A medical device data acquisition terminal

Abstract

This invention relates to the field of medical device monitoring technology and discloses a medical device data acquisition terminal. The terminal comprises modules for multimodal data acquisition, internal data mapping, cross-modal data fusion, dynamic decision-making, and execution feedback. The terminal generates an external health status map and delineates risk areas by collecting surface data from the device; based on this, it targets and deploys internal sensors, generating an internal three-dimensional parameter distribution map using signal penetration and echo analysis; it fuses internal and external data to establish spatiotemporal correlation and outputs a confidence score; it generates medical instructions based on the score and issues them for execution; finally, it dynamically optimizes the data acquisition strategy by monitoring changes in the device's status. This solution achieves precise targeted diagnosis from external abnormalities to internal lesions and improves the accuracy and efficiency of medical device status monitoring through a closed-loop feedback mechanism that adaptively optimizes data acquisition.

Description

Technical Field

[0001] This invention relates to the field of medical device monitoring technology, specifically to a medical device data acquisition terminal. Background Technology

[0002] Existing medical device condition monitoring systems typically employ a discrete detection approach. One common approach relies on externally deployed sensor networks to collect multimodal data such as vibration and temperature from the device's surface, and then triggers alarms based on thresholds. Another approach involves periodically shutting down and opening the medical device for maintenance, or using internal sensors for continuous monitoring to assess the operational status of its core components. These methods form the basis of current mainstream preventative maintenance and fault diagnosis technologies.

[0003] While external monitoring solutions offer non-invasive monitoring, their sensing capabilities are limited to the equipment surface, failing to effectively penetrate the casing to accurately detect early signs of degradation in internal components, such as parameter drift in parts or minor fatigue in mechanical structures. Internal inspection or built-in sensor solutions, while providing direct parameters, require equipment shutdown or structural modifications. Furthermore, comprehensive internal scanning is time-consuming and energy-intensive, lacks specificity, and struggles to achieve real-time, online, and accurate diagnosis. Both approaches suffer from a disconnect between internal and external data, failing to establish a correlation between external anomalies and the root causes of internal lesions.

[0004] Current technology lacks a diagnostic mechanism capable of dynamically linking external macroscopic anomalies with internal microscopic parameters. Potential risk areas identified by external detection cannot guide subsequent internal probing, leading to a blind approach in the internal detection process and wasting computational resources on repetitive scanning of healthy areas. Furthermore, existing data acquisition strategies are often statically preset and cannot self-optimize based on the actual effects of historical decisions. The system cannot assess whether its instructions based on fused data effectively correct the device's state, thus making it difficult to further improve the efficiency and accuracy of data acquisition once a certain level is reached. Summary of the Invention

[0005] The purpose of this invention is to provide a medical device data acquisition terminal to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a medical device data acquisition terminal, the system comprising: Multimodal data acquisition module, internal data mapping module, cross-modal data fusion module, dynamic decision-making module, and execution feedback module; The multimodal data acquisition module is responsible for collecting multimodal data from the surface of the medical device, processing the collected data through a dynamic focus adjustment mechanism, generating an external health status map of the device, and dividing the health status map into risk areas. Based on the risk area division results, the internal data mapping module deploys internal sensing devices, emits detection signals to penetrate the equipment shell, calculates internal physiological parameters based on the echo signals, and generates a three-dimensional physiological parameter distribution map inside the equipment. The cross-modal data fusion module establishes a spatiotemporal correlation between the external health status map of the device and the internal three-dimensional physiological parameter distribution map of the device, performs time synchronization operation, and generates a confidence score through data integration methods; The dynamic decision-making module converts the confidence score into executable medical instructions and sends the instructions to the medical device via the communication interface. The execution feedback module monitors the state changes of the medical device after executing instructions, calculates the deviation between the state changes and a preset threshold, generates a data acquisition strategy effectiveness index, and dynamically optimizes the data acquisition strategy based on the data acquisition strategy effectiveness index.

[0007] Preferably, the method by which the multimodal data acquisition module acquires multimodal medical information includes: The medical equipment data acquisition terminal is equipped with an infrared thermal imager, a high-resolution camera, and a laser scanner to collect temperature data, visual images, and three-dimensional contour data of the equipment surface, respectively. Simulate visual attention mechanisms and dynamically adjust sensor sampling density; based on the digital model of medical equipment, define the key component areas of the equipment as high attention areas and the remaining areas as low attention areas; The real-time data analysis module identifies abnormal features in multimodal medical information and dynamically adjusts the high-concern area based on these features. High-concern areas use dense sampling, while low-concern areas use sparse sampling. Sensor parameters are adjusted by control signals to obtain the final multimodal medical information.

[0008] Preferably, the method for the multimodal data acquisition module to generate an external health status map of the device includes: Standardized preprocessing is performed on multimodal medical information, and the processed data is combined into multidimensional tensors according to channel dimensions; Construct a deep feature extraction network, which includes convolutional layers and pooling layers; Multimodal medical information is input into a deep feature extraction network, and features are extracted step by step through convolutional layers to generate feature maps at different levels. The pooling layer compresses the feature map size while retaining key features; The output layer applies a fully connected layer to integrate features, fusing feature maps from different levels to generate an external health status map of the device.

[0009] Preferably, the method by which the multimodal data acquisition module divides the external health status map of the device into risk areas includes: The health status score of each region in the external health status map of the calculation device is obtained by weighted fusion of temperature data, visual images and three-dimensional contour data; Preset low-risk and high-risk thresholds, and compare the health status score with the thresholds; if the score is lower than the low-risk threshold, the area is marked as a safe zone; if the score is between the low-risk and high-risk thresholds, the area is marked as a warning zone; if the score is higher than the high-risk threshold, the area is marked as a danger zone. Different marking symbols are used to divide the external health status map of the equipment into regions to distinguish risk levels.

[0010] Preferably, the method for the internal data mapping module to generate a three-dimensional physiological parameter distribution map inside the device includes: Based on the risk zone classification results, an impedance sensor array is deployed on the surface of the equipment casing; a differentiated excitation strategy is adopted for different risk zones, and an alternating current is applied to the inside of the equipment through the impedance sensors to perform global detection; The detection frequency is dynamically adjusted based on the risk level of the equipment's external health status map. High-frequency detection is used in dangerous areas, medium-frequency detection is used in warning areas, and low-frequency detection is used in safe areas. The detection parameters are optimized by applying a frequency adaptive algorithm, a multi-parameter coupled model is introduced, and the model parameters are corrected by an iterative reconstruction algorithm until convergence. Physiological parameters are calculated based on electric field distribution, and a three-dimensional physiological parameter distribution map of the equipment with risk markers is generated.

[0011] Preferably, the method by which the cross-modal data fusion module establishes a spatiotemporal mapping between external and internal data includes: A three-dimensional global coordinate system is defined with a point on the surface of the medical device as the origin; the vision sensor is calibrated using a calibration tool to obtain internal and external parameters; The pixel coordinates of the device's external health status map are transformed to the sensor coordinate system using internal parameters, and then transformed to the global coordinate system using external parameters. The impedance sensor is calibrated to obtain its external parameters, and the three-dimensional physiological parameter distribution map inside the device is transformed into the global coordinate system. Align the spatial coordinates of the external health status map and the internal three-dimensional physiological parameter distribution map of the device in the global coordinate system; The time source of the vision sensor is set as the master clock, and the time source of the impedance sensor is set as the slave clock. Time synchronization is achieved through a network time protocol.

[0012] Preferably, the method by which the cross-modal data fusion module generates confidence scores through data integration includes: A multimodal data map is constructed based on the device's external health status map and the device's internal three-dimensional physiological parameter distribution map in the global coordinate system. A multi-layer network structure employing an attention mechanism is used to construct a fusion network, which includes an input layer, a feature transformation layer, an attention layer, and an output layer. The multimodal data graph is input into the fusion network. Features are projected through the feature transformation layer, cross-modal weights are calculated through the attention layer, the data is integrated through the fusion layer, and the confidence score is generated through the output layer.

[0013] Preferably, the method for constructing a multimodal data graph includes: Extract the feature vector of each risk region from the device's external health status map in the global coordinate system as the external node feature; The feature vector of each parameter concentration region is extracted from the three-dimensional physiological parameter distribution map inside the device in the global coordinate system as the internal node feature; A node set is formed by combining all external and internal nodes; Calculate the Euclidean distance between external nodes; if the distance is less than a set threshold, add a connecting edge. Calculate the Euclidean distance between internal nodes; if the distance is less than a set threshold, add a connecting edge. Find the nearest internal node for each external node and add cross-modal connection edges; construct a multimodal data graph based on the node set and edge set.

[0014] Preferably, the method by which the dynamic decision-making module converts confidence scores into executable medical instructions includes: The internal space of the medical device is divided into voxel units, and each unit records physiological parameter values ​​and confidence scores; adjustable device parameters are listed, including parameter range, energy consumption impact, and historical weights. Establish safety and efficiency goals; set both hard and flexible constraints; Multiple sets of parameter adjustment schemes are generated using optimization algorithms. The degree of achievement of safety and efficiency objectives is evaluated for each scheme, and a safety score and an efficiency score are obtained. The safety score and efficiency score are weighted and summed to obtain a comprehensive score. The scheme with the highest comprehensive score is selected as the optimal scheme. The optimal scheme is then converted into executable medical instructions, including parameter adjustment instructions and control instructions.

[0015] Preferably, the method for the execution feedback module to generate the effectiveness index of the collection strategy includes: Collect medical command parameters executed by medical equipment and simultaneously monitor equipment status change data; calculate the deviation between equipment status changes and preset status change thresholds; The deviation is normalized to generate a standardized data collection strategy effectiveness index; low and high effective thresholds are preset. If the effectiveness index of the standardized data collection strategy is greater than the high effective threshold, the data collection strategy is deemed effective; if the index is between the low and high effective thresholds, the data collection strategy needs to be observed; if the index is less than the low effective threshold, the data collection strategy is deemed ineffective, strategy optimization is initiated, and an early warning message is generated.

[0016] Compared with the prior art, the beneficial effects of the present invention are: Based on the results of external risk zone delineation, internal sensing devices are dynamically deployed for targeted detection. This method first identifies risk areas in the external health status map by analyzing multimodal data from the equipment surface. Then, based on the spatiotemporal information of these areas, it precisely guides the transmission and reception of internal detection signals. This surface-to-interior mapping mechanism changes the traditional model where internal and external detection are independent. External detection results serve as leading information, providing a clear focus and target for internal detection, making the deployment of internal sensing resources and signal transmission highly targeted. This improves the efficiency and accuracy of internal detection, concentrating computational and energy resources on the areas most likely to fail, avoiding the resource consumption and time delays caused by indiscriminate full scanning of the entire equipment, and achieving early and accurate location of potential internal faults.

[0017] By calculating the deviation between state changes and preset thresholds through the feedback module, an effectiveness index for the data acquisition strategy is generated, and the front-end data acquisition strategy is dynamically optimized based on this index. This constructs a closed-loop learning system that optimizes the data source from the decision-making effect. The actual effect after each decision execution is quantitatively evaluated, and the evaluation result directly reflects the effectiveness of the initial data acquisition strategy. The effect is that the entire terminal has self-learning and adaptive capabilities. Data acquisition no longer relies on fixed preset parameters, but can continuously adjust the focus, frequency, and modality weight of multimodal data acquisition based on feedback from actual application scenarios. The system can automatically strengthen the acquisition of data features that contribute positively to decision-making and weaken the acquisition of invalid or redundant data, thereby continuously optimizing data quality in long-term operation, improving the accuracy and reliability of decision-making, and enabling system performance to adaptively evolve over time. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the working principle of the medical device data acquisition terminal described in this invention; Figure 2 A flowchart of a method for generating an external health status diagram of a device; Figure 3 A flowchart for establishing a spatiotemporal mapping method for external and internal data; Figure 4 A diagram for cross-modal data fusion analysis; Figure 5 Optimize the analysis diagram for dynamic decision-making. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides a medical device data acquisition terminal. The system includes: a multimodal data acquisition module responsible for acquiring multimodal data from the surface of the medical device, processing the acquired data through a dynamic focus adjustment mechanism, generating an external health status map of the device, and dividing the health status map into risk areas. An internal data mapping module deploys internal sensing devices based on the risk area division results, emits detection signals to penetrate the device shell, calculates internal physiological parameters based on the echo signals, and generates a three-dimensional physiological parameter distribution map of the device's interior. A cross-modal data fusion module establishes a spatiotemporal correlation between the external health status map and the internal three-dimensional physiological parameter distribution map, performs time synchronization operations, and generates a confidence score through a data integration method. A dynamic decision-making module converts the confidence score into executable medical instructions and sends the instructions to the medical device via a communication interface. An execution feedback module monitors the state changes of the medical device after executing the instructions, calculates the deviation between the state changes and a preset threshold, generates an acquisition strategy effectiveness index, and dynamically optimizes the data acquisition strategy based on the acquisition strategy effectiveness index.

[0021] Example 1: See Figure 2In specific implementations, the multimodal data acquisition module is equipped with an infrared thermal imager, a high-resolution camera, and a laser scanner on the medical device data acquisition terminal. The infrared thermal imager is used to acquire temperature data of the device surface, the high-resolution camera is used to acquire visual images of the device surface, and the laser scanner is used to acquire three-dimensional contour data of the device surface. A simulated visual attention mechanism dynamically adjusts the sensor sampling density. Based on the digital model of the medical device, key component areas are defined as high-interest areas (HIAs), and other areas are defined as low-interest areas (LIAs). A real-time data analysis module identifies abnormal features in the multimodal medical information and dynamically adjusts the HIAs range based on these abnormal features. Dense sampling is used in HIAs, and sparse sampling is used in LIAs. The final multimodal medical information is obtained by adjusting sensor parameters through control signals. These control signal adjustments involve the thermal sensitivity parameters of the infrared thermal imager, the exposure time parameters of the high-resolution camera, and the scanning rate parameters of the laser scanner. In some embodiments, the digital model, containing the device's three-dimensional structure and component functional information, is pre-stored in the terminal memory. The real-time data analysis module uses a sliding window algorithm to analyze the data stream in real time, identifying features such as temperature anomalies, image texture changes, or contour deformations. When dynamically adjusting the area of ​​high concern, if new abnormal features are identified, the boundary of the high concern area is expanded; if the abnormal features disappear, the boundary of the high concern area is narrowed, ensuring that sampling resources are focused on potential problem areas.

[0022] In practical implementation, multimodal medical information undergoes standardized preprocessing, including data normalization, noise reduction, and format unification. The processed data is then combined into multidimensional tensors according to channel dimensions, with each channel corresponding to a different sensor data source. A deep feature extraction network is constructed, comprising convolutional and pooling layers. The convolutional layers use 3D convolutional kernels to extract features step-by-step, while the pooling layers employ max pooling to compress the feature map size. Multimodal medical information is input into the deep feature extraction network. Convolutional layers generate feature maps at different levels, pooling layers retain key features, and a fully connected layer integrates the features at the output layer, fusing the feature maps from different levels to generate an external health status map. The kernel size and stride of the convolutional layers are optimized based on the input data size, the pooling layer window size is set to 2x2 to reduce the spatial resolution of the feature map, and the fully connected layer maps the feature vectors to a health status scoring space. The external health status map represents the health status of each location on the device surface in a grid format, with each grid cell containing a comprehensive health index value. Optionally, the standardization preprocessing stage linearly scales the temperature data to the 0-1 range, performs histogram equalization on the visual image to enhance contrast, and performs smoothing filtering on the 3D contour data to remove noise, ensuring that the multimodal data have consistent scale and quality before fusion.

[0023] The health status score of each region in the device's external health status map is calculated. The score is obtained by weighted fusion of temperature data, visual images, and 3D contour data. In the weighted fusion, the weight of temperature data is set to 0.4, the weight of visual images to 0.35, and the weight of 3D contour data to 0.25. Preset low-risk and high-risk thresholds are used. The health status score is compared to these thresholds. If the score is below the low-risk threshold, the region is marked as a safe zone; if the score is between the low-risk and high-risk thresholds, the region is marked as a warning zone; and if the score is above the high-risk threshold, the region is marked as a danger zone. Different marking symbols are used to divide the device's external health status map into regions to distinguish risk levels. The marking symbols use a color-coding system: safe zones are displayed in green, warning zones in yellow, and danger zones in red. In some embodiments, the low-risk threshold is set to 0.3, the high-risk threshold is set to 0.7, and the health status score is calculated based on the weighted average of the data for each region. The region size is preset to a fixed grid based on the device surface partitioning. It can be understood that after the risk regions are divided, the device's external health status map is output as a two-dimensional image with color overlay for visualization and subsequent processing. Optionally, the weighted fusion formula is a linear weighted sum, ensuring that the score value is within the range of 0-1. Threshold comparison is implemented using if-else logic conditions, and the marker symbol is stored in the additional data channel of the layer.

[0024] In practical implementation, the sensor configuration of the multimodal data acquisition module requires initial calibration. Infrared thermal imaging calibration uses a blackbody radiation source to calibrate temperature measurement accuracy, high-resolution camera calibration uses a calibration board to correct lens distortion, and laser scanner calibration verifies accuracy by measuring objects at known distances. The core of the dynamic focus adjustment mechanism is the real-time data analysis module, which runs lightweight machine learning models, such as decision trees or support vector machines, to quickly classify abnormal patterns in the data stream. The deep feature extraction network is trained using historical medical device data, optimizing network parameters through supervised learning, with the training objective being to minimize health status prediction errors. Risk area segmentation results are fed back to the sensor control loop, dynamically adjusting the sampling strategy; for example, increasing the sampling frequency to 10 times per second in dangerous areas and decreasing it to once per second in safe areas. The entire implementation process is executed in real-time on an embedded processor, employing multi-threaded programming to process the sensor data stream in parallel, ensuring low latency and high reliability. Optionally, the generated external health status map is stored in a terminal database for use by the internal data mapping module; risk area boundaries are represented by vector polygons for easy spatial querying.

[0025] Example 2: In specific implementation, the internal data mapping module, based on the risk area division results of the device's external health status map provided by the multimodal data acquisition module, deploys an impedance sensor array on the surface of the medical device's casing. The impedance sensor array covers the entire surface of the device casing in a rectangular grid arrangement, and each sensor node includes a pair of driving electrodes and a pair of measuring electrodes. A differentiated excitation strategy is adopted for different risk areas. An alternating current is applied to the inside of the device through the impedance sensors for global detection. In the differentiated excitation strategy, the driving current amplitude is set to 10 mA for the dangerous area, 5 mA for the warning area, and 1 mA for the safe area. The detection frequency is dynamically adjusted based on the risk level of the device's external health status map: high-frequency detection is used for dangerous areas, medium-frequency detection for warning areas, and low-frequency detection for safe areas. The high-frequency detection frequency is set to 100 kHz, the medium-frequency detection frequency to 50 kHz, and the low-frequency detection frequency to 10 kHz. A frequency-adaptive algorithm is applied to optimize the detection parameters. This algorithm dynamically adjusts the excitation frequency of each region based on the initial detection results. The algorithm aims to minimize the impact of measurement noise on the calculation of physiological parameters. A multi-parameter coupling model is introduced, simplifying the internal structure of the device as an anisotropic conductive medium. The model parameters include conductivity, dielectric constant, and permeability distribution. An iterative reconstruction algorithm is used to correct the model parameters until convergence. This algorithm uses the Gauss-Newton method to solve the nonlinear inverse problem. Each iteration updates the model parameters to minimize the error between the calculated electric field distribution and the measured data. The convergence criterion is set to a rate of change of less than one-thousandth between two adjacent iterations of the model parameters. Physiological parameters are calculated based on the electric field distribution, generating a three-dimensional physiological parameter distribution map of the device's interior with risk labels. The physiological parameters include tissue impedance and blood perfusion index. The voxel size of the three-dimensional physiological parameter distribution map is set to 1 cubic millimeter, and each voxel contains a physiological parameter value and its corresponding risk level label.

[0026] In practical implementation, the deployment process of the impedance sensor array first reads the risk area division data from the external health status map of the device. This risk area division data includes the spatial coordinate boundaries of safe zones, warning zones, and danger zones. Then, the positions of the impedance sensor nodes are planned according to the spatial distribution of the risk areas. The impedance sensor nodes are arranged at equal intervals on the surface of the device housing. The node spacing is set to 5 mm based on the device size and detection accuracy requirements. Each node is connected to the central processing unit via a flexible circuit board. An alternating current signal is applied to the driving electrode. The frequency of the alternating current signal is dynamically selected according to the risk level. The signal waveform is a sine wave, and the voltage amplitude is adaptively adjusted with the current amplitude to ensure signal stability. The measuring electrode detects the voltage response. The voltage response signal is amplified and filtered before being digitized by an analog-to-digital converter. The digitized data is stored in a buffer for subsequent analysis. During global detection, all impedance sensor nodes work synchronously. The central processing unit coordinates the excitation and measurement timing of each node to avoid signal interference. The detection data is uploaded to the fusion processing module in real time. In some embodiments, the driving electrodes of the impedance sensor array are made of silver-silver chloride material to reduce polarization effects, the measuring electrodes use high-impedance input amplifiers to improve measurement accuracy, and the signal transmission lines use shielded twisted-pair cables to suppress electromagnetic interference. It is understandable that the differentiated excitation strategy adjusts the driving current amplitude through a digital potentiometer. The current amplitude setting is based on an empirical value of risk level to ensure that high-risk areas receive higher detection energy to penetrate deep tissues.

[0027] The implementation of the frequency adaptive algorithm begins with initial probes, which collect baseline data using default frequency settings and then analyze the signal-to-noise ratio (SNR) and spectral characteristics of the baseline data. For each risk region, the algorithm calculates measurement quality indicators (MCIs) at the current frequency, such as the signal amplitude variation coefficient and noise power. If the MCIs fall below a threshold, frequency adjustment is triggered. The frequency adjustment step size is set according to the risk level: 10 kHz for dangerous regions, 5 kHz for warning regions, and 2 kHz for safe regions. The adjustment direction is towards maximizing the SNR. The construction of the multi-parameter coupled model is based on prior knowledge of the device's internal structure, derived from a digital model of the medical device. This digital model provides the geometry and material properties of the internal components. Model parameters are initialized using standard biological tissue electrical property values: conductivity is set to 0.5 Siemens per meter, dielectric constant to 80, and permeability to 1. These parameters are then optimized through inversion using probe data. The Gauss-Newton method of the iterative reconstruction algorithm requires calculating the Jacobian matrix, which is approximated using the finite difference method. Each iteration solves the linearized normal equations to update the parameter vector. The convergence judgment module monitors the rate of change of parameters. If the rate of change is less than one-thousandth for three consecutive iterations, convergence is determined, and the final model parameters are output. Optionally, the frequency adaptive algorithm integrates machine learning components, uses historical data to train the frequency selection model, improving adjustment efficiency, and the multi-parameter coupled model supports multi-frequency band data fusion, improving the accuracy of parameter estimation.

[0028] When calculating physiological parameters based on electric field distribution, a converged multi-parameter coupled model is used to simulate the current distribution inside the device. The Poisson equation is solved to obtain the electric potential field, and then the electric field intensity distribution is calculated. Tissue impedance is derived using Ohm's law, i.e., impedance equals the ratio of voltage to current. Local impedance values ​​are calculated based on the voltage and driving current of the measuring electrode pair. The blood perfusion index is estimated based on the impedance variation with frequency, and perfusion parameters are obtained by fitting an equivalent circuit model using multi-frequency probe data. The process of generating a three-dimensional physiological parameter distribution map inside the device discretizes the internal space into a voxel grid, with each voxel fixed at 1 cubic millimeter. Each voxel is assigned a calculated physiological parameter value. Risk markers are mapped to corresponding spatial locations based on the risk level of the external health status map: dangerous areas are marked in red, warning areas in yellow, and safe areas in green. The marking information is stored as an additional attribute of the voxel. The three-dimensional distribution map is visualized using volume rendering technology, supporting rotation and scaling operations for easy observation of the internal state. In some embodiments, the electric field distribution calculation employs the finite element method to discretize the solution domain, with the mesh generation consistent with the voxel mesh. The solver uses the conjugate gradient method to improve computational efficiency. It is understood that the physiological parameter calculation module verifies the rationality of the data; if parameter values ​​exceed the normal range of biological tissue, they are marked as abnormal and require re-detection. The three-dimensional distribution map data is exported in a standard format for use by the cross-modal data fusion module. Optionally, the risk-marked internal three-dimensional physiological parameter distribution map supports time-series analysis, recording historical trends for long-term monitoring, and data compression algorithms reduce storage space usage.

[0029] Example 3: See Figure 3In practical implementation, the cross-modal data fusion module defines a three-dimensional global coordinate system with a point on the surface of the medical device as the origin. This three-dimensional global coordinate system is a right-handed Cartesian coordinate system, with the X-axis along the major axis of the device, the Y-axis along the minor axis, and the Z-axis perpendicular to the device surface outwards. The vision sensor is calibrated using a black and white checkerboard calibration tool to obtain internal and external parameters. The checkerboard squares are precisely 5 mm × 5 mm in size, and the vision sensor captures at least 20 images of the calibration board from different angles. Internal parameters include focal length, principal point coordinates, and distortion coefficients. External parameters include the rotation matrix and translation vector of the vision sensor relative to the three-dimensional global coordinate system. The pixel coordinates of the device's external health status image are transformed to the sensor coordinate system using the internal parameters, employing a pinhole camera model. Then, the external parameters are used to transform the image back to the global coordinate system. The coordinate transformation is achieved using homogeneous coordinates and a transformation matrix. The impedance sensor is calibrated to obtain its external parameters, including the position and orientation matrix of the sensor array center point in the three-dimensional global coordinate system. The calibration method uses a laser tracker to measure the spatial coordinates. The internal 3D physiological parameter distribution map of the device is transformed to a global coordinate system. During the transformation, a rigid body transformation is applied to the center point of each voxel in the distribution map. The spatial coordinates of the external health status map and the internal 3D physiological parameter distribution map are aligned in the global coordinate system, with an alignment accuracy requirement of sub-millimeter level. The time source of the visual sensor is set as the master clock, and the time source of the impedance sensor is set as the slave clock. Time synchronization is achieved through a network time protocol, using Precision Time Protocol version 2, with a synchronization accuracy of microsecond level.

[0030] In practical implementation, feature vectors for each risk region are extracted from the external health status map of the device in the global coordinate system as external node features. Risk regions are divided into safe zones, warning zones, and danger zones based on the segmentation results. The feature vector extracted for each region has an 8-dimensional dimension. The feature vector includes average temperature, temperature gradient magnitude, image texture energy, image texture entropy, average surface curvature, surface Gaussian curvature, risk level code, and region area percentage. Feature vectors for each parameter set region are extracted from the internal 3D physiological parameter distribution map of the device in the global coordinate system as internal node features. Parameter set regions are identified using a clustering algorithm. The feature vector extracted for each region has a 6-dimensional dimension. The feature vector includes the physiological parameter mean, physiological parameter variance, spatial gradient magnitude, region volume, distance to the nearest danger zone, and risk level code. All external and internal nodes are combined to form a node set, which is stored using a linked list data structure. Each node records its spatial coordinates, timestamp, and feature vector. The Euclidean distance between external nodes is calculated based on the node's spatial coordinates in the 3D global coordinate system. If the distance is less than a set threshold, a connecting edge is added. The set threshold is determined based on 1.5 times the average node density. Calculate the Euclidean distance between internal nodes. If the distance is less than a set threshold, add a connecting edge. The internal node distance threshold is set independently to accommodate different spatial distribution densities. For each external node, find the nearest internal node and add a cross-modal connecting edge. The weight of the cross-modal connecting edge is inversely proportional to the distance between nodes. Construct a multimodal data graph based on the node set and edge set. The multimodal data graph adopts an attribute graph model, where nodes contain feature attributes and edges contain connection weights and type attributes.

[0031] In some embodiments, risk region feature extraction of the external health status map uses a sliding window method, with the window size adaptively adjusted according to the region size. Each window extracts statistical and textural features. For the identification of parameter concentration regions in the internal 3D physiological parameter distribution map, a density-based clustering algorithm is used, with a cluster radius of 3 mm and a minimum number of voxels of 5. The threshold for connection edges between external nodes is set to 5 mm, and the threshold for connection edges between internal nodes is set to 4 mm. These thresholds ensure that spatially adjacent nodes establish connections. The weight calculation for cross-modal connection edges uses a Gaussian kernel function, expressed by the following formula: ; in; Represents external nodes With internal nodes Connection weights between them Represents external nodes With internal nodes The Euclidean distance between them This indicates the rate at which the bandwidth parameter controls the weight decays with distance.

[0032] During time synchronization, the visual sensor acts as the master clock, periodically broadcasting synchronization messages containing timestamps and clock identifiers. The impedance sensor acts as the slave clock, receiving synchronization messages and adjusting its local clock. Clock skew calculation employs a delay request-response mechanism: the slave clock sends a delay request message, the master clock returns the received timestamp, and the slave clock calculates the clock skew based on the round-trip time. The network time protocol automatically selects the master clock using an optimal master clock algorithm; if the visual sensor fails, the impedance sensor switches to master clock. During coordinate transformation, the pixel coordinates of the device's external health status map are first distorted using internal parameters, then mapped to the sensor coordinate system via inverse perspective transformation, and finally transformed to the global coordinate system using an external transformation matrix. The global coordinates of each voxel center point in the device's internal 3D physiological parameter distribution map are obtained by multiplying by the impedance sensor's external transformation matrix. Coordinate alignment is finely adjusted using an iterative nearest-point algorithm to minimize distance errors between point clouds. In some embodiments, the visual sensor's internal parameter calibration uses the Zhang Zhengyou calibration method to solve for the camera's intrinsic parameter matrix and distortion coefficients, while the impedance sensor's external parameter calibration uses least-squares fitting of the sensor array plane equations. Understandably, once the spatiotemporal correlation is established, the multimodal data graph contains complete spatiotemporal information, with each node having global coordinates and a precise timestamp, and edge connections reflecting spatial proximity and modal correlation. Optionally, the multimodal data graph supports dynamic updates, incrementally updating nodes and edges when new data arrives, and automatically archiving and saving outdated data.

[0033] See Figure 4 This paper demonstrates the core functionality of the cross-modal data fusion module. The chart, presented in a three-dimensional spatial distribution, visually illustrates the spatiotemporal correlation between external monitoring data and internal sensor data of medical equipment. The chart includes two types of key data nodes: external monitoring nodes and internal sensor nodes. External nodes are generated based on multimodal acquisition data from the device surface, incorporating various features such as temperature, visual texture, and surface curvature. Node colors reflect the risk level assessment results of different areas. Internal nodes originate from the detection data of the impedance sensor array, providing spatial distribution information of the device's internal physiological parameters. Node colors indicate the magnitude changes of parameter values. Connecting lines between nodes establish a spatial mapping relationship between the cross-modal data. This correlation is achieved based on a unified global coordinate system and precise time synchronization. Through this spatial correlation network, the system can accurately correlate externally observed abnormal features with changes in internal physiological parameters, providing a reliable spatiotemporal benchmark for subsequent confidence scoring. The spatial distribution patterns shown in the chart reveal the complex characteristics of the device's health status. Patterns such as the clustering of high-risk areas and the spatial correlation of parameter anomalies provide important evidence for medical decision-making. This cross-modal fusion analysis method significantly improves the accuracy and reliability of medical equipment status monitoring.

[0034] Example 4: In specific implementation, the dynamic decision-making module first divides the internal space of the medical device into voxel units. The size of each voxel unit is set to 1 cubic millimeter according to the resolution requirements of the internal structure of the device. Each voxel unit records physiological parameter values ​​and confidence scores. The physiological parameter values ​​come from the three-dimensional physiological parameter distribution map of the device's internal structure generated by the internal data mapping module, and the confidence scores come from the output of the cross-modal data fusion module. Adjustable device parameters are listed, including infusion rate, drug concentration, and electrical stimulation intensity. Each parameter has a parameter range, energy consumption impact, and historical weight. The parameter range defines the minimum and maximum values ​​that the parameter can be adjusted to. The energy consumption impact is calculated based on the device power consumption model and represents the degree of influence of parameter adjustment on the device's energy consumption. The historical weight is determined based on the adjustment frequency and effect of the parameter over a period of time, reflecting the importance or stability of the parameter. For a clear display of parameter attributes, refer to Table 1: Table 1: Adjustable Equipment Parameter Table In practical implementation, safety and efficiency goals are established. The safety goal is defined as maintaining the stability of physiological parameters within a safe range and preventing abnormal states during equipment operation. The efficiency goal is defined as achieving the best therapeutic effect with minimal energy consumption and optimizing resource utilization. Hard constraints and flexible constraints are set. Hard constraints include absolute safety boundaries for physiological parameters, such as maintaining heart rate within the range of 60-100 beats / minute. Flexible constraints allow parameters to fluctuate within a controllable range, such as allowing a temperature deviation of ±1°C. Multiple parameter adjustment schemes are generated using an optimization algorithm. The optimization algorithm employs a multi-objective particle swarm optimization (PSO) algorithm with a population size of 50 particles and 100 iterations. Each particle position represents a set of parameter adjustment schemes, and the particle velocity represents the direction and magnitude of parameter adjustment. The degree of achievement of the safety and efficiency goals is evaluated for each scheme, resulting in a safety score and an efficiency score. The safety score considers the degree and duration of parameter deviation from the safe range, while the efficiency score comprehensively evaluates the therapeutic effect and energy consumption.

[0035] In some embodiments, voxel unit partitioning employs an octree data structure, with unit size adaptively adjusted based on local feature complexity. Each voxel stores a historical sequence of physiological parameters from the last 10 samples, which is used to analyze trends and predict changes. Hard constraints are embedded in the optimization algorithm through constraint processing techniques, such as using penalty functions to transform constraint violations into penalty terms in the objective function. Flexible constraints are processed using fuzzy logic, allowing for a certain degree of relaxation. The particle update formula of the multi-objective particle swarm optimization algorithm considers both individual optimal and global optimal positions. Inertia weights decrease linearly to balance exploration and utilization, and crossover and mutation operations introduce diversity to prevent premature convergence. Safety score calculation uses a piecewise function, with the highest score when parameters are within the safety range and the lower the score as deviation increases. Efficiency score is based on the Pareto optimality concept, balancing treatment effectiveness and energy consumption ratio.

[0036] Optionally, a weighted summation of the safety score and efficiency score yields a comprehensive score. The safety score is weighted at 0.6, and the efficiency score at 0.4. These weights can be adjusted based on clinical needs. The weighted summation formula is a linear combination. The scheme with the highest comprehensive score is selected as the optimal scheme. The selection process uses a sorting and selection mechanism; if multiple schemes have the same comprehensive score, the scheme with the higher safety score is prioritized. The optimal scheme is then converted into executable medical instructions, including parameter adjustment instructions and control instructions. Parameter adjustment instructions specify the parameter identifier, target value, adjustment rate, and effective time. Control instructions trigger the device to perform specific operations, such as starting or stopping treatment. The instructions are encapsulated in JSON format, including a timestamp and device identifier, ensuring traceability and verification.

[0037] In practical implementation, the dynamic decision-making module connects to the medical equipment control system via a communication interface that uses standard protocols such as HL7 or DICOM to ensure reliable command transmission. The parameter adjustment scheme evaluation process is performed in real time, processing new data every second. The comprehensive score calculation uses a normalization method to ensure comparability within the 0-1 range. After the optimal scheme is selected, the effect of the command is simulated and verified through a digital twin system. The simulation results are fed back to the decision-making module to revise the scheme. Only after successful verification is the command issued to the actual equipment. Optionally, the historical weight update mechanism is based on reinforcement learning. After each adjustment, the weights are adjusted according to the effect, and the effect is evaluated through the effectiveness index of the collection strategy collected by the execution feedback module, achieving adaptive optimization.

[0038] See Figure 5This diagram illustrates the optimization and evaluation process of the dynamic decision-making module. The chart, using a two-dimensional evaluation coordinate system, clearly presents the trade-off between safety and efficiency goals for different parameter adjustment schemes. The horizontal axis represents the safety performance score of the scheme, reflecting the impact of parameter adjustments on equipment operational stability and patient safety. The vertical axis represents the treatment efficiency score, evaluating the scheme's performance in resource utilization and treatment effectiveness. Each data point represents a specific parameter adjustment combination scheme, and the color depth of the point indicates the scheme's overall evaluation score. The decision boundary line in the diagram defines the balance between safety and efficiency, helping to identify the ideal operating area under different clinical priorities. The selection of the optimal scheme is based on a weighted comprehensive evaluation method, comprehensively considering the dual needs of safety and efficiency, ensuring the best treatment effect while guaranteeing basic safety. This visual evaluation method provides clinical operators with intuitive decision support, enabling rapid identification of the optimal parameter configuration suitable for the current patient condition and equipment operating conditions. Through real-time scheme optimization and evaluation, the system can dynamically adjust the operating parameters of medical equipment to achieve personalized and precise medical treatment.

[0039] Example 5: In specific implementation, the execution feedback module collects medical instruction parameters executed by the medical device. These parameters include instruction type, parameter settings, execution timestamp, and expected effect. These parameters are obtained from the instruction log issued by the dynamic decision-making module. Device status change data is monitored synchronously. This data is continuously acquired through a multimodal data acquisition module and an internal data mapping module. The monitoring cycle is strictly synchronized with the instruction execution cycle. Each monitoring cycle collects a status data sequence of 100 time points. This data sequence includes updated data from the device's external health status map and updated data from the device's internal three-dimensional physiological parameter distribution map. The deviation between the device status change and a preset status change threshold is calculated using the root mean square error method. The difference between the actual status change curve and the preset status change threshold curve is compared. The actual status change curve is generated by fitting the monitored status data sequence, while the preset status change threshold curve is set based on the ideal operating model of the medical device.

[0040] The deviation is normalized to generate a standardized acquisition strategy effectiveness index. Normalization maps the calculated root mean square error value to the 0-1 range using a linear scaling method. The standardized acquisition strategy effectiveness index is calculated as 1 minus the normalized deviation value. Preset low and high effectiveness thresholds are used: a low effectiveness threshold of 0.7 and a high effectiveness threshold of 0.9. If the standardized acquisition strategy effectiveness index is greater than the high effectiveness threshold, the acquisition strategy is considered effective. If the index is between the low and high effectiveness thresholds, the acquisition strategy needs observation. If the index is less than the low effectiveness threshold, the acquisition strategy is considered ineffective, strategy optimization is initiated, and an early warning message is generated. The strategy optimization module re-evaluates the sensor configuration and sampling parameters of the multimodal data acquisition module, adjusts the sensitivity settings of the dynamic focus adjustment mechanism, and reports the early warning message to the central monitoring system via the medical device data bus. The message includes the device identifier, the abnormal time point, the effectiveness index value, and suggested handling measures.

[0041] In some embodiments, medical instruction parameter records are stored using a structured database. Each record contains a unique instruction identifier, a parameter list, and a timestamp index. Device status change data is stored in a circular buffer to support real-time streaming processing. Preset status change threshold curves are personalized based on device type and treatment goals; for example, for infusion pump devices, the threshold curve is set as the ideal trajectory of drug delivery volume. The deviation calculation module employs a sliding window technique, with the window size matched to the monitoring period. The deviation value is updated each time new data arrives, ensuring real-time performance. The scaling factor used in normalization processing is dynamically adjusted based on the maximum and minimum values ​​of historical deviation data to avoid extreme values ​​affecting index calculation. It can be understood that the standardized acquisition strategy effectiveness index reflects the degree to which the current data acquisition strategy accurately captures device status changes; a higher index indicates a better match between the acquisition strategy and the actual device status.

[0042] Optionally, the acquisition strategy requires the system to maintain the current acquisition parameters while increasing the monitoring frequency during observation, for example, shortening the monitoring cycle from 1 second to 0.5 seconds, and recording more detailed status data for subsequent analysis. If the acquisition strategy is ineffective, a strategy optimization process is triggered. This process first analyzes the root cause of the ineffectiveness, which may include sensor drift, environmental interference, or changes in the device's dynamic characteristics. The optimization process uses A / B testing, running multiple sets of acquisition parameter configurations in parallel, and selecting the configuration with the highest effectiveness index as the new strategy. After an early warning is generated, the system automatically executes a preset emergency protocol, such as pausing non-critical data acquisition tasks or switching to a backup sensor mode. The optimized acquisition strategy is then updated to each data acquisition module after simulation verification. The simulation verification uses a digital twin system to reproduce the device's operating scenario, ensuring the safety and effectiveness of the new strategy.

[0043] In practical implementation, a specific example is used. Assume the medical device is an intelligent infusion pump, and the dynamic decision-making module issues the medical instruction "adjust the infusion rate from 50 mL / h to 60 mL / h". The execution feedback module collects the instruction parameters, including the instruction type as parameter adjustment, the parameter setting value as 60 mL / h, and the execution timestamp as T0. Simultaneously, the status change data of the infusion pump is monitored. The multimodal data acquisition module monitors the surface temperature distribution and vibration signals of the pump body, while the internal data mapping module monitors the drug flow rate and tubing pressure. Within a monitoring period from T0 to T0+5 minutes, status data is collected at 100 time points, and the actual monitored flow rate gradually changes from 50 mL / h to 59 mL / h. The preset status change threshold curve is the ideal trajectory of linear growth from 50 mL / h to 60 mL / h. The root mean square error between the actual flow rate curve and the ideal trajectory is calculated, assumed to be 2 mL / h. Normalization maps the error value to the 0-1 range. Assuming the mapped value is 0.2, the effectiveness index of the standardized acquisition strategy is 1-0.2=0.8. Since 0.8 falls between the low effective threshold of 0.7 and the high effective threshold of 0.9, the system determines that the current acquisition strategy needs to be observed, automatically increases the monitoring frequency to twice per second, and marks the data for this period for subsequent analysis by engineers.

[0044] Optionally, the strategy optimization process might adjust the sampling frequency of the laser scanner in the multimodal data acquisition module for this example, increasing it from 10Hz to 20Hz to more accurately capture mechanical vibration details. The warning message reads, "Device number Pump-001 is lagging in its flow rate adjustment response at time T0, with an effectiveness index of 0.8; it is recommended to check the drive mechanism." This information is sent to the nurse station monitoring center via the data bus. It can be understood that the execution feedback module uses this closed-loop control mechanism to ensure that the data acquisition strategy always matches the actual operating state of the equipment, providing continuous assurance for the safe and reliable operation of medical equipment. After the optimization strategy is updated, the system continues to monitor the effectiveness index of the next round of instruction execution, achieving a continuous self-optimization cycle.

[0045] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A medical device data acquisition terminal, characterized in that, The terminal includes a multimodal data acquisition module, an internal data mapping module, a cross-modal data fusion module, a dynamic decision-making module, and an execution feedback module; The multimodal data acquisition module is responsible for collecting multimodal data from the surface of the medical device, processing the collected data through a dynamic focus adjustment mechanism, generating an external health status map of the device, and dividing the health status map into risk areas. Based on the risk area division results, the internal data mapping module deploys internal sensing devices, emits detection signals to penetrate the equipment shell, calculates internal physiological parameters based on the echo signals, and generates a three-dimensional physiological parameter distribution map inside the equipment. The cross-modal data fusion module establishes a spatiotemporal correlation between the external health status map of the device and the internal three-dimensional physiological parameter distribution map of the device, performs time synchronization operation, and generates a confidence score through data integration methods; The dynamic decision-making module converts the confidence score into executable medical instructions and sends the instructions to the medical device via the communication interface. The execution feedback module monitors the state changes of the medical device after executing instructions, calculates the deviation between the state changes and a preset threshold, generates a data acquisition strategy effectiveness index, and dynamically optimizes the data acquisition strategy based on the data acquisition strategy effectiveness index.

2. The medical device data acquisition terminal according to claim 1, characterized in that, The method for the multimodal data acquisition module to acquire multimodal medical information includes: The medical equipment data acquisition terminal is equipped with an infrared thermal imager, a high-resolution camera, and a laser scanner to collect temperature data, visual images, and three-dimensional contour data of the equipment surface, respectively. Simulate visual attention mechanisms and dynamically adjust sensor sampling density; based on the digital model of medical equipment, define the key component areas of the equipment as high attention areas and the remaining areas as low attention areas; The real-time data analysis module identifies abnormal features in multimodal medical information and dynamically adjusts the high-concern area based on these features. High-concern areas use dense sampling, while low-concern areas use sparse sampling. Sensor parameters are adjusted by control signals to obtain the final multimodal medical information.

3. The medical device data acquisition terminal according to claim 2, characterized in that, The method for generating an external health status map of the device by the multimodal data acquisition module includes: Standardized preprocessing is performed on multimodal medical information, and the processed data is combined into multidimensional tensors according to channel dimensions; Construct a deep feature extraction network, which includes convolutional layers and pooling layers; Multimodal medical information is input into a deep feature extraction network, and features are extracted step by step through convolutional layers to generate feature maps at different levels. The pooling layer compresses the feature map size while retaining key features; The output layer applies a fully connected layer to integrate features, fusing feature maps from different levels to generate an external health status map of the device.

4. The medical device data acquisition terminal according to claim 3, characterized in that, The method for the multimodal data acquisition module to divide the external health status map of the device into risk areas includes: The health status score of each region in the external health status map of the calculation device is obtained by weighted fusion of temperature data, visual images and three-dimensional contour data; Preset low-risk and high-risk thresholds, and compare the health status score with the thresholds; if the score is lower than the low-risk threshold, the area is marked as a safe zone; if the score is between the low-risk and high-risk thresholds, the area is marked as a warning zone; if the score is higher than the high-risk threshold, the area is marked as a danger zone. Different marking symbols are used to divide the external health status map of the device into regions to distinguish risk levels.

5. The medical device data acquisition terminal according to claim 4, characterized in that, The method for generating a three-dimensional physiological parameter distribution map inside the device by the internal data mapping module includes: Based on the risk zone classification results, an impedance sensor array is deployed on the surface of the equipment casing; a differentiated excitation strategy is adopted for different risk zones, and an alternating current is applied to the inside of the equipment through the impedance sensors to perform global detection; The detection frequency is dynamically adjusted based on the risk level of the equipment's external health status map. High-frequency detection is used in dangerous areas, medium-frequency detection is used in warning areas, and low-frequency detection is used in safe areas. The detection parameters are optimized by applying a frequency adaptive algorithm, a multi-parameter coupled model is introduced, and the model parameters are corrected by an iterative reconstruction algorithm until convergence. Physiological parameters are calculated based on electric field distribution, and a three-dimensional physiological parameter distribution map of the equipment with risk markers is generated.

6. The medical device data acquisition terminal according to claim 5, characterized in that, The method for establishing the spatiotemporal mapping between external and internal data by the cross-modal data fusion module includes: A three-dimensional global coordinate system is defined with a point on the surface of the medical device as the origin; the vision sensor is calibrated using a calibration tool to obtain internal and external parameters; The pixel coordinates of the device's external health status map are transformed to the sensor coordinate system using internal parameters, and then transformed to the global coordinate system using external parameters. The impedance sensor is calibrated to obtain its external parameters, and the three-dimensional physiological parameter distribution map inside the device is transformed into the global coordinate system. Align the spatial coordinates of the external health status map and the internal three-dimensional physiological parameter distribution map of the device in the global coordinate system; The time source of the visual sensor is set as the master clock, and the time source of the impedance sensor is set as the slave clock. Time synchronization is achieved through a network time protocol.

7. The medical device data acquisition terminal according to claim 6, characterized in that, The cross-modal data fusion module generates confidence scores through data integration methods, including: A multimodal data map is constructed based on the device's external health status map and the device's internal three-dimensional physiological parameter distribution map in the global coordinate system. A multi-layer network structure employing an attention mechanism is used to construct a fusion network, which includes an input layer, a feature transformation layer, an attention layer, and an output layer. The multimodal data graph is input into the fusion network. Features are projected through the feature transformation layer, cross-modal weights are calculated through the attention layer, the data is integrated through the fusion layer, and the confidence score is generated through the output layer.

8. The medical device data acquisition terminal according to claim 7, characterized in that, The method for constructing a multimodal data graph includes: Extract the feature vector of each risk region from the device's external health status map in the global coordinate system as the external node feature; The feature vector of each parameter concentration region is extracted from the three-dimensional physiological parameter distribution map inside the device in the global coordinate system as the internal node feature; A node set is formed by combining all external and internal nodes; Calculate the Euclidean distance between external nodes; if the distance is less than a set threshold, add a connecting edge. Calculate the Euclidean distance between internal nodes; if the distance is less than a set threshold, add a connecting edge. Find the nearest internal node for each external node and add cross-modal connection edges; construct a multimodal data graph based on the node set and edge set.

9. The medical device data acquisition terminal according to claim 8, characterized in that, The method by which the dynamic decision-making module converts confidence scores into executable medical instructions includes: The internal space of the medical device is divided into voxel units, and each unit records physiological parameter values ​​and confidence scores; adjustable device parameters are listed, including parameter range, energy consumption impact, and historical weights. Establish safety and efficiency goals; set both hard and flexible constraints; Multiple sets of parameter adjustment schemes are generated using optimization algorithms. The degree of achievement of safety and efficiency objectives is evaluated for each scheme, and a safety score and an efficiency score are obtained. The safety score and efficiency score are weighted and summed to obtain a comprehensive score. The scheme with the highest comprehensive score is selected as the optimal scheme. The optimal scheme is then converted into executable medical instructions, including parameter adjustment instructions and control instructions.

10. The medical device data acquisition terminal according to claim 9, characterized in that, The method for generating the collection strategy effectiveness index by the execution feedback module includes: Collect medical command parameters executed by medical equipment and simultaneously monitor equipment status change data; calculate the deviation between equipment status changes and preset status change thresholds; The deviation is normalized to generate a standardized data collection strategy effectiveness index; low and high effective thresholds are preset. If the effectiveness index of the standardized data collection strategy is greater than the high effective threshold, the data collection strategy is deemed effective; if the index is between the low and high effective thresholds, the data collection strategy needs to be observed; if the index is less than the low effective threshold, the data collection strategy is deemed ineffective, strategy optimization is initiated, and an early warning message is generated.

Images