Visual perception decision closed-loop nerve regulation and control system

Abstract

The invention belongs to the technical field of nerve regulation and control, and discloses a visual perception decision closed-loop nerve regulation and control system, which comprises an individualized functional atlas construction and target spot positioning module for constructing an individualized functional atlas of a region based on a visual perception-decision task normal form, collecting a structure image after an MRI development marker is arranged, and positioning the structure image; positioning and registering the mounting position of the implantable interface base based on the functional map, the structural image and the MRI developing marker; the MRI compatible targeted light stimulation interface module comprises an implantable interface base, a quick-release grid plugboard and a stepping optical fiber pin suite, and the behavior-brain network feedback driven closed-loop regulation and control module is used for executing nerve stimulation, collecting behavior and brain network signals as feedback indexes, modeling stimulation parameters and the feedback indexes, and updating the stimulation parameters according to a modeling result. The system can realize high-precision, systematized and individualized nerve regulation and control.

Description

Technical Field

[0001] The present invention belongs to the field of neural regulation technology, and in particular relates to a visual perception decision-making closed-loop neural regulation system. Background Art

[0002] While existing neuromodulatory brain-computer interface systems have made some progress in target localization and stimulation execution, they still suffer from issues such as poor system integration, low target localization accuracy, and a lack of closed-loop feedback mechanisms, making them difficult to meet the demands of high-precision, systematic neuromodulation. The core of neuromodulatory brain-computer interfaces lies in achieving precise coupling between external stimulation and the brain's functional networks. This process not only relies on matching structural and functional maps but also requires behavioral paradigms with clear input-output relationships and quantifiable feedback indicators as the basis for optimizing control parameters.

[0003] However, most current systems still lack feedback mechanisms based on neural dynamics and behavioral performance under task states, resulting in the inability to quantify regulatory effects and the inability to iterate parameter updates, making the neural regulatory process prone to non-specific intervention, significantly affecting the regulatory effect.

[0004] At the localization level, traditional neuromodulation technologies often rely on standard anatomical atlases (such as the MNI) or resting-state fMRI for target inference (e.g., CN118903688A, CN119949772A). These methods struggle to reflect the neural activation characteristics of individuals during task states, limiting the accuracy of target localization. Furthermore, gadolinium-based markers (e.g., CN119113154B) are often used in traditional MRI-guided localization, but their complex preparation and potential biotoxicity limit their long-term, stable application in neuromodulation.

[0005] Regarding stimulation navigation, target positioning and coordinate conversion in current neuromodulation systems often rely on robotic guidance or camera-assisted registration schemes (e.g., patent application CN116650114A). However, these devices often contain metal components, which are prone to radiofrequency artifacts and electromagnetic interference, severely impacting image quality and positioning accuracy, and preventing them from operating stably in high-field MRI environments.

[0006] To address magnetic compatibility requirements, existing proposals (such as patent application CN221730602U) attempt to use propulsion mechanisms to achieve precise positioning of electrodes or optical fibers. However, these structures are typically large, reaching several centimeters in height, making them difficult to fit within the confined space of an MRI coil. If the MRI coil is too far from the cortex, imaging quality and control accuracy will be severely affected.

[0007] From a system integration perspective, some structures have attempted to introduce magnetically compatible materials such as nylon (PA) and polycarbonate (PC) to replace traditional metal components. However, these materials generally have low rigidity and limited processing precision. This is particularly true in propulsion structures such as micro-slides and screws, where mechanical deformation and wear are prone to occur. This leads to the gradual accumulation of hole tolerances over long-term use, ultimately making it difficult to meet the millimeter-level target navigation accuracy and repeated stimulation stability requirements, limiting their long-term application in high-precision neuromodulation experiments.

[0008] In terms of closed-loop feedback control, current closed-loop neuromodulation systems are primarily used for electrical stimulation therapy in pathological conditions, such as epileptic seizure suppression and Parkinson's tremor control (e.g., patent CN118059386A). These systems typically rely on local field potentials (LFPs) as feedback signals to achieve real-time closed-loop triggering based on neural events. While achieving initial success in disease control, their designs primarily focus on modulating local neural responses to lesion activity, making them difficult to apply to complex tasks requiring the regulation of cognitive functional networks.

[0009] Existing neuromodulation systems, such as CN119989823B, incorporate MRI structural image navigation and task feedback mechanisms, but their core stimulation method remains electrical stimulation. This current spreads widely, with a spatial range of only centimeters, making it difficult to achieve fine-grained local control. While EEG feedback assessment has been introduced, it records superficial electric field signals with a spatial resolution of only a few centimeters, making it incapable of accurately reflecting more detailed local neural dynamics.

[0010] However, the functional organization of the cerebral cortex is characterized by distinct millimeter-scale fine divisions, such as the functional domains of the visual cortex. Therefore, to meet the precise control requirements of perceptual tasks, the system must have millimeter-scale spatial resolution of stimulation and feedback to achieve selective control of specific functional domains.

[0011] Furthermore, existing devices are often limited to a single function and cannot achieve cross-module collaboration within the same system. Therefore, there is an urgent need to develop a neuromodulation system that encompasses the entire process of "functional localization—precise navigation—closed-loop control." This system should operate under a high-field magnetic resonance imaging platform and promote the systematic and refined development of neuromodulation technology. Summary of the Invention

[0012] The present invention provides a visual perception-decision closed-loop neural regulation system. The system constructs a task-induced functional map based on the perception-decision behavior paradigm, combines an MRI-compatible light stimulation interface with a behavior-brain network closed-loop feedback mechanism, and realizes high-precision, systematic and individualized neural regulation.

[0013] To achieve the above object, the present invention provides the following technical solutions: A closed-loop neural control system for visual perception and decision-making, comprising three modules: The module for personalized functional map construction and target positioning constructs a personalized functional map of the region based on the visual perception-decision-making task paradigm. The implantable interface base is first positioned and registered based on the functional map and first structural imaging containing MRI imaging markers. A second positioning and registration is then performed based on the functional map and second structural imaging containing MRI imaging markers on the quick-release grid insert to obtain the stimulation path. An MRI-compatible targeted light stimulation interface module, comprising an implantable interface base, a quick-release grid plate mounted on the implantable interface base, and a step-through fiber optic pin kit that cooperates with the quick-release grid plate; The closed-loop control module, driven by behavior-brain network feedback, controls the step-type fiber optic pin kit to perform neural stimulation on the target functional domain according to the stimulation path, collects behavior and brain network signals as feedback indicators, models the stimulation parameters and feedback indicators, and updates the stimulation parameters based on the modeling results.

[0014] The visual perception and decision-making closed-loop neural regulation system provided by the present invention constructs an individualized functional map based on the perception-decision-making task paradigm, combines an MRI-compatible targeted light stimulation interface and a closed-loop control mechanism driven by behavior-brain network feedback, and is implemented at millimeter-level spatial resolution.

[0015] The individualized functional map construction and target location module includes: Based on the visual perception-decision task paradigm to induce BOLD (Blood Oxygen Level-Dependent) response, functional magnetic resonance imaging signals were collected and individualized functional maps across regions were constructed; By placing vitamin E capsule imaging markers on the skull surface and acquiring T1 high-resolution structural images, the first structural image is obtained and a three-dimensional reference system for spatial registration is established; Based on the one-time positioning and registration of the functional map and the first structural imaging, the installation position and angle of the implantable interface base on the cranial surface are simulated and calculated; Insert the quick-release grid plug-in plate filled with MRI development material into the base of the implantable interface to perform secondary structural imaging and secondary positioning and registration with the functional map, establish a spatial mapping relationship between the position of the socket array in the quick-release grid plug-in plate and the functional activation area, and fine-tune the stimulation path based on the spatial mapping relationship to obtain the optimal stimulation path.

[0016] The specific implementation method for constructing a cross-regional individualized functional map can be: The personalized functional map construction and target localization module uses a well-structured, quantifiable visual perception-decision-making task paradigm as the driving condition for functional map construction. This task paradigm requires participants to discriminate and respond to different visual stimuli presented bilaterally within a time limit. Visual stimulus parameters include contrast, color, motion direction, and spatiotemporal frequency. This task paradigm has a high degree of controllability and behavioral relevance.

[0017] During task execution, the system acquires functional magnetic resonance imaging (fMRI) signals, extracts task-induced BOLD response patterns, and constructs individualized functional maps covering multiple cortical regions, including visual-related cortex, parietal cortex, and prefrontal cortex. Compared with resting-state functional maps or anatomical templates, this method exhibits significant advantages in the following aspects: (1) visual stimulus input parameters are controllable; (2) functional responses are correlated with behavioral performance; and (3) neural synergy dynamics across cortical regions can be quantitatively captured. These features effectively improve the task relevance and spatial accuracy of functional target identification, providing a key basis for precise targeted stimulation of subsequent modules.

[0018] Thanks to the superior spatial resolution of 7T ultra-high-field functional magnetic resonance imaging (fMRI), which surpasses conventional imaging methods, and allows voxel sizes to reach 1 mm or even submillimeter levels, this system is capable of highly precise localization of millimeter-scale functional domains within the cortex. This is particularly applicable to regions with highly differentiated functional columnar structures, such as the visual cortex. In contrast, conventional electroencephalography (EEG) technology is limited in spatial resolution by the scalp volume conductance effect, resulting in actual functional localization capabilities typically at the centimeter level, making it difficult to accurately resolve local neural spatial topology.

[0019] The specific implementation method of one-time positioning and registration can be: By establishing the spatial coordinate conversion relationship between the five geometric reference points on the interface implant base and the development markers, the relative coordinate parameters of the implant base on the skull surface are obtained, and the installation position and angle of the implant base on the skull surface are obtained.

[0020] Among them, the specific implementation method of the secondary positioning registration can be: The second structural imaging is used to obtain the actual three-dimensional position distribution (spatial coordinate information) of the grid array in the magnetic resonance space, perform rigid body registration and affine transformation on the second structural imaging and the functional map, and establish the spatial mapping relationship of the socket array; based on the spatial mapping relationship, calculate the Euclidean distance from the center of each socket to the center of mass of the target functional domain, and determine the optimal stimulation path accordingly to ensure that subsequent stimulation operations act accurately on the functional target area.

[0021] The personalized functional map construction and target positioning module provided by the present invention constructs a personalized functional map by combining task-state fMRI signals induced under the perception-decision task paradigm with high-resolution structural images, which significantly improves the spatial accuracy and behavioral correlation of functional positioning, and provides a quantitative basis and spatial coordinate guarantee for the subsequent light stimulation module to achieve millimeter-level target intervention; subsequently, the functional target positioning with millimeter-level spatial accuracy is achieved by combining the primary positioning alignment and the secondary positioning alignment.

[0022] The MRI-compatible targeted light stimulation interface module includes: The implantable interface base is made of polyetheretherketone (PEEK), with a reinforcing rib structure on the outer wall and a limiting boss inside, and is used in conjunction with bone cement and a quick-release mesh insert. The quick-release mesh insert is made of PEEK and features a regularly arranged array of sockets and asymmetric limiting grooves. The stepped fiber optic ferrule kit includes multiple standardized ferrules with a 0.5mm pitch. The ferrules have ceramic stoppers at the end and an insertion depth of 2-20mm, compatible with the jack array. The protective cover adopts an internal thread connection structure to be sealed and matched with the base of the implantable interface.

[0023] Among them, the sockets of the quick-release grid plug-in plate are filled with MRI contrast agent (such as the contents of vitamin E capsules), and combined with T1 magnetic resonance structural image acquisition, secondary spatial alignment of the socket array position and the task-state functional map is achieved; based on the alignment results, the plane coordinates and insertion angles of the sockets can be customized as needed to meet the spatial intervention requirements of different cortical targets; this component supports multiple pin insertion operations and imaging calibration, has good reusability and spatial stability, and can effectively ensure the spatial accuracy and individual adaptability of the stimulation path.

[0024] Among them, the top of the quick-release grid plug-in plate is provided with a threading hole and is equipped with a nylon wire pulling structure; the outer edge of the quick-release grid plug-in plate is provided with three limiting recessed platforms, which are mechanically matched with the limiting bosses on the inner wall of the implantable interface base; by providing a threading hole on the top of the quick-release grid plug-in plate and cooperating with the nylon wire pulling structure, it is used for plugging and unplugging operations; the mechanical cooperation between the bosses ensures that the plug-in plate has consistent installation directions and good spatial consistency over multiple times.

[0025] The step-by-step fiber optic pin kit can meet the needs of stimulating cortical functional areas at different depths. Each pin is equipped with a ceramic stopper structure at the end, achieving an operating accuracy of better than ±0.5mm. This design avoids the interference of traditional push-in structures in the MRI environment, and it balances magnetic compatibility with precise control capabilities, making it suitable for targeted intervention at different depths of cortical functional areas.

[0026] Among them, the step-by-step fiber optic pin kit is used to perform optical stimulation on the target functional domain. The laser can be transmitted through the extended fiber optic system. The extended section can be precisely connected to the fiber optic pin using a ceramic sleeve, thereby accurately guiding the laser energy to the target functional domain and achieving neural regulation operations with millimeter-level spatial resolution.

[0027] Among them, the protective cover is used for long-term implant protection: after the stimulation experiment is completed, the interface supports the installation of a non-metallic protective cover for closure to reduce the risk of infection and ensure the tissue compatibility and stability of the interface during long-term implantation.

[0028] The MRI-compatible targeted light stimulation interface module is used to perform precise neural stimulation operations in a high-field magnetic resonance environment. It has excellent spatial adaptability, magnetic compatibility, and stimulation targeting accuracy. It is a key component that supports the conversion of the target positioning results of this system into an implementable stimulation path. It works in conjunction with the individualized functional map construction and target positioning module to accurately map the cortical targets identified in the individualized functional map to the surgical operation space, thereby achieving spatially targeted intervention in the target area.

[0029] To meet the needs of neural regulation experiments in high-field magnetic resonance environments, the total height of the MRI-compatible targeted light stimulation interface module is controlled within 8mm. This flat design allows the radio frequency coil to be close to the cerebral cortex, effectively improving the signal-to-noise ratio of fMRI imaging. Through compact structural design and precise control, it ensures millimeter-level targeted intervention of the target cortical functional domain under imaging conditions with strict magnetic compatibility requirements.

[0030] The behavior-brain network feedback-driven closed-loop control module includes: The neural stimulation control actuator sets the stimulation parameters, and the output laser is transmitted through the step-type optical fiber pin kit to perform optical stimulation on the target functional area according to the stimulation path; A behavioral and brain network signal acquisition unit, used to synchronously record feedback indicators (including BOLD signal amplitude in task-related brain regions and functional connectivity between regions): behavioral indicators and fMRI brain network data. The behavioral indicators include choice accuracy, reaction time, and choice bias, and the fMRI brain network data includes activation amplitude and connection strength. The feedback index modeling unit is used to construct a regression model for the mapping relationship between stimulation parameters and feedback indicators, and to perform training and round-by-round updates; The control parameter update unit is used to adjust the stimulation parameters according to the output results of the regression model and update them to the neural stimulation control actuator to achieve closed-loop optimization control.

[0031] Among them, in the neural stimulation control actuator, light stimulation of the target cortical area is performed according to the set parameters. The stimulation parameters include power, frequency and target coordinates. The laser source is a laser with a central wavelength of 1870±10nm, which is introduced into the brain area through a polymer optical fiber with a core diameter of 200μm and NA=0.22; in the feedback indicator modeling unit, the stimulation parameters include power, frequency and target coordinates. Multiple linear regression, Ridge regression or gradient boosting-based XGBoost integrated learning method is used for model training and round-by-round updating. During training and round-by-round updating, the fitting evaluation indicators include R² value and standardized regression coefficient.

[0032] Specifically: the feedback indicator modeling unit is used to establish a mapping relationship between stimulation parameters and behavioral-neural feedback; the modeling unit introduces a multiple regression analysis method, using the parameters set for each round of stimulation, including stimulation power, frequency, and target coordinates, as independent variable inputs; using synchronously collected behavioral indicators, including selection accuracy, reaction time, and selection bias, and fMRI neurofeedback indicators, including the activation amplitude of task-related brain areas and the strength of cross-regional functional connections, as a set of dependent variables to construct a regression model; the model can adopt linear or nonlinear regression form, support round-by-round data updates, and the output is used to evaluate the impact of stimulation setting parameters on behavior and brain networks, thereby dynamically adjusting subsequent stimulation parameters.

[0033] The control parameter updating unit includes: The current stimulation parameter effectiveness evaluation mechanism determines the effectiveness of the stimulation based on whether the change in the feedback indicator exceeds the set threshold. If the stimulation is judged to be ineffective, a parameter search and iteration mechanism is used. Parameter search and iteration mechanism, using the minimum mean square error optimization algorithm to adjust the stimulation parameter combination; A safety threshold setting mechanism is used to limit the maximum stimulation energy and frequency to ensure the safety of the subjects. If the stimulation is judged to be ineffective, a parameter search and iteration mechanism is adopted, and the adjusted stimulation parameter combination is used for the next round of light stimulation.

[0034] The control parameter update unit automatically updates the stimulation parameters based on the modeling results and transmits the new parameters to the neural stimulation control actuator, achieving closed-loop control. Based on the regression model output, the system adjusts key parameters, including stimulation energy and frequency, to enhance the individualized adaptability and control effectiveness of the stimulation. This mechanism significantly enhances the targeting, interpretability, and closed-loop control capabilities of the control strategy.

[0035] The system runs on a 7T ultra-high field magnetic resonance imaging platform and supports BOLD signal acquisition with millimeter-level resolution.

[0036] This invention constructs a closed-loop neural control system for visual perception and decision-making. The overall structure consists of the following three functional modules: an individualized functional map construction and target localization module, which is used to induce functional magnetic resonance imaging signals driven by perception-decision-making tasks, obtain individualized task-state functional maps, and combine MRI imaging markers with image registration algorithms to achieve millimeter-level functional domain identification and target localization; an MRI-compatible targeted light stimulation interface module, suitable for neural stimulation operations in high-field magnetic resonance environments, with millimeter-level targeting accuracy and good spatial adaptability of radiofrequency coils. Its compact structure supports high-precision optical fiber targeted stimulation of specific functional areas of the cortex at the millimeter scale; and a behavior-brain network-driven closed-loop control module, which is used to synchronously record behavioral indicators and fMRI brain network responses, construct a feedback model, and update stimulation parameters to achieve closed-loop optimization of control parameters. Each module is closely logically linked and works together to achieve full-process control, from functional localization to precise stimulation to closed-loop feedback, at millimeter-level spatial resolution.

[0037] The visual perception decision-making closed-loop neural control system provided by the present invention is suitable for neural function control experiments in high-field strength magnetic resonance imaging environments, and is particularly suitable for millimeter-level functional domain positioning and control based on individualized task states; the present invention realizes the systematic integration of "functional positioning-precise navigation-closed-loop control" at millimeter-level spatial resolution, and promotes the development of neural control technology towards high precision, systematization and individualization.

[0038] The visual perception decision-making closed-loop neural control system provided by the present invention solves the following problems in existing neural control technologies: 1) the spatial resolution of functional localization methods is insufficient, making it difficult to accurately identify and target millimeter-scale functional domains in the cortex; 2) the existence of current diffusion in electrical stimulation methods; 3) the lack of an individualized behavior-brain network feedback mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of a closed-loop neural control system for visual perception decision-making provided in an embodiment; Figure 2 A schematic diagram of constructing an individualized functional domain map in an embodiment; Figure 3 Schematic diagram of an MRI-compatible targeted light stimulation interface module in an embodiment; Among them, Figure 3 Middle: 1 is the implantable base, 2 is the quick-release grid plug-in plate, 3 is the step-depth fiber optic pin kit, 4 is the protective cover; 21 is the asymmetric groove mark, 22 is the threading through hole, 23 is the jack array, 24 is the limiting concave platform; 11 is the limiting boss, and 12 is the raised reinforcement rib; Figure 4 Schematic diagram of closed-loop feedback indicators. DETAILED DESCRIPTION

[0040] The closed-loop neural control system for visual perception and decision-making provided in this embodiment includes three major modules: Individualized functional map construction and target positioning module, used to complete functional map construction and target positioning; MRI-compatible targeted light stimulation interface module for precise implementation of physical stimulation pathways; A closed-loop control module driven by behavior-brain network feedback is used to dynamically optimize and adjust stimulation parameters.

[0041] Combine Figure 1 , explaining the specific implementation process of each module one by one.

[0042] Module 1: Individualized functional map construction and stimulation target positioning module This module aims to achieve millimeter-precision functional target identification and targeted stimulation path planning under high-field magnetic resonance imaging conditions: construct an individualized functional map of the region based on the visual perception-decision-making task paradigm, perform a primary positioning and registration of the implantable interface base based on the functional map and the first structural imaging containing MRI imaging markers, and then perform a secondary positioning and registration based on the functional map and the second structural imaging of the quick-release grid insert containing MRI imaging markers to obtain the stimulation path.

[0043] This module specifically includes the following four steps: Step 1-1): Construction of functional maps based on perception-decision tasks In this embodiment, a visual contrast discrimination task was used as a paradigm to induce neural function activation in order to construct a functional map with spatiotemporal specificity in the perception-decision-making process of the individual subject.

[0044] This embodiment uses a bilateral visual perception-decision-making task as a paradigm for constructing functional maps. In each trial, the subject fixates on the central point and is presented with a sinusoidal wave drift grating in the left and right visual fields at the same time. The contrast of the standard stimulus on one side is fixed at 50%, and the contrast of the test stimulus on the other side is set with multiple gradient conditions between 10% and 90%. The stimulus is presented in the lower quadrant on both sides, with an eccentricity of about 4°, and the stimulus area is a circular grating with a diameter of 2°. The contrast conditions are presented alternately in a pseudo-random order, and the sampling density of the conditions near the perception threshold is higher to improve the discrimination of the task and the sensitivity of behavioral measurement. The subject selects the stimulus direction with higher contrast by saccade, and the behavioral monitoring program provides rewards based on the results of their selection.

[0045] In this example, functional MRI data were acquired using a T2*-weighted gradient echo planar imaging (EPI) sequence using a 7T high-field MRI system in combination with a custom coil to achieve high temporal and spatial resolution of the BOLD signal. Key acquisition parameters included a repetition time (TR) of 2000 ms, an echo time (TE) of 25 ms, a flip angle of 90°, a slice thickness of 1.0 mm, a spatial resolution of 1.0 × 1.0 × 1.0 mm³, and 38 coronal slices for full brain coverage. The image matrix size was 96 × 96, with a field of view (FOV) of 96 mm. GRAPPA acceleration mode (acceleration factor 2) was used to shorten acquisition time and improve image signal-to-noise ratio.

[0046] like Figure 2 As shown in a in Figure 1, driven by a visual perception-decision-making task, this embodiment can stably induce blood oxygen level-dependent (BOLD) signal responses in multiple brain regions, including the visual cortex, parietal cortex, and prefrontal cortex. The resulting task-related functional map has excellent spatial resolution and task specificity, serving as a key reference for subsequent personalized functional domain target stimulation. Furthermore, the centroid coordinates of each activated functional domain can be extracted as the spatial coordinates of the control target, enabling millimeter-level precision positioning of the functional target and subsequent stimulation path planning.

[0047] Step 1-2): MRI imaging marker positioning and structural image acquisition like Figure 2 As shown in (b), to achieve spatial mapping between the functional map and the physical stimulation pathway, three MRI imaging markers are pre-placed on the individual's skull surface to form a stable spatial positioning reference system. This embodiment uses vitamin E soft capsules as imaging markers. Their contents are composed of a mixture of dl-α-tocopheryl acetate and vegetable oil. These soft capsules are rich in fat-soluble hydrocarbon chain structures (-CH2-CH3), have high proton density, and exhibit short T1 relaxation properties. Therefore, they appear as bright, high-signal spots in T1-weighted imaging sequences, resulting in clear imaging and excellent MRI compatibility and biosafety.

[0048] The markers are attached to the skull at pre-set locations using PMMA dental bone cement. The contact surfaces at the attachment points can be marked with carbon ink for surgical reference. The marker spacing ensures that the three points are not collinear, thus establishing a stable three-dimensional coordinate system.

[0049] Subsequently, T1-weighted high-resolution structural images were acquired to clearly visualize the landmarks. Specifically, a 7T ultra-high-field MRI system with a custom head coil was used. The acquisition sequence parameters were as follows: an MPRAGE sequence with a spatial resolution of 0.5 × 0.5 × 0.5 mm³, a TR of 2300 ms, a TE of 3.61 ms, a TI of 1100 ms, a flip angle of 7°, and GRAPPA parallel acceleration mode (acceleration factor 2). The acquisition matrix was 192 × 192, with 120 slices, covering a 96 mm brain area.

[0050] like Figure 2 As shown in Figure 2(b), the acquired structural image clearly displays the locations of imaging markers in three-dimensional space. This imaging signal is stable and has high contrast, making it suitable as a reference point for non-invasive MRI spatial positioning and highly practical in experiments. A stable three-dimensional spatial reference system is then constructed, serving as a geometric basis for subsequent image registration and interface installation position calculations.

[0051] Steps 1-3) - Simulation calculation and surgical installation of the implantable stimulation interface base This embodiment aims to accurately map the target coordinates identified in the functional map to the actual surgical operation coordinates, and perform implantable stimulation interface base installation simulation design and surgical implantation based on the spatial alignment results, thereby achieving precise targeted coverage of the functional activation domain.

[0052] The functional atlas acquired in step 1 and the structural image with development markers acquired in step 2 were spatially registered, and the optimal installation angle and plane position of the stimulation interface on the cranial surface were calculated using simulation software. Specifically, First, AFNI and FreeSurfer were used as image processing software to perform rigid body transformation and affine registration on the fMRI functional map obtained in step 1 and the high-resolution T1 structural image acquired in step 2.

[0053] After image registration is completed, the functional activation map is projected onto the structural image with development markers using Planner simulation software to calculate the optimal installation plane and angle of the implantable stimulation interface base, such as Figure 2 The simulation design must meet the following constraints: (1) The implantable stimulation interface base installation area is not covered by large blood vessels; (2) other implants are avoided; (3) the implantable stimulation interface base can completely cover the stimulation target, and a space redundancy of ≥0.5 mm is reserved to compensate for operational errors.

[0054] Furthermore, by establishing the spatial coordinate transformation relationship between the five geometric reference points on the base of the implantable stimulation interface and the imaging markers, a parameter set guiding the surgical installation of the interface was obtained, enabling precise surgical implantation of the interface base. Specifically: The key geometric points on the base of the implantable stimulation interface, including the center and four-way edges, and the three imaging markers form two independent spatial reference systems, such as Figure 2 As shown in Figure d, P1-P3 are schematic diagrams of three imaging landmarks. The red circle in the lower right corner represents the interface base, and the blue dots above the red circle represent key geometric points on the interface base. By extracting its 3D coordinates from the T1 structural image and calculating the transformation matrix between the two coordinate systems, the relative coordinate parameters for the precise installation of the interface on the individual's craniofacial surface can be derived. These parameters can be exported in CSV format for intraoperative reference.

[0055] Finally, based on the simulation results, the implantable stimulation interface base was surgically installed in a sterile environment. PMMA bone cement (MRI-compatible fixation material) was used to adhere the implantable stimulation interface base, providing a stable mounting platform for the subsequent interface module.

[0056] Steps 1-4) - Grid imaging for secondary positioning calibration In this embodiment, after the surgical installation of the implantable stimulation interface base is completed, a quick-release grid insert pre-filled with the contents of a vitamin E capsule as MRI imaging material is inserted into the interface base. The grid insert is arranged in a regular array structure, and the spatial geometric relationship of all the socket positions has been calibrated. The imaging material inside the sockets can be displayed as a clear high-intensity signal in the T1-weighted structural image, forming a recognizable bright spot array. The new structural image is collected to obtain the actual three-dimensional position distribution of the grid array in the magnetic resonance space (see Figure 2 e in step 1), and then the structural map is rigidly registered and affine transformed with the task-related functional map obtained in step 1 to establish the spatial correspondence between the grid socket coordinates and the functional activation area (see Figure 2 Based on this mapping relationship, the system automatically calculates the Euclidean distance from the center of each socket to the centroid of the target functional domain and selects the nearest socket or socket combination as the actual stimulation path, achieving individualized fine-tuning and precise coverage of the functional target area.

[0057] The secondary positioning calibration step in this embodiment optimizes the registration accuracy between the socket array in the grid plug board and the functional map by spatially arranging the development markers and the stimulation path, which is a key step in achieving millimeter-level stimulation targeting.

[0058] Module 2: MRI-compatible targeted light stimulation interface module This embodiment provides a targeted neural stimulation interface module suitable for high-field magnetic resonance environments. It has a compact structure and good magnetic compatibility, and can achieve millimeter-level neural stimulation operations on specific functional domains of the cortex. This module combines the individualized functional map obtained in module 1 with the stimulation target coordinate information to complete the transformation of the map space into the actual operation space, and achieve precise fiber insertion and stimulation path control. Figure 3 As shown in a, it mainly consists of the following four structures: 2-1) Implantable interface base 1 like Figure 3 As shown in Figure c, the implantable interface base 1 is the core support platform of the interface system. Its design goal is to achieve long-term implantation and precise spatial positioning while ensuring MRI compatibility and mechanical stability. Its main structural features include: The material and processing technology used are polyetheretherketone (PEEK) and are formed through high-precision CNC machining. This material has high strength, low magnetic susceptibility, and excellent biocompatibility, making it suitable for long-term skull implantation.

[0059] Mechanically stable structure: four raised reinforcing ribs 12 are evenly distributed on the outer wall of the implantable interface base 1, which can form a mechanical interlocking structure with dental bone cement during surgery, significantly improving adhesion and preventing dislocation.

[0060] The base has an embedded limiting structure and is provided with three limiting bosses 11 inside, which are used to form a stable fit with the quick-release grid plug-in plate 2, thereby preventing the quick-release grid plug-in plate 2 from rotating or tilting during use and improving the repeatability of the socket positioning.

[0061] Implantation method: The size and geometric structure of the implantable interface base 1 support individual customization, and can be fixed in the craniotomy area in combination with ceramic micro bone screws and bone cement to ensure the long-term stability and spatial reference consistency of the stimulation interface.

[0062] 2-2), quick release grid plug-in board 2 like Figure 3 As shown in b in FIG, the quick-detachable grid insert 2 in this embodiment is designed to be installed on the implantable base 1. As a detachable component of the stimulation interface, it has both secondary calibration and stimulation targeting functions, and has the advantages of precise positioning and rapid replacement.

[0063] There are two symmetrically distributed threading holes 22 on both sides of the top of the plugboard, which can be used to insert nylon wire as a pulling structure, making it easy to quickly extract the quick-release grid plugboard 2 in an area with limited space, thereby improving operational convenience and module reuse efficiency.

[0064] In order to ensure that the quick-release grid plug-in plate 2 is installed in the same direction each time, an asymmetric groove structure 21 is provided in the vertical direction of the outer edge of the quick-release grid plug-in plate 2 as a direction identification mark, which cooperates with the alignment and limiting boss structure of the base to ensure the spatial consistency of the quick-release grid plug-in plate 2 in multiple installations and avoid stimulation coordinate errors due to wrong direction.

[0065] A regularly arranged array of sockets 23 is provided in the center of the quick-release grid plug-in plate 2. A limiting structure is provided inside each socket to fix the inserted ceramic fiber optic pins, prevent the pins from axially slipping during the experiment, and improve the spatial stability and repeatability of targeted stimulation.

[0066] After functional map registration and target identification are completed in Module 1, MRI imaging material, such as the contents of a vitamin E capsule, can be injected into the socket. The T1 structural image is then used to recalibrate the plug position, achieving secondary spatial registration between the grid array and the functional map, thereby precisely guiding the stimulation path. During the actual stimulation phase, the selected socket angle and coordinates can be customized to the individual target to match the desired stimulation path, enabling personalized targeted intervention at the millimeter level.

[0067] Furthermore, three retaining recesses 24 are provided on the outer edge of the quick-release grid insert 2, which mechanically engage with the retaining protrusions 11 on the inner wall of the base. During installation, this structure allows for rapid alignment and positioning of the quick-release grid insert 2, preventing rotational deviation and axial misalignment, effectively improving the modular interface's assembly efficiency and overall structural stability.

[0068] 2-3), step-type fiber optic pin kit 3 To meet the need for precise stimulation of cortical functional areas at different depths, this embodiment is equipped with a set of stepped fiber optic pin kits 3, which include multiple standardized pins with a step spacing of 0.5mm, covering a depth range of 2-20mm. Each pin is precision-machined and molded, with a ceramic stopper structure installed at the end. The insertion depth is determined by the length of the pin itself, avoiding magnetic compatibility interference caused by traditional propulsion mechanisms, ensuring stable operation within the limited area of ​​the magnetic resonance coil, and ensuring an insertion error of less than ±0.5mm. The pin body is constructed from polymer-coated quartz fiber, offering excellent light transmission performance and structural rigidity. The optically polished tip ensures uniform light output, effectively transmitting laser light for optical stimulation of cortical targets. The precise ceramic retaining structure at the tail end securely engages the socket of the quick-release grid plate.

[0069] In this embodiment, light stimulation is based on laser output, which is coupled into the pin through an optical fiber path, and ultimately focuses the laser energy on a millimeter-scale functional domain, achieving high spatial precision neural regulation without the risk of current diffusion.

[0070] 2-4), protective cover 4 The protective cover 4 in this embodiment is used to close the interface during the non-stimulation phase to prevent contamination and foreign matter from entering, thereby improving the stability and biosafety of long-term implantation. The protective cover 4 is made of polyetheretherketone (PEEK) material and adopts an internal thread design, which can be tightly screwed together with the implantable base 1 to form a good seal. A vertical groove is provided on the top of the protective cover 4, which can be easily screwed in or out using a plastic ruler tool, making it suitable for operations under space-constrained conditions. The overall structure is compact and has good magnetic resonance compatibility and reusability.

[0071] All interface components can be sterilized with ethylene oxide to meet the use and safety requirements under long-term experimental conditions.

[0072] The MRI-compatible targeted light stimulation interface module structure provided in this embodiment can correspond to the interface-functional area registration result established in module one, thereby realizing complete mapping and execution from individualized atlas to spatial targeting path.

[0073] Module 3: Closed-loop control module driven by behavior-brain network feedback The module described in this embodiment is designed to construct a stimulation parameter control model by collecting behavioral and brain network feedback signals in real time under the visual perception-decision-making task paradigm, thereby realizing a closed-loop control mechanism of stimulation-feedback-optimization. This module consists of the following four functional units: Unit 1: Neural Stimulation Control of Actuators The neural stimulation control actuator in this embodiment is used to set parameter control instructions and execute neural stimulation operations on the target cortical area accordingly. Taking near-infrared light stimulation as an example, the system is configured with a laser with a central wavelength of 1870±1nm as the stimulation source. The laser is transmitted through a flexible polymer optical fiber with a core diameter of 200μm and a numerical aperture (NA) of 0.22, and is accurately introduced into the target functional domain through the MRI-compatible targeted light stimulation interface of module 2, thereby achieving spatial alignment between the stimulation site and the target in the task-induced functional map. The laser output parameters can be flexibly modulated according to the preset power, frequency and pulse width, and are precisely controlled by the control software through TTL signals or digital communication interfaces to ensure that the stimulation waveform is highly synchronized with the experimental task.

[0074] Unit 2: Behavior and brain network signal acquisition like Figure 4 As shown, the behavior and brain network signal acquisition unit in this embodiment is designed to synchronously obtain multimodal feedback data during the execution of visual perception-decision-making tasks, providing a quantitative basis for subsequent feedback modeling and stimulation parameter updating.

[0075] In terms of behavioral feedback collection, behavioral feedback includes quantifiable behavioral indicators such as choice accuracy, reaction time, and choice bias. This example is equipped with an MRI-compatible infrared eye tracker (Eyelink 1000 Plus, SRResearch) with a sampling frequency of 250 Hz to record the subject's eye movement trajectory during the task. By setting a fixed fixation point and a choice response window, this example calculates key behavioral indicators such as choice accuracy, reaction time, and left-right choice bias for each trial. The results are recorded and output on a trial-by-trial basis.

[0076] For brain network feedback acquisition, functional magnetic resonance imaging (fMRI) signals are used to extract activation amplitudes and cross-regional functional connectivity strengths in task-related brain regions. This example simultaneously acquires whole-brain BOLD signals on a 7T high-field MRI platform using a T2*-weighted gradient echo EPI sequence (TR = 2000ms, TE = 25ms, flip angle = 90°, 1mm equidistant voxel resolution, and 38 slices covering the entire brain), ensuring excellent spatiotemporal resolution of task-related brain regions.

[0077] The collected behavioral and brain network data will be aligned according to the trial timestamps, bound to the stimulation setting parameters for each round, and stored in a CSV standardized format for use by the feedback modeling unit.

[0078] This embodiment can automatically detect abnormal responses or low-quality signals, set data cleaning rules, and exclude trials with gaze loss exceeding a threshold, so as to improve the reliability of subsequent model construction.

[0079] The synchronous data acquisition capability of this embodiment ensures that multi-dimensional, quantifiable behavior-brain network feedback indicators can be obtained after each round of stimulation, forming a key data support path in the system closed-loop control mechanism.

[0080] Unit 3: Feedback indicator modeling unit The feedback indicator modeling unit in this embodiment is used to establish a functional relationship between neural stimulation parameters and simultaneously acquired behavioral performance and brain network feedback, thereby enabling quantitative assessment and adjustment of stimulation effects. This unit is a key computing module that implements the core closed-loop control logic. Its main functions include data preprocessing, feature extraction, regression modeling, and model updating.

[0081] This embodiment first standardizes the data of each round of stimulation experiments. The stimulation parameter set of each round of experiments includes: light stimulation power, stimulation frequency, three-dimensional coordinates of the target point, and stimulation execution time as additional meta-information, and these parameters serve as independent variables of the model.

[0082] At the same time, this embodiment synchronously reads the corresponding feedback indicator set for each round from the behavioral and brain network signal acquisition unit as the model's dependent variable. This set includes behavioral indicator dimensions such as choice accuracy, reaction time, and choice bias probability; and brain network feedback dimensions such as the BOLD signal amplitude of each activated region under the behavioral paradigm and the functional connectivity indicator between each region and other regions.

[0083] These variables are standardized and fed into the modeling process. This modeling unit employs a variety of regression algorithms to construct the mapping between stimulation parameters and feedback metrics. The basic model is a multivariate linear regression model. When the feedback structure involves nonlinear relationships, ridge regression or gradient boosting-based XGBoost ensemble learning methods can be used.

[0084] Regression model training is performed in a round-by-round update manner: that is, after each round of stimulation experiments, the system appends the new stimulation-feedback data pairs to the historical data to refit the model, and calculates the goodness of fit of each predicted dependent variable, such as R², and the parameter importance coefficient, such as the standardized β value, to evaluate the regulatory effect of different stimulation parameters on behavior and brain network status.

[0085] Furthermore, this embodiment also features a mechanism for setting control thresholds and indicator weights. For example, if a specific feedback indicator, such as visual area activation intensity, is the primary control target, the system can increase the weight of that indicator in the model evaluation function, causing the parameter optimization strategy to converge in that direction.

[0086] The final output model not only evaluates the regulatory effect of the current stimulation setting, but also provides directional suggestions for unit 4 on updating the stimulation parameters, such as "whether increasing the stimulation power will help improve the selection accuracy", thereby realizing model-driven closed-loop regulation.

[0087] In summary, this embodiment structures, quantifies, and predicts the effects of neural stimulation by introducing statistical modeling and a round-by-round update mechanism, providing technical support for efficient and individualized neural intervention strategies.

[0088] Unit 4: Control Parameter Update The control parameter updating unit in this embodiment is used to dynamically adjust the neural stimulation parameters according to the output results of the behavior-neural feedback regression model established in unit 3, thereby realizing a closed-loop neural control strategy for the individual state.

[0089] The control parameter update unit receives the modeling estimation results and performs the following two operations: 1. Current parameter effectiveness assessment: The system first determines whether the current stimulation has achieved the expected regulatory effect based on a comparison of the model's predicted value and the actual feedback data. The configurable evaluation criteria include: behavioral indicator changes: a change of more than 2% in the accuracy of the choice; neural indicator changes: a significant change in the BOLD amplitude of the target functional domain (p < 0.05); and enhanced functional connectivity between functional domains: significantly higher than the baseline round. 2. Parameter Search and Update: If the expected results are not met, the system will perform an iterative parameter search within the current parameter neighborhood. For example, based on the minimum mean square error optimization algorithm, it will adjust the stimulation power, frequency, or target coordinates to form the next round of recommended stimulation parameter configurations.

[0090] The updated stimulation parameters, after user evaluation, are transmitted via TTL or digital communication via the control software to the laser control interface in Unit 1, where they are used to set the stimulation protocol for the next round of experiments. This process forms a complete closed-loop control pathway, supporting multiple rounds of feedback-based parameter optimization iterations.

[0091] During the parameter update process, this embodiment also supports setting safety range limit conditions (such as maximum stimulation energy not exceeding 1.0J / cm² and frequency not exceeding 400Hz) to ensure subject safety and avoid stimulation overload or cumulative risks.

[0092] This module implements closed-loop control and iterative optimization of neural stimulation parameters during visual perception-decision-making tasks, a crucial component of the system's precise neural control. Building on the individualized functional map construction and target location achieved in Module 1, and the targeted stimulation performed in Module 2, the system simultaneously collects behavioral indicators and functional magnetic resonance imaging (fMRI) brain network data through Module 3, constructing a quantitative stimulation effect evaluation model. This model is then used to dynamically update stimulation parameters, forming a closed-loop control pathway of stimulation, feedback, and optimization.

Claims

1. A closed-loop neural control system for visual perception decision making, characterized by: The system consists of three modules: The module for personalized functional map construction and target positioning constructs a personalized functional map of the region based on the visual perception-decision-making task paradigm. The implantable interface base is first positioned and registered based on the functional map and first structural imaging containing MRI imaging markers. A second positioning and registration is then performed based on the functional map and second structural imaging containing MRI imaging markers on the quick-release grid insert to obtain the stimulation path. An MRI-compatible targeted light stimulation interface module, comprising an implantable interface base, a quick-release grid plate mounted on the implantable interface base, and a step-through fiber optic pin kit that cooperates with the quick-release grid plate; The closed-loop control module, driven by behavior-brain network feedback, controls the step-type fiber optic pin kit to perform neural stimulation on the target functional domain according to the stimulation path, collects behavior and brain network signals as feedback indicators, models the stimulation parameters and feedback indicators, and updates the stimulation parameters based on the modeling results.

2. The visual perception decision closed-loop neural control system according to claim 1, characterized in that: The individualized functional map construction and target location module includes: Based on the visual perception-decision-making task paradigm to induce BOLD response, collect functional magnetic resonance imaging signals and construct individualized functional maps across regions; By placing vitamin E capsule imaging markers on the skull surface and acquiring T1 high-resolution structural images, the first structural image is obtained and a three-dimensional reference system for spatial registration is established; Based on the one-time positioning and registration of the functional map and the first structural imaging, the installation position and angle of the implantable interface base on the cranial surface are simulated and calculated; Insert the quick-release grid plug-in plate filled with MRI development material into the base of the implantable interface to perform secondary structural imaging and secondary positioning and registration with the functional map, establish a spatial mapping relationship between the position of the socket array in the quick-release grid plug-in plate and the functional activation area, and fine-tune the stimulation path based on the spatial mapping relationship to obtain the optimal stimulation path.

3. The visual perception decision closed-loop neural control system according to claim 2, characterized in that: In one-time positioning and registration, the relative coordinate parameters of the implantable interface base on the cranial surface are obtained by establishing the spatial coordinate conversion relationship between the five geometric reference points on the interface implantable interface base and the development markers.

4. The visual perception decision closed-loop neural control system according to claim 2, characterized in that: In the secondary positioning registration, the second structural imaging and the functional map are rigidly registered and affine transformed to establish the spatial mapping relationship of the jack array; Based on the spatial mapping relationship, the Euclidean distance from the center of each socket to the centroid of the target functional domain is calculated, and the optimal stimulation path is determined accordingly.

5. The visual perception decision closed-loop neural control system according to claim 2, characterized in that: The MRI-compatible targeted light stimulation interface module includes: The implantable interface base is made of polyetheretherketone (PEEK), with a reinforcing rib structure on the outer wall and a limiting boss inside, and is used in conjunction with bone cement and a quick-release mesh insert. The quick-release mesh insert is made of PEEK and features a regularly arranged array of sockets and asymmetric limiting grooves. The stepped fiber optic ferrule kit includes multiple standardized ferrules with a 0.5mm pitch. The ferrules have ceramic stoppers at the end and an insertion depth of 2-20mm, compatible with the jack array. The protective cover adopts an internal thread connection structure to be sealed and matched with the base of the implantable interface.

6. The visual perception decision closed-loop neural control system according to claim 5, characterized in that: The top of the quick-release grid plug-in plate is provided with a threading hole and cooperates with a nylon wire pulling structure; the outer edge of the quick-release grid plug-in plate is provided with three limiting concave structures, which are mechanically matched with the limiting bosses on the inner wall of the implantable interface base.

7. The visual perception decision closed-loop neural control system according to claim 1, characterized in that: The behavior-brain network feedback-driven closed-loop control module includes: The neural stimulation control actuator sets the stimulation parameters, and the output laser is transmitted through the step-type optical fiber pin kit to perform optical stimulation on the target functional area according to the stimulation path; A behavioral and brain network signal acquisition unit, used to synchronously record feedback indicators: behavioral indicators and fMRI brain network data, wherein the behavioral indicators include selection accuracy, reaction time, and selection bias, and the fMRI brain network data includes activation amplitude and connection strength; The feedback index modeling unit is used to construct a regression model for the mapping relationship between stimulation parameters and feedback indicators, and to perform training and round-by-round updates; The control parameter update unit is used to adjust the stimulation parameters according to the output results of the regression model and update them to the neural stimulation control actuator to achieve closed-loop optimization control.

8. The visual perception decision closed-loop neural control system according to claim 7, characterized in that: In the neural stimulation control actuator, the stimulation parameters include power, frequency, and target coordinates. In the feedback indicator modeling unit, multiple linear regression, Ridge regression, or gradient boosting-based XGBoost ensemble learning methods are used for model training and round-by-round updating. During training and round-by-round updating, fitting evaluation indicators include R² value and standardized regression coefficient.

9. The visual perception decision closed-loop neural control system according to claim 8, characterized in that: The control parameter updating unit includes: The current stimulation parameter effectiveness evaluation mechanism determines the effectiveness of stimulation based on whether the change in feedback indicators exceeds the set threshold; Parameter search and iteration mechanism, using the minimum mean square error optimization algorithm to adjust the stimulation parameter combination; Safety threshold setting mechanism, used to limit the maximum stimulation power and frequency to ensure subject safety; If the stimulation is judged to be ineffective, a parameter search and iteration mechanism is adopted, and the adjusted stimulation parameter combination is used for the next round of light stimulation.

10. The visual perception decision closed-loop neural control system according to any one of claims 1 to 9, characterized in that: The system runs on a 7T ultra-high field magnetic resonance imaging platform and supports BOLD signal acquisition with millimeter-level resolution.

Images