Soft tissue puncture path planning and monitoring method and system based on electrical impedance imaging
By combining multi-frequency domain electrical impedance tomography with ultrasound and magnetic resonance imaging, along with reinforcement learning and mechanical sensors, the problems of imaging resolution and intelligence in soft tissue puncture path planning have been solved, achieving efficient and safe puncture path planning and monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN YINGTAI LIANKANG MEDICAL SCI & TECH CO LTD
- Filing Date
- 2025-05-07
- Publication Date
- 2026-06-26
AI Technical Summary
Existing EIT-based soft tissue puncture path planning methods suffer from low imaging resolution, poor path planning flexibility, and insufficient intelligence, which affect surgical success rate and patient safety.
By combining multi-frequency domain electrical impedance tomography with ultrasound and magnetic resonance imaging, an image reconstruction model was established and a prior auxiliary network was trained. Reinforcement learning was used for path planning, and a mechanical sensor was installed at the tip of the puncture needle for intraoperative monitoring and dynamic adjustment of the path.
It improves imaging resolution and the intelligence of path planning, significantly reduces tissue damage rate, shortens operation time, and provides standardized and intelligent technical support.
Smart Images
Figure CN120168105B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of medicine and computer technology, specifically to a method and system for planning and monitoring soft tissue puncture paths based on electrical impedance imaging. Background Technology
[0002] Electrical impedance tomography (ETM) is a functional imaging technique based on differences in the electrical conductivity of biological tissues. Its basic principle involves injecting a weak alternating current into an array of electrodes on the body surface, measuring changes in the boundary voltage, and then reconstructing the conductivity distribution within the tissue using an inverse problem-solving algorithm. Because different soft tissues exhibit significant differences in conductivity, ETM offers a non-invasive, real-time, and radiation-free imaging method, making it valuable for applications in medical monitoring and intraoperative navigation.
[0003] In recent years, electrical impedance tomography (EIT) technology has been introduced into the field of soft tissue puncture path planning to assist surgical navigation. However, existing EIT-based puncture path planning methods still suffer from key problems such as low imaging resolution, poor path planning flexibility, and insufficient intelligence. Therefore, there is an urgent need to develop high-precision, adaptive, and intelligent EIT-guided puncture path planning methods to improve surgical success rates and patient safety. Summary of the Invention
[0004] To address the shortcomings of existing detection methods, a soft tissue puncture path planning and monitoring method based on electrical impedance imaging is proposed, and a soft tissue puncture path planning and monitoring system based on electrical impedance imaging is built.
[0005] A method for soft tissue puncture path planning and monitoring based on electrical impedance imaging is characterized by:
[0006] Step S1: Before puncture, soft tissue information in different frequency domains is obtained through multi-frequency domain electrical impedance imaging, while richer structural information is obtained through ultrasound imaging and magnetic resonance imaging.
[0007] Step S2: Establish an image reconstruction model, fuse multi-frequency domain electrical impedance imaging information, ultrasound imaging information and nuclear magnetic resonance imaging information, train a prior auxiliary network, and improve resolution with the help of supplementary information.
[0008] Step S3: Preoperative path planning modeling. The preoperative path planning model is trained based on reinforcement learning. The needle tip position and its relative position to the target point during the puncture process are used as state information. An appropriate reward function is defined, such as giving a positive reward when the needle tip approaches the target point or successfully reaches it, and giving a negative reward when colliding with obstacles or deviating from the path. Through continuous interaction with the environment and trial-and-error learning, the agent can gradually master the optimal action strategy in different states.
[0009] Step S4: Install a mechanical sensor at the tip of the puncture needle to monitor the resistance of the puncture path in real time during the operation. Analyze the condition of the surrounding soft tissue through the needle tip sensor and dynamically adjust the path.
[0010] A needle-tip sensor data N is introduced for real-time monitoring of intraoperative path rationality and anomaly handling; the local reward function is optimized as follows:
[0011] Localr = Localr + R - Resistance Anomaly
[0012] R_resistance anomaly degree = -d·(Ft-Fsafe)
[0013] This value measures whether the resistance Ft at the current time t exceeds the expected safety threshold Fsafe. A penalty is imposed according to the proportion of the excess, and d is the penalty weight of the resistance anomaly.
[0014] Step S5: Iterate through steps S3-S4 until the model converges, thus achieving the optimal route design.
[0015] This invention also relates to a soft tissue puncture path planning and monitoring system based on electrical impedance imaging, the system being used to execute the soft tissue puncture path planning and monitoring method based on electrical impedance imaging as described above, the system comprising:
[0016] This invention also relates to a soft tissue puncture path planning and monitoring system based on electrical impedance imaging, the system being used to execute the soft tissue puncture path planning and monitoring method based on electrical impedance imaging as described above, the system comprising:
[0017] The data acquisition module is used to inject multi-frequency alternating current through a body surface electrode array to obtain conductivity distribution data in different frequency domains; at the same time, it acquires ultrasound imaging data and nuclear magnetic resonance imaging data to provide a multimodal information basis for subsequent path planning, and ensures the spatiotemporal consistency and availability of each modality of data through data preprocessing.
[0018] The image reconstruction module is used to construct a priori auxiliary network, which extracts key anatomical features such as blood vessels and tumors based on ultrasound and MRI data, and generates a probability map matrix. At the same time, it guides the EIT reconstruction network to fuse multi-frequency domain electrical impedance imaging information with the prior feature matrix and uses weighted and correlation calculation strategies to improve the resolution and microstructure recognition capability of electrical impedance imaging.
[0019] The route planning module is used to mark high-risk areas and target tissues based on image reconstruction results through threshold segmentation and morphological operations; construct a reinforcement learning model, define a state space including impedance features and needle tip position, and an action space including forward direction angle and propulsion force; design global and local reward functions, and combine factors such as patient age, target proximity, and distance to danger zone, and iteratively optimize through the Value function to plan a safe and efficient puncture path;
[0020] The intraoperative monitoring module is used to collect resistance, resistance change rate and lateral force data in real time using the mechanical sensor at the tip of the puncture needle; integrate the sensor data into the local reward function to dynamically adjust the puncture path; analyze data trends, predict potential risks, and realize real-time dynamic monitoring and path correction of the puncture process.
[0021] The abnormal alarm module is used to monitor abnormal situations such as excessive resistance, path deviation, and sensor failure during the puncture process in real time. When an abnormality occurs, it will issue an alarm through sound and light, pop-up window, etc., automatically suspend the puncture operation or execute safety protection actions, and provide medical staff with handling suggestions to ensure surgical safety.
[0022] The present invention also discloses a non-volatile storage medium, characterized in that the non-volatile storage medium includes a stored program, wherein the program, when running, controls the device where the non-volatile storage medium is located to execute the above-described method.
[0023] The present invention also discloses an electronic device, characterized in that it comprises a processor and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein the computer-readable instructions execute the method described above.
[0024] Beneficial effects
[0025] This invention relates to a soft tissue puncture path planning and monitoring system and method based on electrical impedance imaging. By fusing multimodal imaging techniques including electrical impedance, ultrasound, and MRI, it utilizes prior auxiliary networks and optimization algorithms to improve imaging resolution. A reinforcement learning-based intelligent path planning model is constructed, balancing efficiency and safety through global and local reward functions. Furthermore, a mechanical sensor is introduced to achieve dynamic intraoperative monitoring and path correction. The system balances high efficiency and strategy adjustment capabilities, significantly reducing tissue damage rates and shortening operation time compared to traditional methods. It is applicable to various minimally invasive interventional surgeries, providing standardized and intelligent technical support for precision medicine. Attached Figure Description
[0026] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation
[0027] Example 1
[0028] A method for soft tissue puncture path planning and monitoring based on electrical impedance imaging includes the following steps:
[0029] Step S1: Before puncture, soft tissue information in different frequency domains is obtained through multi-frequency domain electrical impedance imaging, while richer structural information is obtained through ultrasound imaging and magnetic resonance imaging.
[0030] Step S11: Inject multi-frequency alternating current through the body surface electrode array to obtain the conductivity distribution in different frequency domains, which is used to better distinguish different tissues. The corresponding frequency domain image is output as E_f, where f is the frequency.
[0031] Step S12: Acquire ultrasound imaging data for preoperative supplementation of tissue structure information, denoted as U;
[0032] Step S13: Acquire magnetic resonance imaging data to supplement high-resolution structural and functional information before surgery, denoted as M;
[0033] Step S2: Establish an image reconstruction model, fuse multi-frequency domain electrical impedance imaging information, ultrasound imaging information and nuclear magnetic resonance imaging information, and improve resolution with the help of supplementary information.
[0034] Step S21: Train a priori auxiliary network by combining ultrasound data and MRI data to extract tissue features of key anatomical structures such as blood vessels and tumors, which serve as prior constraints for electrical impedance tomography reconstruction. The priori auxiliary network uses an encoder architecture, with ultrasound imaging data U and MRI imaging data M as inputs, and outputting the predicted features of key areas, denoted as the probability map matrix MP, ranging from [0,1]. Tightening 0 represents the background, and tightening 1 represents key structures such as blood vessels or tumors. The calculation formula is as follows:
[0035] MP = sigmoid(NETWORK(M,U))
[0036] In this context, NETWORK represents the encoder layer, sigmoid represents the activation function, and ultrasound imaging data U and magnetic resonance imaging data M are processed through the encoder layer and activated to obtain the prediction probability map MP of the key parts.
[0037] The purpose of modeling the loss function is to make the distribution of the predicted value MP close to the true value MP', as shown in the following formula:
[0038]
[0039] Where MP represents the feature data predicted by modeling, MP' represents the true feature label, which has only 0 and 1 values, · represents the dot product operation, || represents the summation operation, and μ represents the balancing parameter.
[0040] Step S22: Guide the optimization of the EIT reconstruction network by intelligently fusing the multi-frequency domain electrical impedance imaging information E_f and the prior key feature matrix MP. Utilize the prior probability map and multi-frequency domain information to improve resolution and enhance the ability to identify minute structures. Considering the sparsity of the data, the fusion formula is designed as follows:
[0041] E = MEAN(E_f)
[0042] Where E represents the electrical impedance imaging data after fusing multiple frequency domains, and MEAN represents the mean operation;
[0043] Q = Wq·E
[0044] Where Q represents the mapped electrical impedance imaging data, and Wq is the weight matrix;
[0045] K = (Wk·MP) T
[0046] Where K represents the mapped prior data MP, which is used to calculate the correlation. For ease of calculation, the T transpose operation is used to transform it, and Wk is the corresponding weight matrix.
[0047] V = Wv·E_f
[0048] Where V is used to calculate the weighted output of the correlation, and Wv is the corresponding weight matrix;
[0049]
[0050] Where A represents the correlation sparse matrix, i, j represent the corresponding positions in the matrix, exp represents the exponential operation, l represents taking only the top-k elements, k is a manually set value, and d represents the scaling dimension; the correlation sparse matrix can focus more on key areas;
[0051] E' = E + softmax(A)·V
[0052] Where E' represents the optimized electrical impedance imaging feature output, softmax is the standardization function, and the original information E is added to the feature map softmax(A)·V after calculating the correlation to obtain the optimized electrical impedance imaging feature E'.
[0053] Step S3: Preoperative path planning modeling. The preoperative path planning model is trained based on reinforcement learning. The needle tip position and its relative position to the target point during the puncture process are used as state information. An appropriate reward function is defined, such as giving a positive reward when the needle tip approaches the target point or successfully reaches it, and giving a negative reward when colliding with obstacles or deviating from the path. Through continuous interaction with the environment and trial-and-error learning, the agent can gradually master the optimal action strategy in different states.
[0054] Step S31: Based on the optimized electrical impedance imaging features E, key regions are detected and labeled through threshold segmentation and morphological operations. Key regions are divided into high-risk regions (such as blood vessels and nerves) and target tissues (such as tumors).
[0055] Step S32, Path Planning Modeling
[0056] Modeling objective: To achieve optimal path planning
[0057] Define the motion space A: {forward direction angle, propulsion force}.
[0058] Define the state space S: {impedance imaging features E', needle tip position, high-risk area and target tissue}
[0059] Global reward function design
[0060] Globalr = R_target proximity + R_danger area penalty + R_age coefficient. The global reward function guides the puncture path to avoid obstacles macroscopically and approach the target efficiently. The target proximity encourages rapid approach to the target tissue, while the penalty for approaching the danger area penalizes the path to approach high-risk areas (such as blood vessels and nerves) to ensure a safe distance. The age coefficient adjusts the strategy according to the patient's age.
[0061] R_target proximity = -a·||p_cur-p_obj||2
[0062] R target proximity encourages rapid approach to the target tissue, calculated as the Euclidean distance from the needle tip to the target tissue, where p_cur represents the needle tip position, p_obj represents the target position, ||||2 represents the Euclidean distance, and a is the weight of this item;
[0063] R_dangerous area penalty degree = -ρ∑ i∈高危区域 ||p_cur-p_danger_i||2
[0064] R represents the penalty for the path to the high-risk area, i represents the high-risk area number, p_danger_i represents the position of the i-th high-risk area, ||||2 represents the Euclidean distance, and ρ represents the penalty coefficient.
[0065] R-age coefficient = -c·|age-30|
[0066] The R-age coefficient adjusts the advancement strategy based on the patient's age. age represents the patient's age, c represents the weight of the age item, and || is the absolute value. When the patient is an elderly person or a child, the advancement intensity should be reduced.
[0067] Design of local reward functions;
[0068] Localr = R path smoothness
[0069] Local reward functions are used for refined design to balance path efficiency and safety. R-path smoothness encourages gradual changes in puncture direction between adjacent actions, reducing the risk of tissue tearing caused by sharp turns.
[0070] R path smoothness = -τ∑ i∈动作序列 |θcur_i-θpre_i|
[0071] R path smoothness encourages smooth changes in puncture direction between adjacent action sequences, where i represents a certain action, θcur_i represents the forward direction angle at the i-th action, θcur_i represents the previous forward direction angle at the i-th action, || belongs to the absolute value operation, and τ represents the weight of smoothness.
[0072] The reward function is the sum of the global and local reward functions;
[0073] R = Globalr + Localr
[0074] The update strategy for the Value function is as follows:
[0075] V(s t ,a t )←V(s t ,a t )+[R+σmax a' V(s t+1 ,a')-V(s t ,a t )]
[0076] The value function V is key to modeling and learning, where t represents time t and s. t a represents the state at time t. t Let R represent the action taken at time t, R be the reward function mentioned above, σ be the discount factor, and max be the value of the reward. a' V(s t+1 (,a') represents selecting an action a' and the state s at the next moment after executing the action. t+1 The corresponding Value function has the largest value;
[0077] Step S4: Install a mechanical sensor at the tip of the puncture needle to monitor the resistance of the puncture path in real time during the operation. Analyze the condition of the surrounding soft tissue through the needle tip sensor and dynamically adjust the path.
[0078] A needle-tip sensor data N is introduced for real-time monitoring of intraoperative path rationality and anomaly handling; the intraoperative local reward function is optimized as follows:
[0079] Localr = Localr + R (resistance anomaly)
[0080] R_resistance anomaly = -d·(Ft - Fsafe)
[0081] R-resistance anomaly measures whether the resistance Ft at the current time t exceeds the expected safety threshold Fsafe. A penalty is imposed according to the proportion of the exceedance, and d is the penalty weight of the resistance anomaly.
[0082] Step S5: Iterate through steps S3-S4 until the model converges, thus achieving the optimal route design.
[0083] Example 2
[0084] A soft tissue puncture path planning and monitoring system based on electrical impedance imaging is provided, wherein the system is used to execute the soft tissue puncture path planning and monitoring method based on electrical impedance imaging as described above, such as... Figure 1 Its characteristics are:
[0085] The data acquisition module is used to inject multi-frequency alternating current through a body surface electrode array to obtain conductivity distribution data in different frequency domains; at the same time, it acquires ultrasound imaging data and nuclear magnetic resonance imaging data to provide a multimodal information basis for subsequent path planning, and ensures the spatiotemporal consistency and availability of each modality of data through data preprocessing.
[0086] The data acquisition module further includes:
[0087] Electrical impedance data acquisition submodule: used to inject multi-frequency alternating current into the human body through a body surface electrode array, measure the boundary voltage change, obtain conductivity distribution data in different frequency domains, and reflect tissue differences from the perspective of electrophysiological characteristics;
[0088] Ultrasound data acquisition submodule: Utilizing the reflection and refraction characteristics of ultrasound waves propagating in human tissues, it acquires high frame rate two-dimensional tomographic image data to quickly provide morphological and structural information of soft tissues;
[0089] Magnetic resonance imaging (MRI) data acquisition submodule: Utilizes strong magnetic fields and radio frequency pulses to acquire high-resolution three-dimensional image data, providing rich information on tissue anatomy and function;
[0090] The data preprocessing submodule performs time synchronization, spatial alignment, noise reduction, and format conversion on the collected multimodal data to ensure data consistency and usability.
[0091] The image reconstruction module is used to construct a priori auxiliary network, which extracts key anatomical features such as blood vessels and tumors based on ultrasound and MRI data, and generates a probability map matrix. At the same time, it guides the EIT reconstruction network to fuse multi-frequency domain electrical impedance imaging information with the prior feature matrix and uses weighted and correlation calculation strategies to improve the resolution and microstructure recognition capability of electrical impedance imaging.
[0092] The image reconstruction module further includes: a priori feature extraction submodule and an EIT reconstruction optimization submodule;
[0093] The prior feature extraction submodule is based on ultrasound and MRI data to construct a prior auxiliary network, extract key anatomical features such as blood vessels and tumors, and generate a probability map matrix to provide prior constraints for electrical impedance imaging reconstruction.
[0094] The EIT reconstruction optimization submodule fuses multi-frequency domain electrical impedance imaging information with prior feature matrix and uses strategies such as weighting and correlation calculation to optimize electrical impedance imaging features, thereby improving image resolution and the ability to identify small structures.
[0095] The route planning module is used to mark high-risk areas and target tissues based on image reconstruction results through threshold segmentation and morphological operations; construct a reinforcement learning model, define a state space including impedance features and needle tip position, and an action space including forward direction angle and propulsion force; design global and local reward functions, and combine factors such as patient age, target proximity, and distance to danger zone, and iteratively optimize through the Value function to plan a safe and efficient puncture path;
[0096] The route planning module further includes: a key area marking submodule and a reinforcement learning modeling submodule;
[0097] The key region labeling submodule, based on optimized electrical impedance imaging features, detects and labels high-risk regions and target tissues through threshold segmentation and morphological operations.
[0098] The reinforcement learning modeling submodule is used to build reinforcement learning models, define a state space including impedance features, needle tip position, etc., and an action space including forward direction angle and propulsion force; design global and local reward functions, combine multiple factors, and iteratively optimize through the Value function to plan a safe and efficient puncture path;
[0099] The intraoperative monitoring module is used to collect resistance, resistance change rate and lateral force data in real time using the mechanical sensor at the tip of the puncture needle; integrate the sensor data into the local reward function to dynamically adjust the puncture path; analyze data trends, predict potential risks, and realize real-time dynamic monitoring and path correction of the puncture process.
[0100] The intraoperative monitoring module further includes: a mechanical data acquisition submodule and a path dynamic adjustment submodule;
[0101] Mechanical data acquisition submodule: Utilizes a mechanical sensor at the tip of the puncture needle to collect resistance data in real time during the puncture process, reflecting the interaction between the puncture needle and the surrounding soft tissue;
[0102] The path dynamic adjustment submodule integrates mechanical sensor data into the local reward function and automatically adjusts the forward direction angle and propulsion force of the puncture needle based on feedback to achieve dynamic correction of the puncture path.
[0103] The abnormal alarm module is used to monitor abnormal situations such as excessive resistance, path deviation, and sensor failure during the puncture process in real time. When an abnormality occurs, it will issue an alarm through sound and light, pop-up window, etc., automatically suspend the puncture operation or execute safety protection actions, and provide medical staff with handling suggestions to ensure surgical safety.
[0104] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A method for soft tissue puncture path planning and monitoring based on electrical impedance imaging, characterized by: Step S1: Before puncture, soft tissue information in different frequency domains is obtained by multi-frequency domain electrical impedance imaging, and structural information is obtained by ultrasound imaging and magnetic resonance imaging. Step S11: Inject multi-frequency alternating current through a surface electrode array to obtain the conductivity distribution in different frequency domains, which is used to distinguish different tissues. The corresponding frequency domain image is output and denoted as E_f, where f is the frequency. Step S12: Acquire ultrasound imaging data for preoperative supplementation of tissue structure information, denoted as U; Step S13: Acquire magnetic resonance imaging data to supplement high-resolution structural and functional information before surgery, denoted as M; Step S2: Establish an image reconstruction model, fuse multi-frequency domain electrical impedance imaging information, ultrasound imaging information and nuclear magnetic resonance imaging information, train a prior auxiliary network, and improve resolution with the help of supplementary information; Step S21: Train a priori auxiliary network by combining ultrasound data and MRI data to extract tissue features of blood vessels and key anatomical structures of tumors as prior constraints for electrical impedance tomography reconstruction. The priori auxiliary network uses an encoder architecture, with ultrasound imaging data U and MRI imaging data M as inputs, and outputting the predicted features of key areas, denoted as the probability map matrix MP, ranging from [0,1]. A tendency towards 0 represents background, and a tendency towards 1 represents key structures of blood vessels or tumors. The calculation formula is as follows: MP = sigmoid(NETWORK(M,U)) In this context, NETWORK represents the encoder layer, sigmoid represents the activation function, and ultrasound imaging data U and magnetic resonance imaging data M are processed through the encoder layer and activated to obtain the prediction probability map MP of the key parts. The purpose of modeling the loss function is to make the distribution of the predicted value MP close to the true value MP', as shown in the following formula: ； Where MP represents the predicted feature data, MP' represents the true feature label (with only 0 or 1 values), · represents the dot product operation, and || represents the summation operation. Represents the equilibrium parameters; Step S22: Guide the optimization of the EIT reconstruction network by intelligently fusing the multi-frequency domain electrical impedance imaging information E_f and the prior key feature matrix MP. Utilize the prior probability map and multi-frequency domain information to improve resolution and enhance the ability to identify minute structures. Considering the sparsity of the data, the fusion formula is designed as follows: E=MEAN(E_f) Where E represents the electrical impedance imaging data after fusing multiple frequency domains, and MEAN represents the mean operation; Q=Wq·E Where Q represents the mapped electrical impedance imaging data, and Wq is the weight matrix; K=(Wk·MP) T Where K represents the mapped prior data MP, which is used to calculate the correlation. For ease of calculation, the T transpose operation is used to transform it, and Wk is the corresponding weight matrix. V=Wv·E_f Where V is used to calculate the relevance-weighted output, and Wv is the corresponding weight matrix; ； Where A represents the correlation sparse matrix, i, j represent the corresponding positions in the matrix, exp represents the exponential operation, l represents taking only the top-k elements, k is a manually set value, and d represents the scaling dimension; the correlation sparse matrix can focus more on key areas; E' = E + softmax(A)·V Where E' represents the optimized electrical impedance imaging feature output, softmax is the standardization function, and the original information E is added to the feature map softmax(A)·V after calculating the correlation to obtain the optimized electrical impedance imaging feature E'. Step S3: Preoperative path planning modeling. The preoperative path planning model is trained based on the reinforcement learning approach. The position of the needle tip during the puncture process and its relative position to the target point are used as state information. An appropriate reward function is defined. Positive rewards are given when the needle tip approaches the target point or reaches it successfully, and negative rewards are given when it collides with obstacles or deviates from the path. Through continuous interaction with the environment and trial and error learning, the agent can gradually master the optimal action strategy in different states. Step S31: Based on the optimized electrical impedance imaging feature E, key regions are detected and marked through threshold segmentation and morphological operations. Key regions are divided into high-risk regions and target tissues. Step S32, Path Planning Modeling Modeling objective: To achieve optimal path planning Define the action space A: {forward direction angle, propulsion force}. Define the state space S: {impedance imaging features E', needle tip position, high-risk area and target tissue} Global reward function design Globalr = R target proximity + R danger zone penalty + R age coefficient The global reward function guides the puncture path to avoid obstacles and efficiently approach the target. Target proximity encourages rapid approach to the target tissue, while proximity to dangerous areas penalizes the path to high-risk areas to ensure a safe distance. The age coefficient adjusts the strategy according to the patient's age. R_target proximity = -a·||p_cur - p_obj||2 R target proximity encourages rapid approach to the target tissue, calculated as the Euclidean distance from the needle tip to the target tissue, where p_cur represents the needle tip position, p_obj represents the target position, || ||2 represents the Euclidean distance, and a is the item weight; R Dangerous Area Penalty = - ; R represents the penalty for dangerous areas, where the path is close to a high-risk area, and i represents the high-risk area number. Represents the location of the i-th high-risk area, || ||2 represents the Euclidean distance. Represents the penalty coefficient; R-age coefficient = -c * |age - 30| The R-age coefficient adjusts the advancement strategy based on the patient's age. age represents the patient's age, c represents the weight of the age item, and || is the absolute value. When the patient is an elderly person or a child, the advancement intensity should be reduced. Design of local reward functions; Localr = R Path smoothness Local reward functions are used for refined design to balance path efficiency and safety. R-path smoothness encourages gradual changes in puncture direction between adjacent actions, reducing the risk of tissue tearing caused by sharp turns. R path smoothness = - ; R-path smoothness encourages gradual changes in puncture direction between adjacent action sequences, where i represents a specific action. This represents the forward direction angle at the i-th iteration. This represents the previous forward direction angle at the i-th iteration. | This belongs to absolute value operations. Weights representing smoothness; The reward function is the sum of the global and local reward functions; R = Globalr + Localr The update strategy for the Value function is as follows: ; The value function V is key to modeling and learning, where t represents time t. This represents the state at time t. R represents the action taken at time t, and R is the reward function mentioned above. It is a discount factor. This represents the state at the next moment after selecting an action a' and executing that action. The corresponding Value function has the largest value; Step S4: Install a mechanical sensor at the tip of the puncture needle to monitor the resistance of the puncture path in real time during the operation. Analyze the condition of the surrounding soft tissue through the needle tip sensor and dynamically adjust the path. Introducing needle tip sensor data N and optimizing the local reward function are used for real-time monitoring of intraoperative path rationality and abnormal handling. Step S5: Iterate through steps S3-S4 until the model converges, thus achieving the optimal route design.
2. A soft tissue puncture path planning and monitoring system based on electrical impedance imaging, the system being used to execute the soft tissue puncture path planning and monitoring method based on electrical impedance imaging as described in claim 1, characterized in that: The data acquisition module is used to inject multi-frequency alternating current through a body surface electrode array to obtain conductivity distribution data in different frequency domains; at the same time, it acquires ultrasound imaging data and nuclear magnetic resonance imaging data to provide a multimodal information basis for subsequent path planning, and ensures the spatiotemporal consistency and availability of each modality of data through data preprocessing. The image reconstruction module is used to construct a priori auxiliary network, which extracts key anatomical features of blood vessels and tumors based on ultrasound and MRI data, and generates a probability map matrix. At the same time, it guides the EIT reconstruction network to fuse multi-frequency domain electrical impedance imaging information with the prior feature matrix and uses weighted and correlation calculation strategies to improve the resolution and microstructure recognition capability of electrical impedance imaging. The route planning module is used to mark high-risk areas and target tissues based on image reconstruction results through threshold segmentation and morphological operations; A reinforcement learning model is constructed, defining a state space containing impedance features and needle tip position elements, and an action space containing forward direction angle and propulsion force; global and local reward functions are designed, and a safe and efficient puncture path is planned by iteratively optimizing the Value function in combination with factors such as patient age, target proximity, and distance to the danger zone. The intraoperative monitoring module is used to collect data on resistance, rate of change of resistance, and lateral force in real time using a mechanical sensor at the tip of the puncture needle. Sensor data is incorporated into a local reward function to dynamically adjust the puncture path; Analyze data trends, predict potential risks, and achieve real-time dynamic monitoring and path correction during the puncture process; The abnormal alarm module is used to monitor resistance exceeding limits, path deviation, and sensor malfunctions in real time during the puncture process. When an abnormality occurs, it will issue an alarm through sound and light or pop-up window, automatically suspend the puncture operation or execute safety protection actions, and provide medical staff with handling suggestions to ensure surgical safety.
3. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored program, wherein the program, when running, controls the device where the non-volatile storage medium is located to execute the method of claim 1.
4. An electronic device, characterized in that, It includes a processor and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein the computer-readable instructions, when executed, perform the method of claim 1.