" closed-loop experience that conforms to the principles of modern cognitive behavioral therapy. Therefore, constructing a novel virtual reality intelligent training system that integrates real-time physiological monitoring and prediction, personalized graded exposure management, and embedded adaptive psychological training has become a key challenge for improving clinical efficacy and accessibility in this field. Summary of the Invention
[0006] This application provides a VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders to address issues such as inaccurate treatment, discontinuous experience, and high patient dropout rates caused by delayed physiological feedback, rigid exposure procedures, and lack of psychological intervention during the process in related technologies. Based on AI (Artificial Intelligence) stress inflection point prediction, millisecond-level visual smooth transition, personalized graded exposure scene management, and embedded adaptive training protocols, it forms a safe, effective, and highly personalized treatment closed loop, improving the safety, effectiveness, and user tolerance of VR exposure therapy.
[0007] This application provides a VR graded exposure treatment system for anxiety disorders and trauma and stress-related disorders, including: a wearable skin conductance sensor module, an AI edge computing node, a graded exposure scene management module, a scene parameter adjuster, and a virtual reality head-mounted display rendering unit. The wearable skin conductance sensor module is attached to the patient's skin surface and is used to continuously collect skin conductance signals at a preset sampling frequency. The AI edge computing node is deployed locally in the treatment room and is communicatively connected to the wearable skin conductance sensor module. It is used to receive the skin conductance signal and process the skin conductance signal based on a lightweight temporal prediction neural network model to output a stress risk index. The graded exposure scenario management module stores a sequence of virtual reality exposure scenarios sorted by anxiety level; The scene parameter adjuster is communicatively connected to the AI edge computing node; The virtual reality headset rendering unit is connected to the scene parameter adjuster and is used to render and present the virtual reality scene; The AI edge computing node is configured to generate a first type of intervention instruction when the stress risk index reaches or exceeds the dynamic safety threshold, and the scene parameter regulator is configured to execute the first type of intervention instruction to perform multi-level progressive adjustment of the visual and auditory parameters of the current exposure scene according to multi-level progressive parameters. The system further includes an adaptive training trigger module, configured to trigger a second type of intervention instruction when the exposure process needs to be paused based on the stress risk index or preset rules. The scene parameter regulator is also configured to execute the second type of intervention instruction to control the virtual reality headset rendering unit to switch from the current exposure scene to a preset relaxation training scene and start a guided psychological adjustment program.
[0008] Optionally, in some embodiments, the wearable skin conductance sensor module includes: a flexible patch substrate, a pair of Ag / AgCl electrodes disposed on the flexible patch substrate, a conductive gel layer coated on the surface of the electrodes, and a dual-channel acquisition circuit integrated on a flexible printed circuit board. The dual-channel acquisition circuit includes a main channel and an auxiliary channel. The main channel is used to acquire the absolute value of the skin conductance signal, and the auxiliary channel is used to filter out electromyographic signal interference.
[0009] Optionally, in some embodiments, the AI edge computing node is equipped with an embedded microcontroller and runs a lightweight temporal prediction neural network model; The lightweight temporal prediction neural network model includes an LSTM (Long Short-Term Memory) layer and an Attention layer. The LSTM layer is used to extract the temporal features of the skin conductance signal, and the Attention layer is set with two attention heads to focus on the key time frames of the inflection point of the skin conductance signal. The stress risk index has an output range of 0 to 100 points.
[0010] Optionally, in some embodiments, the dynamic safety threshold is periodically or automatically calibrated based on the patient's personalized historical stress profile; The initial value of the dynamic security threshold is set to 85 points, and the calibration algorithm uses the exponentially weighted moving average method, calculated as follows: ; in, The threshold is the previous day's threshold. This represents the current baseline mean of skin conductance. The standard deviation of skin conductance during the three most recent effective exposure phases.
[0011] Optionally, in some embodiments, the multi-level gradient parameters include: a first-level gradient parameter, a second-level gradient parameter, and a third-level gradient parameter, wherein, The first-level gradient parameters are used to reduce the intensity of visual stimulation in the current virtual reality scene and introduce soothing auditory elements. The second-level gradient parameters are used to generate visual transition content from the current virtual reality scene to the target scene and reduce the complexity of scene rendering. The third-level gradient parameter is used to switch to the pre-stored safe scene and maintain the overall stimulation level of the scene in a stable state.
[0012] Optionally, in some embodiments, the exposure scenario management module for treating school adjustment disorder includes the following levels from low anxiety to high anxiety: an empty school exterior, an empty classroom corridor, a classroom with static virtual characters, and a classroom scene with mild interactive elements.
[0013] Optionally, in some embodiments, the guided psychological adjustment procedure includes one or more of the following: guided deep breathing with visual breathing rhythm indicators, mindfulness meditation audio training, and encouraging verbal and visual feedback provided after the training.
[0014] Optionally, in some embodiments, the scene parameter regulator is integrated into the virtual reality headset driver layer in the form of a field-programmable gate array (FPGA) logic unit; the AI edge computing node is directly connected to the scene parameter regulator via a PCIe 4.0 interface.
[0015] Optionally, in some embodiments, the virtual reality headset rendering unit uses a binocular organic light-emitting diode display and runs a rendering engine based on the Unity high-definition rendering pipeline; The rendering engine is configured with a dual buffering mechanism, including a main buffer for displaying the current frame and a secondary buffer for preloading the content of the next frame. Buffer switching is triggered by a vertical synchronization signal. When the rendering unit performs the multi-level progressive adjustment, it follows a pixel-level gradient constraint algorithm to control the upper limit of the value changes of each color channel of the same pixel in adjacent rendering frames.
[0016] Optionally, in some embodiments, a hardware-level interrupt path is provided between the wearable skin conductance sensor module and the AI edge computing node; The wearable skin conductance sensor module is configured to send an interrupt signal to the AI edge computing node through the hardware-level interrupt path when the transient change characteristics of the skin conductance signal are detected in real time and meet the preset emergency conditions.
[0017] The beneficial effects of the embodiments of this application are as follows: (1) Significantly improve treatment safety and intervention timeliness: This application can trigger intervention before the patient’s subjective panic is formed by predicting the stress inflection point based on the LSTM-Attention model, effectively avoiding treatment interruption caused by stimulus overload, and providing reliable safety guarantee for exposure therapy.
[0018] (2) Achieving a smooth transition that is "unobtrusive" to human eye perception: This application adopts a multi-level gradient protocol and pixel-level change rate control to make scene switching visually continuous and smooth, avoiding secondary discomfort that may be caused by sudden changes in the screen, and ensuring the immersion and comfort of the treatment experience.
[0019] (3) Personalized graded exposure and dynamic adaptation: The system pre-sets graded exposure scene sequences based on the anxiety level theory (such as from school exterior to classroom interaction), and can dynamically adjust the exposure intensity and safety threshold according to the patient's real-time physiological feedback and historical data to achieve precise dose control of "one person, one policy".
[0020] (4) A highly reliable and low-latency hardware closed loop has been constructed: This application uses the architecture of "hardware interrupt triggering - edge computing node prediction - FPGA (Field-Programmable Gate Array) regulator execution" and PCIe (Peripheral Component Interconnect Express) direct connection and double buffer rendering mechanism to ensure that the end-to-end latency from physiological signal acquisition to scene rendering adjustment is less than 1 millisecond, thus ensuring the real-time and deterministic nature of the system response.
[0021] (5) Provides a structured progressive exposure therapy path: The built-in graded exposure scenario management module provides a standardized progressive therapy path from low anxiety to high anxiety for specific disorders (such as school adjustment disorder), which reduces the design burden of therapists and improves the operability and standardization of treatment.
[0022] (6) Transforming exposure interruption into a skill learning opportunity: When the system determines that exposure needs to be paused, it can automatically start embedded psychological adjustment programs (such as guided deep breathing and mindfulness meditation) to transform possible treatment setbacks into an emotional regulation skill training, thereby enhancing the patient's self-management ability and treatment confidence.
[0023] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0024] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of the structure of a VR graded exposure therapy system for anxiety disorders, trauma and stress-related disorders provided in the embodiments of this application. Detailed Implementation
[0025] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.
[0026] The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders, according to embodiments of this application, is described below with reference to the accompanying drawings. Specifically, Figure 1 This is a schematic diagram of the structure of the VR graded exposure therapy system for anxiety disorders, trauma and stress-related disorders provided in the embodiments of this application.
[0027] like Figure 1 As shown, the VR graded exposure therapy system 10 for anxiety disorders and trauma and stress-related disorders (hereinafter referred to as "System 10") includes: a wearable skin conductance sensor module 100, an AI edge computing node 200, a graded exposure scene management module 300, an adaptive training trigger module 400, a scene parameter adjuster 500, and a virtual reality head-mounted display rendering unit 600.
[0028] Specifically, the closed-loop workflow of system 10 is as follows: the wearable skin conductance sensor module 100 collects skin conductance signals in real time, which are transmitted to the AI edge computing node 200 via a low-latency communication link. The lightweight LSTM-Attention model built into the AI edge computing node 200 processes the signals in real time, calculates the stress risk index, and makes decisions in combination with the scene sequence information pre-stored in the graded exposure scene management module 300.
[0029] System 10's intervention includes parallel two-path logic: One approach is real-time adjustment: If the stress risk index exceeds the dynamic safety threshold, the AI edge computing node 200 generates a first-type intervention instruction (i.e., a scene adjustment instruction containing multi-level gradual parameters), which is sent to the scene parameter regulator 500 via a high-speed interface (such as PCIe 4.0). The scene parameter regulator 500 parses and executes the instruction, driving the virtual reality headset rendering unit 600 to complete a smooth gradual transition of the currently exposed scene within milliseconds.
[0030] The second is the adaptive training path: If the adaptive training trigger module 400 determines that exposure needs to be suspended based on the stress risk index or preset rules (such as therapist instructions), it generates a second type of intervention instruction and sends it to the scene parameter regulator 500. The scene parameter regulator 500 then controls the virtual reality headset rendering unit 600 to switch from the current exposure scene to a preset relaxation training scene and initiates a guided psychological adjustment program (such as mindfulness breathing training).
[0031] After any intervention is completed, system 10 enters the closed-loop verification phase: the wearable skin conductance sensor module 100 continuously monitors physiological signals within subsequent time windows, comparing the monitoring results with the patient's individual baseline to objectively assess the effectiveness of the intervention. The assessment results will be used to update the patient's personalized treatment model. If several pre-ineffective or stress-induced intensification occurs, a higher-level safety plan will be triggered, thus forming an intelligent treatment closed loop integrating monitoring, prediction, personalized intervention, and effect verification.
[0032] The above components form a millisecond-level closed-loop feedback system through a low-latency communication link and coordinated control logic, namely system 10 in this embodiment. The specific implementation of this application will be described in detail below with reference to the accompanying drawings.
[0033] The wearable skin conductance sensor module 100 is attached to the patient's skin surface to continuously collect skin conductance signals at a preset sampling frequency.
[0034] Optionally, in some embodiments, the wearable skin conductance sensor module 100 includes: a flexible patch substrate, a pair of Ag / AgCl electrodes disposed on the flexible patch substrate, a conductive gel layer coated on the electrode surface, and a dual-channel acquisition circuit integrated on a flexible printed circuit board; the dual-channel acquisition circuit includes: a main channel and an auxiliary channel, the main channel being used to acquire the absolute value of the skin conductance signal, and the auxiliary channel being used to filter out electromyographic signal interference.
[0035] In this embodiment, the wearable skin electrical response sensor module 100 adopts a flexible medical patch structure with an overall thickness of no more than 1.2 mm. It is attached to the skin area between the flexor carpi ulnaris and palmaris longus muscles on the inner side of the patient's left wrist to ensure that the electrode contact surface avoids the main muscle group activity area and reduces motion artifact interference.
[0036] The wearable skin conductance sensor module 100 of this application embodiment has a pair of Ag / AgCl electrodes with a diameter of 8 mm. The electrodes are coated with a conductive gel layer. The gel composition includes sodium chloride, glycerin and polyvinyl alcohol, which has high ionic conductivity and low skin irritation.
[0037] The dual-channel acquisition circuit of this embodiment is integrated on a flexible printed circuit board. The main channel is equipped with a constant current source excitation circuit, with a constant output current of 0.5μA and a sampling frequency set to 1000Hz, used to continuously acquire the absolute value of skin conductance signals. The auxiliary channel is connected in series with a 10 to 100Hz bandpass filter. This filter adopts a second-order Butterworth topology with a cutoff slope of -40dB per decade, effectively suppressing 50Hz power frequency interference and high-frequency electromyographic noise. After analog-to-digital conversion, the raw signal is transmitted to the AI edge computing node 200 in broadcast mode via Bluetooth 5.0 Low Energy protocol. The measured end-to-end transmission delay at a distance of 3 meters without obstacles is 1.6ms with a standard deviation of 0.12ms.
[0038] AI edge computing node 200 is deployed locally in the treatment room and communicates with wearable skin conductance sensor module 100 to receive skin conductance signals and process the skin conductance signals based on a lightweight temporal prediction neural network model to output a stress risk index.
[0039] AI edge computing node 200 is configured to generate a first-class intervention instruction when the stress risk index reaches or exceeds the dynamic safety threshold.
[0040] Edge computing node 200 refers to a dedicated computing device deployed locally in the treatment room, close to the data source (patient). Its core feature is localized processing to avoid network latency caused by uploading data to the cloud or remote server. In this embodiment, the AI edge computing node 200 receives physiological signals from the wearable skin conductance sensor module 100, runs a lightweight artificial intelligence model, and can output a stress risk index in a very short time (milliseconds).
[0041] In this embodiment, the lightweight temporal prediction neural network model is deployed on the edge computing node 200. It can analyze skin conductance temporal signals in real time with limited storage and computing resources (such as embedded microcontrollers) and output a characterization of the patient's instantaneous stress risk level.
[0042] The first type of intervention instruction in this application embodiment is generated by the AI edge computing node 200 after internal calculation. When the stress risk index calculated in real-time by the AI edge computing node 200 reaches or exceeds the dynamic safety threshold set for the current patient, the generation of this instruction is immediately triggered. Its core design goal is to achieve a smooth and imperceptible decay of stimulus intensity to avoid panic, interruption, or secondary psychological trauma to the patient caused by abrupt changes in the virtual reality scene content. In other words, the core logic of the first type of intervention instruction is "dynamically adjusting the stimulus intensity during the exposure process." It does not interrupt the overall process of exposure therapy but optimizes the experience within the process through technical means, belonging to "technical safety control."
[0043] Optionally, the edge computing node 200 in this embodiment is deployed in a dedicated cabinet in the treatment room, with its physical location no more than 3m from the center of the patient's seat to minimize wireless signal propagation delay. The edge computing node 200 is equipped with a customized embedded motherboard, and the main control chip is an ARM Cortex-M7 architecture microcontroller with a clock frequency of 480 MHz. It integrates 512 kilobytes of SRAM and 2 megabytes of flash memory on-chip, running a lightweight temporal prediction neural network model built based on the TensorFlow Lite Micro framework. The total number of parameters in this lightweight temporal prediction neural network model is 48,768 bytes, satisfying local storage and real-time inference constraints. The model input is GSR (Galvanic Skin Response) time-series data within a 200ms sliding window, containing 200 sampling points, which are normalized and then fed into the LSTM layer.
[0044] Optionally, in some embodiments, the AI edge computing node 200 is equipped with an embedded microcontroller and runs a lightweight temporal prediction neural network model; the lightweight temporal prediction neural network model includes an LSTM layer and an Attention layer, the LSTM layer is used to extract the temporal features of the skin conductance signal, and the Attention layer sets two attention heads to focus on the key time frames of the inflection point of the skin conductance signal.
[0045] In this embodiment of the application, the LSTM layer (Long Short-Term Memory layer) is a lightweight temporal prediction neural network model that is specifically responsible for extracting and memorizing temporal dynamic features related to stress response from the historical sequence of skin conductance signals, such as the rise slope and fluctuation pattern of the signal.
[0046] In this embodiment of the application, the Attention layer is located after the LSTM layer in a lightweight temporal prediction neural network model, and is used to weight the temporal features extracted by the LSTM. Its two attention heads work together to specifically identify and focus on key time frames about 100ms before the skin conductance signal undergoes a sharp change (i.e., the "inflection point"), thereby improving the accuracy of inflection point prediction.
[0047] In this embodiment, the LSTM layer is configured with 24 hidden units and uses the tanh activation function. Its internal state update formula is as follows: in, For the first Input at any time, The state was hidden in the previous moment. In cellular state, For the sigmoid function, This represents element-wise multiplication. The 24-dimensional feature vector output from the LSTM layer is fed into the Attention layer, which has two attention heads. Each head independently computes the query, key, and value vectors, each with a dimension of 12. The attention weights are calculated using a scaled dot product mechanism.
[0048] in, The dimension of the key vector.
[0049] This application's embodiments utilize an attention mechanism to focus the model on the critical time frame within 100ms before the GSR signal inflection point, enhancing its sensitivity to sudden changes in the rising slope. The final output is a stress risk index ranging from 0 to 100, with higher values indicating a greater probability that the patient is about to enter a high-stress state. In actual execution, after a GSR surge event occurs, the model can complete the entire inference process within 280μs.
[0050] Optionally, in some embodiments, the stress risk index output ranges from 0 to 100 points.
[0051] In this embodiment, a higher stress risk index indicates a greater risk of the model predicting a patient experiencing a strong stress response (panic, anxiety, etc.). This index is the core basis for system 10 to determine whether intervention needs to be triggered. Specifically, it is used in two parallel judgment processes: first, it is compared with a dynamic safety threshold to determine whether to generate a first type of intervention instruction (for scene gradual adjustment); second, it is sent in real time to the adaptive training triggering module 400 as a key input parameter for the module to determine whether to pause the current exposure and trigger a second type of intervention instruction (for initiating relaxation training).
[0052] Optionally, in some embodiments, the dynamic safety threshold is periodically or automatically calibrated based on the patient's personalized historical stress profile; The initial value of the dynamic safety threshold is set to 85 points. The calibration algorithm uses the exponentially weighted moving average method, and the calculation formula is as follows: ; in, The threshold is the previous day's threshold. The mean baseline skin conductance (the mean baseline skin conductance of the patient at rest before treatment). The standard deviation of skin conductance during the three most recent effective exposure phases.
[0053] In this embodiment, the initial value of the dynamic safety threshold is set to 85 points, but it will be periodically or automatically calibrated after each treatment based on the patient's personalized historical stress profile. For example, it can be set to be calibrated once a day. When the "stress risk index is greater than or equal to the dynamic safety threshold", the system 10 determines that immediate intervention is required, thereby achieving personalized and adaptive treatment.
[0054] The personalized historical stress map in this application embodiment is a database or data structure that records the patient's baseline skin conductance during each treatment. ), standard deviation of signal fluctuation ( This includes dynamic safety thresholds, intervention effects, etc. Personalized historical stress maps serve as the data foundation for System 10 to perform dynamic safety threshold calibration and optimize future intervention strategies.
[0055] The graded exposure scenario management module 300 stores a sequence of virtual reality exposure scenarios sorted by anxiety level.
[0056] The graded exposure scenario management module 300 in this embodiment is the core content engine for the system 10 to realize standardized and personalized treatment plans. This module is an intelligent management system built based on the "exposure hierarchy" theory in cognitive behavioral therapy.
[0057] All virtual reality exposure scenarios in this module are pre-assessed and graded based on their potential to induce subjective anxiety in patients. The grading criteria typically reference the Subjective Units of Distress Scale (SUDS), usually ranging from 0 to 100 points. System 10 maps SUDS score ranges to different scenario levels, thereby constructing a progressive "exposure ladder" from low-anxiety stimuli to high-anxiety stimuli.
[0058] The graded exposure scenario management module 300 exists in the form of a database or configuration file. Each record is associated with a virtual scenario resource and the following metadata: a unique scenario identifier, the theme or type of obstacle (e.g., school phobia, social anxiety, traffic phobia, etc.), a preset anxiety level (SUDS range), a description of the scenario content and key stimulus cues, and the associated safe scenario or relaxation training scenario ID.
[0059] Optionally, in some embodiments, the graded exposure scenario management module 300 includes a sequence of exposure scenarios for treating school adjustment disorder, ranging from low anxiety to high anxiety, such as: an empty school exterior, an empty classroom corridor, a classroom with static virtual characters, and a classroom scene with mild interactive elements.
[0060] This application uses the treatment of "school adjustment disorder" as an example to demonstrate a typical sequence of graded exposure scenarios, which is designed to follow a progressive principle from "non-threatening environments" to "simulated real social stress": Level 1 (SUDS 10-30): An unoccupied school exterior. This scene presents a static, sunny 3D exterior view of a school building on a weekend or in the evening, without any people, vehicles, or sounds of activity. In this scene, the patient establishes initial adaptation by visually encountering the concept of "school" at a safe distance.
[0061] Level 2 (SUDS 31-50): An empty classroom corridor. In this scenario, the patient is in an empty school building corridor with soft lighting and can hear faint sounds of bells or ventilation systems in the distance. Entering the school's interior space, the patient adapts to the relatively enclosed indoor environment and is introduced with mild ambient sound stimulation.
[0062] Level 3 (SUDS 51-70): A classroom containing static virtual characters. In this scenario, the patient sits in the back row of a classroom with 5-8 static virtual student models sitting with their backs to the patient or to the side. The models have neutral expressions and do not make any movements. This scenario introduces the social pressure element of "the presence of others," but controls it to a safe level of static and non-interactive elements.
[0063] Level 4 (SUDS 71-85): A classroom scene with light interactive elements. Building upon the Level 3 scene, the virtual student model adds occasional slight movements (such as turning their head or turning pages), and the teacher model may stand at the podium and once glance across the patient's area without a specific target. This scene increases the dynamism and unpredictability of the environment, simulates the feeling of being noticed, and further increases the anxiety challenge level.
[0064] Level 5 (SUDS 86-100): Simulates a classroom scenario of roll call or simple question-and-answer. In this scenario, the teacher model turns to the patient to call roll or ask a simple question with a fixed answer (such as "What day is it today?"), and waits for the patient to respond (via microphone or controller). The virtual classmate model may provide simple feedback actions (such as nodding). This scenario simulates realistic, low-risk social interaction, preparing the patient for simple social conversations in real life.
[0065] The graded exposure scenario management module 300 in this embodiment creates a treatment plan for each patient. Essentially, it selects and customizes a personalized exposure path that matches the patient's current condition from the module's "general library." When the AI edge computing node 200 performs stress risk assessment, it can call upon the level information of the current exposure scenario. For example, in a Level 5 (high anxiety level) scenario, the system 10 may use a more sensitive intervention threshold to achieve scenario-based adaptation of risk control. When a patient triggers adaptive training (type II intervention) due to excessive stress and recovers, the system 10 can combine the level information from the graded exposure scenario management module 300 to intelligently recommend, or the therapist can decide, whether to revert to the previous level for consolidation or repeat the current level for another attempt, thereby scientifically managing the treatment process.
[0066] The adaptive training trigger module 400 is configured to trigger a second type of intervention instruction when the exposure process needs to be paused based on the stress risk index or preset rules. The scene parameter regulator is also configured to execute the second type of intervention instruction to control the virtual reality headset rendering unit 600 to switch from the current exposure scene to a preset relaxation training scene and start a guided psychological adjustment program.
[0067] The adaptive training trigger module 400 in this embodiment of the application acts as a "scheduler" for the treatment logic of the system 10. It is responsible for assessing the appropriateness of the exposure process in real time and initiating a complete psychological skills training process when it is determined that the patient needs to temporarily detach from the current stimulus to regulate emotions. Its fundamental design philosophy is to transform a possible "exposure failure" (forced to stop due to excessive stress) into a successful opportunity to "learn emotions regulation skills".
[0068] The adaptive training triggering module 400 makes decisions based on multi-source information, and its triggering conditions are configured to include, but are not limited to, the following situations (any one of them needs to be met): Condition A (based on AI prediction): Continuously receive real-time stress risk index from AI edge computing node 200. When this index remains above a set pause threshold for a short period (e.g., within 2 seconds), the trigger module determines that the exposure intensity may have exceeded the critical point of the patient's current emotional regulation ability. The pause threshold can be set as a percentage of the dynamic safety threshold (e.g., 110%) or as an independent fixed value.
[0069] Condition B (based on preset rules): First, a timed rule, that is, when the duration of a single continuous exposure reaches the preset safety limit (such as 90-120 seconds), a short rest and adjustment is forcibly initiated regardless of physiological indicators to prevent fatigue accumulation; Second, an external instruction rule, which is to receive a clear "pause" instruction from the therapist's control terminal, or to actively pause the request issued by the patient through the physical "safety key" on the handheld controller.
[0070] Condition C (based on system state): When system 10 determines that the intervention is ineffective (i.e., physiological signals have not recovered) after executing the "first type of intervention instruction" (scenario change), the adaptive training trigger module 400 can be activated as an upgrade plan.
[0071] Once the determination is successful, the adaptive training trigger module 400 immediately generates a second type of intervention instruction. This instruction is a structured command package, the core contents of which include at least: instruction type identifier (clearly "adaptive training"), target relaxation scene ID (specifying the preset relaxation training scene to switch to), and training program identifier (specifying the type and parameters of the guided psychological adjustment program to be activated). This instruction is immediately sent to the scene parameter regulator 500 via the internal communication bus of system 10.
[0072] Optionally, in some embodiments, the guided mental conditioning procedure includes one or more of the following: guided deep breathing with visual breathing rhythm indicators, mindfulness meditation audio training, and encouraging verbal and visual feedback provided after the training.
[0073] Upon receiving a second type of intervention instruction, the scene parameter adjuster 500 interrupts any ongoing gradation process (if any) with the highest priority, and coordinates with the virtual reality headset rendering unit 600 to perform the following inseparable and coherent operations: Operation 1: Smooth scene exit: Control the currently exposed scene to fade out within 300-500 milliseconds (e.g., gradually dimming the brightness to a black screen).
[0074] Step Two: Relaxation Scene Presentation: Seamlessly switch to the preset relaxation training scene. This scene is designed to minimize cognitive and emotional load, such as: a 360-degree tranquil starry sky (with slowly moving nebulae), a quiet forest corner where sunlight filters through the leaves, or a space with a uniform and soft color scheme. There are no dynamic narrative elements in the scene that could trigger associations.
[0075] Operation 3: Activating the Guided Psychological Adjustment Program: Almost simultaneously with the scene transition, System 10 activates the pre-stored guided psychological adjustment program. This program is a digital integration of various evidence-based psychological intervention methods. Specifically, it includes guided visual deep breathing, where a dynamic, soothing visual element appears in the scene (such as a slowly expanding and contracting halo, a flower opening and closing in sync), while simultaneously playing a matching rhythmic guided audio: "Please follow the rhythm of the halo, inhale deeply... hold your breath... exhale slowly and evenly..." The strict synchronization of vision and hearing greatly reduces the cognitive threshold for patients to perform relaxation exercises; it also includes mindfulness meditation audio training, providing standardized mindfulness meditation audio guidance on various themes, such as " The program includes features such as "body scan relaxation," "observing thoughts drifting like clouds," and "safe imagery construction." The audio is recorded by professional therapists with a steady tone and background soothing white noise or natural sound effects. It also includes a positive feedback mechanism that automatically triggers encouraging feedback after a training session (e.g., 3 minutes) or when the system detects through sensors that the patient's physiological indicators (e.g., GSR) have significantly leveled off. For example, it presents a gentle visual message saying "Well done" and plays affirmative voice messages such as "You have successfully calmed yourself down, which is an important step forward" to enhance the patient's self-efficacy.
[0076] Optionally, during and after the guided psychological adjustment procedure, the system 10 enters a special "standby and assessment" state, continuously monitoring physiological signals to confirm the relaxation effect. Through gentle voice prompts (such as "When you feel completely calm and ready to continue, please nod gently / press the confirmation button"), the control and choice of whether and how to continue treatment are returned to the patient and therapist. Based on the data from this pause and training, the therapist can discuss with the patient and decide on subsequent steps through the system 10 control terminal: return to the previous lower-level exposure scenario, repeat the current scenario, or continue after adjusting parameters.
[0077] Scene parameter adjuster 500, which is communicatively connected to AI edge computing node 200, is configured to execute first-type intervention instructions to perform multi-level progressive adjustment of visual and auditory parameters of the current exposed scene according to multi-level progressive parameters.
[0078] Optionally, in some embodiments, the multi-level gradient parameters include: a first-level gradient parameter, a second-level gradient parameter, and a third-level gradient parameter, wherein the first-level gradient parameter is used to reduce the visual stimulation intensity of the current virtual reality scene and introduce soothing auditory elements; the second-level gradient parameter is used to generate visual transition content from the current virtual reality scene to the target scene and reduce the complexity of scene rendering; and the third-level gradient parameter is used to complete the switch to a pre-stored safe scene and maintain the overall stimulation level of the scene in a stable state.
Claims
1. A VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders, characterized in that, include: The device comprises a wearable skin conductance sensor module, an AI edge computing node, a graded exposure scene management module, a scene parameter adjuster, and a virtual reality headset rendering unit. The wearable skin conductance sensor module is attached to the patient's skin surface and is used to continuously collect skin conductance signals at a preset sampling frequency. The AI edge computing node is deployed locally in the treatment room and is communicatively connected to the wearable skin conductance sensor module. It is used to receive the skin conductance signal and process the skin conductance signal based on a lightweight temporal prediction neural network model to output a stress risk index. The graded exposure scenario management module stores a sequence of virtual reality exposure scenarios sorted by anxiety level; The scene parameter adjuster is communicatively connected to the AI edge computing node; The virtual reality headset rendering unit is connected to the scene parameter adjuster and is used to render and present the virtual reality scene; The AI edge computing node is configured to generate a first type of intervention instruction when the stress risk index reaches or exceeds the dynamic safety threshold, and the scene parameter regulator is configured to execute the first type of intervention instruction to perform multi-level progressive adjustment of the visual and auditory parameters of the current exposure scene according to multi-level progressive parameters. The system further includes an adaptive training trigger module, configured to trigger a second type of intervention instruction when the exposure process needs to be paused based on the stress risk index or preset rules. The scene parameter regulator is also configured to execute the second type of intervention instruction to control the virtual reality headset rendering unit to switch from the current exposure scene to a preset relaxation training scene and start a guided psychological adjustment program.
2. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, The wearable skin electroreactivity sensor module includes: a flexible patch substrate, a pair of Ag / AgCl electrodes disposed on the flexible patch substrate, a conductive gel layer coated on the surface of the electrodes, and a dual-channel acquisition circuit integrated on a flexible printed circuit board. The dual-channel acquisition circuit includes a main channel and an auxiliary channel. The main channel is used to acquire the absolute value of the skin conductance signal, and the auxiliary channel is used to filter out electromyographic signal interference.
3. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, The AI edge computing node is equipped with an embedded microcontroller and runs a lightweight time-series predictive neural network model; The lightweight temporal prediction neural network model includes an LSTM layer and an Attention layer. The LSTM layer is used to extract the temporal features of the skin conductance signal, and the Attention layer is set with two attention heads to focus on the key time frames of the inflection point of the skin conductance signal. The stress risk index has an output range of 0 to 100 points.
4. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, The dynamic safety threshold is periodically or automatically calibrated after each treatment based on the patient's personalized historical stress profile. The initial value of the dynamic security threshold is set to 85 points, and the calibration algorithm uses the exponentially weighted moving average method, calculated as follows: ； in, The threshold is the previous day's threshold. This represents the current baseline mean of skin conductance. The standard deviation of skin conductance during the three most recent effective exposure phases.
5. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, The multi-level gradient parameters include: first-level gradient parameters, second-level gradient parameters, and third-level gradient parameters, wherein, The first-level gradient parameters are used to reduce the intensity of visual stimulation in the current virtual reality scene and introduce soothing auditory elements. The second-level gradient parameters are used to generate visual transition content from the current virtual reality scene to the target scene and reduce the complexity of scene rendering. The third-level gradient parameter is used to switch to the pre-stored safe scene and maintain the overall stimulation level of the scene in a stable state.
6. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, In the graded exposure scenario management module, the exposure scenario sequence for treating school adjustment disorder includes the following levels from low anxiety to high anxiety: an empty school exterior, an empty classroom corridor, a classroom with static virtual characters, and a classroom scene with mild interactive elements.
7. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, The guided psychological adjustment program includes one or more of the following: guided deep breathing with visual breathing rhythm indicators, mindfulness meditation audio training, and encouraging verbal and visual feedback provided after the training.
8. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, The scene parameter regulator is integrated into the virtual reality headset driver layer in the form of a field-programmable gate array (FPGA) logic unit; the AI edge computing node is directly connected to the scene parameter regulator via a PCIe 4.0 interface.
9. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, The virtual reality head-mounted rendering unit uses a binocular organic light-emitting diode display and runs a rendering engine based on the Unity high-definition rendering pipeline. The rendering engine is configured with a dual buffering mechanism, including a main buffer for displaying the current frame and a secondary buffer for preloading the content of the next frame. Buffer switching is triggered by a vertical synchronization signal. When the rendering unit performs the multi-level progressive adjustment, it follows a pixel-level gradient constraint algorithm to control the upper limit of the value changes of each color channel of the same pixel in adjacent rendering frames.
10. The VR graded exposure therapy system for anxiety disorders and trauma and stress-related disorders according to claim 1, characterized in that, A hardware-level interrupt path is provided between the wearable skin electroreactivity sensor module and the AI edge computing node; The wearable skin conductance sensor module is configured to send an interrupt signal to the AI edge computing node through the hardware-level interrupt path when the transient change characteristics of the skin conductance signal are detected in real time and meet the preset emergency conditions.