Self-driven micro-robot for liver tumor targeted therapy and real-time monitoring system

Abstract

The invention provides a self-driven micro-robot for a liver tumor targeted therapy and real-time monitoring system, and relates to the technical field of biomedical engineering. The monitoring main system comprises a self-driven navigation and positioning module, an intelligent drug delivery module, a tumor monitoring and state evaluation module, a multi-modal data fusion module, a real-time self-learning decision module, an advanced molecular simulation and drug optimization module, a safety communication module, a power grid environment adaptation module and a fault detection and recovery module; the self-driven navigation and positioning module adopts a mode of combining a quantum genetic algorithm, a fuzzy constraint optimization algorithm and a self-adaptive simulated annealing algorithm, so that the navigation accuracy and positioning efficiency of the robot in a complex environment in the liver are remarkably improved, and the quantum genetic algorithm has the advantages that by means of the strong parallelism of quantum calculation, the positioning accuracy of the robot is improved; according to the method, numerous potential path combinations can be quickly explored, the path planning time is greatly shortened, and meanwhile, the path planning accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical engineering technology, and in particular to a self-driven micro-robot for liver tumor targeted treatment and real-time monitoring system. Background Art

[0002] Traditional treatments for liver tumors mainly include surgical resection, chemotherapy, and radiotherapy. Surgical resection may be effective for early-stage tumors and when the patient's physical condition allows, but for mid- and late-stage tumors, especially when the tumor is located near important blood vessels or tissues, or has already metastasized, the operation is extremely difficult and risky, and it is difficult to completely remove the tumor tissue, and the postoperative recurrence rate is high. Chemotherapy inhibits the growth of tumor cells by using chemical drugs, but due to the lack of specific targeting of the drugs, while killing tumor cells, it will also cause serious toxic side effects to normal liver cells and other organs of the body.

[0003] According to a monitoring system and method for the efficacy of targeted treatment of liver cancer with Chinese patent number CN113140275B, all monitoring information of the same patient at each treatment stage is dynamically associated and compared to form a historical display interface, and the patient's behavioral habit information and subsequent targeted treatment preset plan are collected and added to the historical display interface to obtain the current display interface; the current display interface includes a three-dimensional tumor change trend graph and various key parameter change trend lines; the efficacy evaluation result is obtained based on the three-dimensional tumor change trend graph and various key parameter change trend lines. The present invention effectively solves the problem of evaluating the efficacy of targeted treatment of liver cancer based on the doctor's experience alone, and can intuitively show the efficacy of targeted treatment of liver cancer to doctors and patients at the same time.

[0004] At present, with the rapid development of multidisciplinary fields such as micro-nanotechnology, biomedical engineering, and artificial intelligence, the application research of self-driving micro-robots in the biomedical field has gradually emerged and become a research direction with great potential. Self-driving micro-robots have the advantage of small size and can move autonomously in the complex physiological environment of the human body, such as blood vessels and tissue gaps, and are expected to achieve accurate diagnosis and treatment of diseases. However, the existing self-driving micro-robot technology still has the following problems in the application of liver tumor treatment: 1. In terms of navigation and positioning, traditional microrobot navigation algorithms have difficulty coping with the complex and ever-changing environment inside the liver. The liver's vascular network is intricate, the vascular walls have elastic deformation, the tissue density is uneven, and the blood flow is volatile. These uncertainties make it extremely challenging for robots to accurately navigate the liver. When some microrobots based on simple path planning algorithms navigate in the liver's blood vessels, they cannot adjust their paths in time according to the dynamic changes of the blood vessels, and may fall into areas of local vascular stenosis or misjudge the location of the tumor, making it impossible to effectively deliver drugs to the tumor site.

[0005] 2. In terms of drug delivery, traditional drug delivery methods lack precision and intelligence. The drug release dose and rate are usually set according to a fixed plan, without fully considering the individual differences of tumors, real-time status, and changes in the internal microenvironment of the liver. This may result in insufficient drug concentration at the tumor site, failing to achieve the expected therapeutic effect, or excessive drug concentration in normal liver tissue, aggravating toxic side effects.

[0006] 3. In terms of tumor monitoring and assessment, existing technical means have certain limitations. Single-modality imaging technology (such as CT, MRI or ultrasound) can only provide partial information about the tumor and cannot fully and accurately reflect the status of the tumor. For example, although ultrasound imaging has the advantages of good real-time performance and low cost, the image resolution is relatively low, and the subtle structure and pathological characteristics inside the tumor are not clear enough; although CT and MRI can provide high-resolution images, their ability to detect functional information of the tumor (such as the activity of tumor cells, metabolic status, etc.) is limited.

[0007] 4. In terms of data fusion and decision-making, traditional data fusion methods are difficult to efficiently integrate data from different sensors and lack real-time self-learning capabilities. When fusing ultrasound imaging data of liver tumors and tumor marker concentration data detected by electrochemical sensors, due to differences in data dimensions and distribution, some traditional data fusion algorithms cannot effectively remove noise interference and extract valuable information, thus affecting the accuracy of tumor diagnosis and treatment.

[0008] Therefore, self-driving microrobots are needed for targeted liver tumor treatment and real-time monitoring systems to solve the above problems. Summary of the invention

[0009] Technical issues solved In view of the deficiencies in the prior art, the present invention provides a self-driving microrobot for targeted liver tumor treatment and real-time monitoring system, which solves the problems in the above background technology.

[0010] Technical Solution To achieve the above objectives, the present invention is implemented through the following technical solutions: a self-driving microrobot for targeted liver tumor treatment and real-time monitoring system, including a monitoring main system, the monitoring main system includes a self-driving navigation and positioning module, an intelligent drug delivery module, a tumor monitoring and status assessment module, a multimodal data fusion module, a real-time self-learning decision module and an advanced molecular simulation and drug optimization module, a secure communication module, a power grid environment adaptation module and a fault detection and recovery module, wherein: The self-driving navigation and positioning module includes a quantum genetic algorithm, a fuzzy constraint optimization algorithm, and an adaptive simulated annealing algorithm, which are used to achieve accurate navigation and obstacle avoidance of the robot in the complex environment of the liver; The intelligent drug delivery module includes an adaptive particle swarm optimization algorithm based on Lévy flight, a nonlinear feedback neural network, and a multibody dynamics simulation algorithm to optimize the dosage and strategy of drug release; The tumor monitoring and status assessment module includes a Bayesian deep convolutional generative adversarial network, a variational inference algorithm, and an edge detection and feature fusion algorithm, which are used for synthesis, repair, enhancement, and real-time assessment of tumor images; The multimodal data fusion module includes an adaptive density clustering algorithm, reinforcement learning and graph convolutional networks, and simulated annealing and multi-granularity convolutional networks, which are used to efficiently fuse data from different sensors and improve the accuracy of tumor diagnosis and treatment; The real-time self-learning decision module includes a multi-fractal analysis algorithm, deep graph network reinforcement learning, and a temporal generative adversarial network for real-time optimization of treatment decisions; The advanced molecular simulation and drug optimization module includes a quantum particle swarm optimization algorithm, a kinetic Markov decision process and a molecular dynamics simulation algorithm, which are used for the design of drug molecules and the optimization of delivery pathways.

[0011] Preferably, the quantum genetic algorithm (QGA) in the self-driving navigation and positioning module accelerates path search through quantum computing to improve the accuracy and efficiency of navigation path planning. The quantum genetic algorithm performs encoding operations based on the superposition state and entanglement characteristics of quantum bits. During the path planning process, the internal environment of the liver is abstracted into a three-dimensional spatial network, in which nodes represent vascular branch points and tissue gap key points, and edges represent accessible paths between nodes. The quantum genetic algorithm QGA updates chromosomes through quantum rotating gate operations. In each generation of evolution, the parallelism of quantum states is used to simultaneously explore multiple groups of path combinations. The fuzzy constraint optimization algorithm models uncertain factors through fuzzy set theory, where uncertain factors include Including the elastic deformation of the blood vessel wall, the unevenness of tissue density and the volatility of blood flow. During the application process, the elastic deformation of the blood vessel wall, the unevenness of tissue density and the volatility of blood flow are defined as fuzzy variables, and fuzzy constraints are constructed. When planning the path, the robot's kinematic feasibility, the safe distance from the blood vessel wall and surrounding tissues, and the estimated time to reach the tumor are used as optimization goals. The fuzzy membership function is used to quantitatively evaluate the degree of satisfaction of each constraint. When the robot navigates in the blood vessel, the path planning strategy is dynamically adjusted according to the real-time feedback information from the sensor to ensure that the robot navigates safely and accurately in the complex and uncertain liver environment and avoids collision with the blood vessel wall or other important tissues.

[0012] Preferably, the Lévy flight adaptive particle swarm optimization algorithm (LPSO) in the intelligent drug delivery module is used to accelerate the search for the optimal dose and rate of drug release and improve the therapeutic effect of the drug. The Lévy flight adaptive particle swarm optimization algorithm gives the particle swarm the ability to make long-distance jumps in the search space, breaking through the limitation that the traditional particle swarm optimization algorithm is prone to fall into the local optimum. During drug delivery, the Lévy flight adaptive particle swarm optimization algorithm takes the release dose and rate of the drug carried by the robot as the optimization target based on three factors: the diffusion model of the drug in the liver tissue, the uptake rate of the drug by tumor cells, and the metabolic kinetics of the drug. When in use, by simulating the diffusion process of drug molecules in the liver blood vessels and tissue gaps, the Lévy flight adaptive particle swarm optimization algorithm quickly searches for the optimal drug release strategy under the tumor size, location and pathological state, accelerates the search for the optimal dose and rate of drug release, improves the therapeutic effect of the drug, and reduces the toxic side effects on normal liver tissue.

[0013] Preferably, the Bayesian deep convolutional generative adversarial network (BD-CGAN) in the tumor monitoring and state assessment module is used to generate tumor images that are close to reality, enhance image quality and reduce interference factors. The Bayesian deep convolutional generative adversarial network consists of two parts: a generator and a discriminator. The generator is based on a deep convolutional neural network structure and uses Bayesian reasoning to model the uncertainty of network parameters. During the training process, the generator receives a random noise vector as input and attempts to generate a tumor image that is close to reality. The discriminator distinguishes and judges the generated image from the real tumor image. Through the adversarial training of the two, the quality of the image generated by the generator is continuously improved. In actual operation, in the ultrasound image monitoring of liver tumors, the Bayesian deep convolutional generative adversarial network uses high-quality ultrasound image samples for training.

[0014] Preferably, the reinforcement learning and graph convolutional network (RL-GCN) in the multimodal data fusion module is adaptively adjusted according to the spatial relationship of different sensor data to improve the accuracy of data fusion. The adaptive density clustering algorithm identifies local density changes in data from ultrasonic sensors, optical sensors, and electrochemical sensors, and determines cluster centers. In actual operation, when fusing ultrasonic imaging data of liver tumors and tumor marker concentration data detected by electrochemical sensors, the local density around each data point is first analyzed, and data points with high density and interconnected are divided into a cluster. For isolated points with low density and noise data, clustering parameters are adaptively adjusted to assign them to the nearest cluster or mark them as abnormal data. This can effectively integrate effective information in different modal data, remove noise interference, and improve the accuracy and reliability of data fusion.

[0015] Preferably, the temporal generative adversarial network (TGAN) in the real-time self-learning decision module predicts the future evolution of the tumor, dynamically adjusts the treatment plan, models and predicts the development process of the liver tumor, and predicts the size change of the tumor in the next week and whether metastasis will occur based on the patient's tumor marker change curve and ultrasound image sequence in the past month.

[0016] Preferably, the quantum particle swarm optimization algorithm (QPSO) in the advanced molecular simulation and drug optimization module optimizes the design of drug molecules through quantum computing, thereby improving the targeting and therapeutic effects of drugs. The advanced molecular simulation and drug optimization module combines the characteristics of quantum computing with the particle swarm optimization algorithm. In terms of drug molecule design, quantum bits are used to encode the structural characteristics and chemical properties of drug molecules, including the bond length, bond angle and charge distribution of the molecules. Through quantum gate operations and the information sharing mechanism of the particle swarm, the drug molecular structure with the best targeting and activity is searched in the chemical space. For specific receptors on the surface of liver tumor cells, the quantum particle swarm optimization algorithm can optimize the design of drug molecular structures that bind to them with high affinity, thereby improving the targeting of drugs. At the same time, the metabolic stability and toxicity of drug molecules are considered, the overall performance of drug molecules is optimized, and the therapeutic effects of drugs are improved.

[0017] Preferably, the secure communication module uses the Advanced Encryption Standard (AES) encryption algorithm and elliptic curve cryptography (ECC) to encrypt the data transmitted between the robot and the external monitoring device, ensure the confidentiality of the data during transmission, prevent the patient's privacy information and treatment monitoring data from being stolen, and use the message authentication code (MAC) technology to verify the integrity of the data. During each data transmission, the sender uses the key to generate a message authentication code and attaches it to the data. The receiver uses the same key to verify the received data. If the message authentication code does not match, it indicates that the data may have been tampered with. The receiver will refuse to receive the data and notify the sender to resend it, thereby ensuring the integrity and reliability of the data and ensuring the safe and stable operation of the entire treatment monitoring process.

[0018] Preferably, the power grid environment adaptation module is equipped with a voltage regulating device to monitor the fluctuation of the power grid voltage in real time. When the power grid voltage is higher than a threshold or lower than a threshold, the voltage is adjusted to the range required for the normal operation of the robot through an automatic transformer.

[0019] Preferably, the fault detection and recovery module monitors the system operating status in real time, detects potential faults, and automatically restores system functions when a fault occurs.

[0020] Beneficial Effects The present invention provides a self-driving micro robot for targeted liver tumor treatment and real-time monitoring system. It has the following beneficial effects: 1. The self-driving navigation and positioning module of the present invention adopts a combination of quantum genetic algorithm, fuzzy constraint optimization algorithm and adaptive simulated annealing algorithm, which significantly improves the navigation accuracy and positioning efficiency of the robot in the complex environment of the liver. The quantum genetic algorithm, with the help of the powerful parallelism of quantum computing, can quickly explore many potential path combinations, greatly shorten the path planning time, and improve the accuracy of path planning. The fuzzy constraint optimization algorithm accurately models and effectively processes uncertain factors such as elastic deformation of vascular walls, tissue density heterogeneity, and blood flow volatility, so that the robot can dynamically adjust the path according to the real-time collected environmental information during navigation, effectively avoiding falling into local vascular stenosis areas or misjudging the location of tumors.

[0021] 2. The intelligent drug delivery module of the present invention integrates an adaptive particle swarm optimization algorithm based on Lévy flight, a nonlinear feedback neural network and a multibody dynamics simulation algorithm, realizing intelligent and precise control of drug release dosage and strategy. The adaptive particle swarm optimization algorithm based on Lévy flight breaks through the limitation that traditional algorithms are prone to fall into local optimality, and can quickly search for the optimal drug release dosage and rate combination according to the size, location, pathological state of the tumor and liver tissue microenvironment information, effectively improving the concentration accumulation of the drug in the tumor site and enhancing the treatment effect. The nonlinear feedback neural network receives the tumor microenvironment information and drug concentration information collected by the robot sensor in real time, and accurately predicts the apoptosis rate of tumor cells and the changing trend of tumor volume under different drug release strategies through complex feature extraction and mapping.

[0022] 3. The tumor monitoring and state assessment module in the present invention utilizes the Bayesian deep convolutional generative adversarial network, variational reasoning algorithm, and edge detection and feature fusion algorithm to effectively solve many problems existing in existing tumor monitoring technologies, and realize high-quality synthesis, restoration, enhancement, and comprehensive and accurate real-time assessment of tumor images. The Bayesian deep convolutional generative adversarial network uses an innovative method of combining a deep convolutional neural network structure with Bayesian reasoning. It only requires a small amount of high-quality ultrasound image samples to train and generate a large number of tumor images that are close to reality and rich in details, significantly enhancing the quality of ultrasound images, effectively reducing interference factors such as noise and artifacts, and providing doctors with clearer and more accurate tumor image information.

[0023] 4. The adaptive density clustering algorithm, reinforcement learning and graph convolutional network, and simulated annealing and multi-granularity convolutional network in the multimodal data fusion module of this aspect work together to achieve efficient fusion of different sensor data, significantly improve the diagnosis and treatment accuracy of tumors, and provide reliable data support for real-time self-learning decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1It is a specific framework flow chart of the present invention; Figure 2 This is a simulation diagram of the monitoring main system of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Specific embodiment one: like Figure 1-2 As shown, the self-driving micro robot is used for targeted treatment and real-time monitoring of liver tumors, including a main monitoring system, which includes a self-driving navigation and positioning module, an intelligent drug delivery module, a tumor monitoring and status assessment module, a multimodal data fusion module, a real-time self-learning decision module, an advanced molecular simulation and drug optimization module, a secure communication module, a power grid environment adaptation module, and a fault detection and recovery module, among which: The self-driving navigation and positioning module includes quantum genetic algorithm, fuzzy constraint optimization algorithm and adaptive simulated annealing algorithm, which are used to achieve accurate navigation and obstacle avoidance of the robot in the complex environment of the liver; The intelligent drug delivery module includes an adaptive particle swarm optimization algorithm based on Lévy flight, a nonlinear feedback neural network, and a multibody dynamics simulation algorithm to optimize the dosage and strategy of drug release; The tumor monitoring and status assessment module includes a Bayesian deep convolutional generative adversarial network, a variational inference algorithm, and an edge detection and feature fusion algorithm, which are used for the synthesis, restoration, enhancement, and real-time assessment of tumor images. The multimodal data fusion module includes adaptive density clustering algorithm, reinforcement learning and graph convolutional network, and simulated annealing and multi-granularity convolutional network, which are used to efficiently fuse data from different sensors and improve the accuracy of tumor diagnosis and treatment; The real-time self-learning decision-making module includes multi-fractal analysis algorithms, deep graph network reinforcement learning, and temporal generative adversarial networks, which are used to optimize treatment decisions in real time; The Advanced Molecular Simulation and Drug Optimization module includes quantum particle swarm optimization algorithm, kinetic Markov decision process and molecular dynamics simulation algorithm, which are used for the design of drug molecules and optimization of delivery pathways.

[0027] The quantum genetic algorithm (QGA) in the self-driving navigation and positioning module accelerates path search through quantum computing and improves the accuracy and efficiency of navigation path planning. The quantum genetic algorithm performs encoding operations based on the superposition and entanglement characteristics of quantum bits. During the path planning process, the internal environment of the liver is abstracted into a three-dimensional spatial network, in which nodes represent vascular branch points and tissue gap key points, and edges represent accessible paths between nodes. The quantum genetic algorithm QGA updates chromosomes through quantum revolving gate operations. In each generation of evolution, the parallelism of quantum states is used to simultaneously explore multiple groups of path combinations. The fuzzy constraint optimization algorithm models uncertain factors through fuzzy set theory, where uncertain factors include the elasticity of the vascular wall. In the application process, the elastic deformation of the blood vessel wall, the unevenness of tissue density and the volatility of blood flow are defined as fuzzy variables, and fuzzy constraints are constructed. When planning the path, the robot's kinematic feasibility, the safe distance from the blood vessel wall and surrounding tissues, and the estimated time to reach the tumor are used as optimization goals. The fuzzy membership function is used to quantitatively evaluate the degree of satisfaction of each constraint. When the robot navigates in the blood vessel, the path planning strategy is dynamically adjusted according to the real-time feedback from the sensor to ensure that the robot navigates safely and accurately in the complex and uncertain liver environment and avoids collision with the blood vessel wall or other important tissues.

[0028] The Lévy flight adaptive particle swarm optimization algorithm (LPSO) in the intelligent drug delivery module is used to accelerate the search for the optimal dose and rate of drug release and improve the therapeutic effect of drugs. The Lévy flight adaptive particle swarm optimization algorithm gives the particle swarm the ability to make long-distance jumps in the search space, breaking through the limitation of traditional particle swarm optimization algorithms that are prone to fall into local optimality. During drug delivery, the Lévy flight adaptive particle swarm optimization algorithm takes the release dose and rate of the drug carried by the robot as the optimization target based on three factors: the diffusion model of the drug in the liver tissue, the uptake rate of the drug by tumor cells, and the metabolic kinetics of the drug. When in use, by simulating the diffusion process of drug molecules in the liver blood vessels and tissue gaps, the Lévy flight adaptive particle swarm optimization algorithm quickly searches for the optimal drug release strategy under the tumor size, location and pathological state, accelerates the search for the optimal dose and rate of drug release, improves the therapeutic effect of the drug, and reduces the toxic side effects on normal liver tissue.

[0029] The nonlinear feedback neural network constructs a neural network structure with multiple hidden layers. The input layer receives information about the tumor microenvironment from the robot sensor, including physiological parameters such as the local temperature, pH value, and oxygen partial pressure of the tumor, as well as the real-time concentration information of the drug in the body. The nonlinear activation function performs complex feature extraction and mapping on these input information, and the output layer predicts key indicators such as the apoptosis rate of tumor cells and the change trend of tumor volume under different drug release strategies. The back propagation algorithm is used to continuously adjust the weights and thresholds of the neural network to achieve accurate prediction and intelligent regulation of drug delivery effects. When acidification of the local tumor environment is detected, the nonlinear feedback neural network can adjust the drug release rate according to the pre-trained model to adapt to changes in tumor cell sensitivity to drugs.

[0030] The multi-body dynamics simulation algorithm takes into account the interaction forces between the robot and the surrounding tissues, blood vessel walls, and drug release process when the robot moves in the liver tissue, and establishes a multi-body dynamics model of the robot, drug carrier, and liver tissue. It simulates the robot's swimming process in the blood vessels, the impact force during drug release on the surrounding tissues, and the reaction force of tissue deformation on the robot's movement, and other complex mechanical behaviors. Through simulation analysis, the robot's structural design and drug release mechanism are optimized to ensure that the drug can be accurately delivered to the tumor site and the release process is stable and reliable. When designing the robot's propeller structure, the multi-body dynamics simulation algorithm can predict the thrust and torque generated by propellers of different shapes and sizes in the liver blood vessels, as well as the pressure on the blood vessel wall, so as to select the optimal propeller design scheme to ensure the efficiency and safety of the robot in the drug delivery process.

[0031] The Bayesian Deep Convolutional Generative Adversarial Network (BD-CGAN) in the tumor monitoring and status assessment module is used to generate tumor images that are close to reality, enhance image quality and reduce interference factors. The Bayesian Deep Convolutional Generative Adversarial Network consists of two parts: a generator and a discriminator. The generator is based on a deep convolutional neural network structure and uses Bayesian reasoning to model the uncertainty of network parameters. During the training process, the generator receives a random noise vector as input and tries to generate tumor images that are close to reality. The discriminator distinguishes between the generated images and the real tumor images. Through adversarial training between the two, the quality of the images generated by the generator is continuously improved. In actual operation, in the ultrasound image monitoring of liver tumors, the Bayesian Deep Convolutional Generative Adversarial Network uses high-quality ultrasound image samples for training.

[0032] Variational inference algorithms target the high dimensionality and complexity of tumor monitoring data, such as the fusion analysis of multimodal imaging data (CT, MRI, PET) and biomarker detection data. Variational inference algorithms map complex high-dimensional data to low-dimensional latent space by building a probabilistic model, and estimate the posterior distribution of the data. In this latent space, operations such as data clustering, classification, and anomaly detection can be performed more conveniently. For example, the CT image features, tumor marker concentrations, and gene expression data of patients with liver tumors are fused and analyzed through variational inference algorithms to mine the potential associations between different data modalities, identify important information such as tumor subtypes and grades, and predict the risk of tumor metastasis, thereby achieving a comprehensive and accurate assessment of tumor status.

[0033] In tumor image analysis, edge detection algorithms use information such as grayscale changes and texture features of images to accurately extract the boundary contours between tumor tissue and surrounding normal tissue. For example, gradient-based edge detection operators, such as the Sobel operator and the Canny operator, are used to detect edge information on tumor images of different resolutions, and then the edge features are fused with the texture, density and other features inside the tumor through a feature fusion algorithm. The fused feature vector can more comprehensively describe the morphology, structure and growth characteristics of the tumor, providing a richer basis for accurate diagnosis and treatment effect evaluation of the tumor. For example, when evaluating the effect of tumor radiotherapy, edge detection and feature fusion algorithms can accurately judge the tumor shrinkage and treatment response by comparing the clarity of the tumor edge before and after treatment, the changes in internal texture and the changes in overall morphology.

[0034] The reinforcement learning and graph convolutional network (RL-GCN) in the multimodal data fusion module is adaptively adjusted according to the spatial relationship of different sensor data to improve the accuracy of data fusion. The adaptive density clustering algorithm identifies local density changes in data from ultrasonic sensors, optical sensors, and electrochemical sensors, and determines the cluster center. In actual operation, when fusing the ultrasonic imaging data of liver tumors and the tumor marker concentration data detected by electrochemical sensors, the local density around each data point is first analyzed, and the data points with high density and interconnected are divided into a cluster. For isolated points with low density and noise data, the clustering parameters are adaptively adjusted to assign them to the nearest cluster or mark them as abnormal data. This can effectively integrate the effective information in different modal data, remove noise interference, and improve the accuracy and reliability of data fusion.

[0035] The reinforcement learning part enables the agent to continuously learn the optimal data fusion strategy in the process of interaction with the environment by defining the agent (i.e., data fusion system), environment (multimodal data space), action (data fusion strategy selection) and reward function (accuracy and usefulness evaluation index of fused data). The graph convolutional network is used to model the spatial relationship and semantic association between different sensor data. For example, the anatomical structure information of the liver is constructed as a graph structure, each node represents an organ tissue or region, and the edge represents the spatial adjacent relationship or functional connection between them. The sensor data is associated with the corresponding graph node, and the spatial correlation of the data can be fully utilized for fusion processing through the information transmission mechanism of the graph convolutional network. For example, when fusing liver vascular ultrasound data and metabolite concentration data of tumor tissue, RL-GCN can adaptively adjust the weights and fusion methods of different data according to the spatial position relationship between blood vessels and tumor tissue, improve the accuracy of data fusion, and thus provide more accurate information support for tumor diagnosis and treatment.

[0036] The simulated annealing algorithm is used in multimodal data fusion to avoid falling into the local optimal solution. In the data fusion process, as the iteration proceeds, a poor fusion result is accepted with a certain probability, so as to jump out of the local optimum and explore a better fusion solution. The multi-granularity convolutional network can extract and fuse data features from different scales. For example, for CT image data of liver tumors, the multi-granularity convolutional network can perform convolution operations at the local pixel level, regional block level, and overall image level at the same time, extract image features at different levels, and then optimize the fusion weights of different granularity features through the simulated annealing algorithm to obtain a more comprehensive and representative fusion feature vector, thereby improving the accuracy of tumor diagnosis and treatment.

[0037] The temporal generative adversarial network (TGAN) in the real-time self-learning decision-making module models and predicts the development process of liver tumors by predicting the future evolution of the tumor and dynamically adjusting the treatment plan. Based on the patient's tumor marker change curve and ultrasound image sequence over the past month, it predicts the size change of the tumor in the next week and whether metastasis will occur.

[0038] The multi-analysis algorithm performs multi-fractal feature analysis on physiological signals of liver tumors (such as heart rate variability, blood pressure fluctuations, etc.) and tumor-related indicators (such as tumor volume change rate, tumor marker concentration fluctuations, etc.). By calculating the multi-fractal parameters of the signal, such as the fractal dimension and singular spectrum, the self-similarity and complexity changes of the signal at different time scales and spatial scales are revealed. For example, during tumor treatment, the multi-fractal analysis algorithm can discover intermittent acceleration or deceleration during tumor growth, as well as changes in the multi-fractal characteristics of tumor physiological signals after therapeutic intervention. This information helps to judge the development trend and treatment effect of the tumor, and provides an important basis for subsequent treatment decisions.

[0039] Deep graph network reinforcement learning constructs a deep graph network structure to represent the state space, action space, and transfer probability during the treatment of liver tumors. The state space includes information such as the current size, location, pathological type, and physical condition of the patient, while the action space covers various treatment methods (such as drug dose adjustment, radiotherapy intensity change, surgical plan selection, etc.) and their combinations. Through the reinforcement learning algorithm, the agent (i.e., the decision-making system) continuously explores different treatment strategies in the deep graph network, obtains rewards based on environmental feedback (such as tumor shrinkage, patient adverse reactions, etc.), and gradually learns the optimal treatment decision sequence. For example, when faced with complex liver tumor cases, deep graph network reinforcement learning can comprehensively consider the heterogeneity of the tumor, the individual differences of the patient, and the dynamic changes during the treatment process, and formulate a personalized, real-time optimized treatment plan to improve the treatment effect and reduce complications.

[0040] The quantum particle swarm optimization algorithm (QPSO) in the advanced molecular simulation and drug optimization module optimizes the design of drug molecules through quantum computing, improving the targeting and therapeutic effects of drugs. The advanced molecular simulation and drug optimization module combines the characteristics of quantum computing with the particle swarm optimization algorithm. In terms of drug molecule design, quantum bits are used to encode the structural characteristics and chemical properties of drug molecules, including the bond length, bond angle and charge distribution of the molecules. Through quantum gate operations and the information sharing mechanism of particle swarms, the drug molecular structure with the best targeting and activity is searched in the chemical space. For specific receptors on the surface of liver tumor cells, the quantum particle swarm optimization algorithm can optimize the design of drug molecular structures that bind to them with high affinity, improve the targeting of drugs, and at the same time consider the metabolic stability and toxicity of drug molecules, optimize the overall performance of drug molecules, and improve the therapeutic effects of drugs.

[0041] The kinetic Markov decision process constructs a kinetic model of the drug molecule delivery process in the liver, and regards the absorption, distribution, metabolism and excretion process of the drug as a Markov decision process. The state variables include the concentration of the drug in different tissues and organs, the chemical form of the drug molecules, etc., and the action is the drug delivery method (such as injection site, injection rate, drug dosage form selection, etc.). According to the pharmacokinetic model, the transition probability and reward function of taking different actions under different states are calculated (such as the accumulation of the drug in the tumor tissue, the toxicity to normal tissues, etc.). By solving the kinetic Markov decision process, the optimal drug delivery path and strategy are determined to achieve the maximum accumulation of drugs in the liver tumor site and the minimum side effects. For example, when designing a chemotherapy drug delivery plan for liver tumors, the kinetic Markov decision process can determine whether to use arterial injection or intravenous injection, as well as the injection time interval and dose adjustment strategy based on the blood perfusion characteristics of the liver, the vascular distribution of the tumor, and the pharmacokinetic characteristics of the drug, so as to improve the therapeutic effect of the drug and reduce systemic toxicity.

[0042] Molecular dynamics simulation algorithms use molecular dynamics simulation technology to study the interaction mechanism between drug molecules and liver tumor cell targets and surrounding biological molecules (such as proteins, nucleic acids, etc.). By establishing a detailed molecular model, the movement trajectory of drug molecules in a physiological environment, the binding process with the target, and the impact on intracellular signaling pathways are simulated. For example, the binding process of anticancer drug molecules with specific enzyme proteins in liver tumor cells is simulated, and the conformational changes of proteins before and after binding, changes in active sites, and the blocking effect on downstream signal transduction are analyzed. Molecular dynamics simulation algorithms can provide a theoretical basis at the microscopic level for the optimal design of drugs, help design more efficient and specific drug molecules, and also help understand the mechanism of action and potential adverse reactions of drugs, providing strong support for drug research and development and clinical applications.

[0043] The secure communication module uses the Advanced Encryption Standard (AES) encryption algorithm and elliptic curve cryptography (ECC) to encrypt the data transmitted between the robot and the external monitoring device, ensuring the confidentiality of the data during transmission and preventing the patient's privacy information and treatment monitoring data from being stolen. At the same time, the message authentication code (MAC) technology is used to verify the integrity of the data. During each data transmission, the sender uses the key to generate a message authentication code and attaches it to the data. The receiver uses the same key to verify the received data. If the message authentication code does not match, it indicates that the data may have been tampered with. The receiver will refuse to receive the data and notify the sender to resend it, thereby ensuring the integrity and reliability of the data and ensuring the safe and stable operation of the entire treatment monitoring process.

[0044] The grid environment adaptation module is equipped with a voltage regulating device to monitor the fluctuation of grid voltage in real time. When the grid voltage is higher or lower than the threshold, the voltage is adjusted to the range required for the normal operation of the robot through an automatic transformer.

[0045] The fault detection and recovery module monitors the system operation status in real time, detects potential faults, and automatically restores system functions when a fault occurs. Specific embodiment 2: like Figure 1-2 As shown, the key algorithm mentioned in Example 1 is analyzed in detail below, including its core mathematical formula and explanation: 1. Quantum Genetic Algorithm (QGA) Qubit representation: ; in, and is the ground state of the qubit, and is a complex number that satisfies the normalization condition .

[0047] Fitness evaluation: In each generation, according to the fitness function Evaluate the pros and cons of each quantum individual: ; in, It is decision variables, is its corresponding weight.

[0048] Quantum genetic operation: Quantum genetic algorithm generates new individuals through quantum gates, quantum superposition and quantum interference operations. In each generation, individuals with higher fitness are selected for crossover and mutation.

[0049] in represents the probability amplitude of the two ground states in the quantum superposition state, which determines the state of the quantum bit, is the fitness function, which is used to evaluate the quality of the current solution.

[0050] 2. Fuzzy Constrained Optimization Algorithm (FCA) Fuzzy constraint processing: ; in, is the constraint function that represents the constraints of the optimization problem. is the fuzzification factor, is the threshold value.

[0051] Objective function optimization: ; in, are the coefficients of the objective function, is the weight of the fuzzy constraint.

[0052] Represents the fuzzy form of the constraint function, and uses the Sigmoid function to transform the traditional hard constraints into soft constraints. is the objective function, in addition to considering the target variable In addition to the coefficients, it also includes the weighted terms of the constraints.

[0053] 3. Lévy flight adaptive particle swarm optimization algorithm (Lévy-PSO) Particle Update: ; ; in, It is The position of a particle, It is The speed of a particle, is the inertia weight, is the learning factor, is a random number, is the step length of the Lévy flight.

[0054] Lévy flight step length: ; in, are the parameters of the Lévy flight, is the step size, is the Gamma function, Represents the velocity of the particle, and its update formula includes the inertia term , the attractiveness of individual optimal solutions and global optimal solutions (through and Adjustments) and exploratory adjustments to Lévy flight , This makes the movement of particles have a strong exploration ability, which helps to escape from the local optimal solution.

[0055] 4. Nonlinear Feedback Neural Network (NFNN) Network output: ; in, is the input signal, is the weight, is the bias term, It is a non-linear activation function (such as Sigmoid, ReLU, etc.).

[0056] Feedback Mechanism: ; in, is the feedback output, is the feedback intensity.

[0057] It is the output of the neural network, which is affected by the weighted and biased terms of the input signal and mapped through a nonlinear activation function. The feedback mechanism affects the current input through the previous output, thereby adjusting the behavior of the network and enhancing its adaptability.

[0058] 5. Bayesian Deep Convolutional Generative Adversarial Network (BD-CGAN) Generator Network: ; in, is the generator network, is the input noise, are the parameters of the generator, is the mapping function of the generator.

[0059] Discriminator network: ; in, is the discriminator network, is the input sample, are the parameters of the discriminator, is the Sigmoid function.

[0060] Generate adversarial loss: ; in, is the expectation, which represents the mean of the loss function, is the judgment of the discriminator on the real sample, is the discriminator’s decision on the generated sample, The generator is based on the input noise and parameters Generate data, The discriminator is based on .

[0061] 6. Reinforcement Learning with Graph Convolutional Networks (RL-GCN) Graph Convolutional Network (GCN) Update: ; in It is The node representation of the layer, is the normalized adjacency matrix of the graph, It is The weight matrix of the layer, is the activation function.

[0062] Reinforcement Learning (RL) Value Function: ; in, Yes Status The value function of It is a cumulative reward.

[0063] Q-learning update formula: ; in, It's action In Status The value of It’s an instant reward. is the discount factor, is the learning rate.

[0064] The GCN part passes through the adjacency matrix of the graph Node information is propagated to generate richer representations in the network. The reinforcement learning part uses Q-learninq to evaluate each state-action pair and continuously update the strategy to maximize the reward.

[0065] 7. Simulated Annealing Algorithm (SA) Status Update: ; in, is the current solution, is a random perturbation in the search space.

[0066] Probability of acceptance: ; in, is the change in energy (objective function value), is the current temperature.

[0067] Temperature drop: ; in, is the cooling factor, usually , is a random change, allowing the algorithm to jump out of the local optimal solution, Controls whether to accept solutions that are not as good as the current solution. As decreases, the probability of accepting a worse solution gradually decreases, simulating the physical annealing process.

[0068] 8. Deep Graph Network Reinforcement Learning (DeepGNN-RL) Graph convolution layer update: ; in, Representation Node The neighbor set of is the normalization constant.

[0069] Reinforcement learning Q value update: ; The graph convolution layer updates the state (representation) of each node through the adjacent node information, and uses the relationship between nodes to strengthen the graph structure features. At the same time, the reinforcement learning part selects actions through Q-learning optimization strategy to maximize the cumulative reward.

[0070] 9. Variational Inference Algorithm (VIA) Variational objective function: ; in, is the joint distribution, is the variational distribution, is the Kullback-Leibler divergence, which measures the difference between the variational distribution and the true posterior distribution; ; Variational inference is performed by optimizing the variational distribution To approximate the true posterior distribution and reduce the computational complexity of the inference process, Measures the gap between the variational distribution and the true distribution, and makes the two as close as possible during the optimization process.

[0071] 10. Dynamic Multimodal Visual Analysis Algorithm (DMVA) Feature Fusion: ; in, It is The characteristics of a mode, is the weight of the corresponding mode It is the fusion feature.

[0072] Timing Analysis: ; in, It's time The hidden state of RNN is a recurrent neural network. It's time visual features.

[0073] The features of multiple modalities are fused by weighted sum to obtain more comprehensive visual information. The time series analysis part processes time series data through RNN to capture dynamic changes. Specific embodiment three: like Figure 1-2 As shown, the following is a description of the specific application logic steps of the core modules and algorithms in the self-driving micro-robot for liver tumor targeted treatment and real-time monitoring system: 1. Self-driving navigation and positioning module Quantum Genetic Algorithm (QGA) application logic and steps: Initialization: Construct the internal environment of the liver into a three-dimensional spatial network, clarify the concepts of nodes and edges, and randomly generate an initial population based on quantum bit encoding. Each individual represents a possible path from the robot's starting point to the tumor location; Fitness evaluation: Formulate a fitness function that comprehensively considers factors such as path length, safety, and estimated arrival time, and calculate the fitness value of each individual based on this; Quantum revolving door operation: In each generation of evolution, based on the fitness information of the current individual and the global optimal individual, the chromosome is updated with the help of a quantum revolving door, and the parallelism of the quantum state is used to simultaneously explore multiple path combinations; Termination condition judgment: Set the maximum number of iterations or a satisfactory fitness threshold. When the conditions are met, output the chromosome corresponding to the optimal path, that is, the navigation path planning result.

[0075] Application logic and steps of fuzzy constraint optimization algorithm: Fuzzy variable definition: The elastic deformation of the vascular wall, tissue density inhomogeneity, and blood flow volatility are defined as fuzzy variables, and their corresponding fuzzy sets and membership functions are determined; Fuzzy constraint condition construction: Fuzzy constraint conditions are constructed with the robot's kinematic feasibility, safe distance from the vascular wall and surrounding tissues, and estimated time to reach the tumor as optimization goals; Path planning and adjustment: During the robot's navigation process, the sensor collects relevant information in real time and maps it to the fuzzy set to obtain the degree of satisfaction of the fuzzy constraint conditions. The path planning strategy is dynamically adjusted according to the fuzzy rules to ensure the robot's safe and accurate navigation and avoid collisions with important tissues.

[0076] 2. Intelligent Drug Delivery Module Application logic and steps of adaptive particle swarm optimization algorithm (LPSO) based on Lévy flight: Initialization: Determine the size of the particle swarm, randomly initialize the particle position (representing the combination of drug release dose and rate) and speed, and determine the particle value range and fitness function based on the drug diffusion in liver tissue, tumor cell uptake rate and metabolic kinetics; Particle update: Calculate the particle fitness value, update the individual optimal and global optimal positions, and update the particle speed and position in combination with the Lévy flight strategy to give particles a chance to jump out of the local optimal area; Fitness evaluation and adjustment: Evaluate the fitness of the updated particle position, update the individual and global optimal solutions if there are better solutions, and adaptively adjust the inertia weight and learning factor to balance the global and local search capabilities; Termination condition judgment: Set the maximum number of iterations or the satisfactory fitness threshold, and output the drug release strategy corresponding to the global optimal position when it is met, so as to improve the treatment effect and reduce toxic side effects.

[0077] Nonlinear feedback neural network application logic and steps: Network construction and initialization: Construct a multi-hidden layer neural network, determine the number of nodes in the input, hidden and output layers, and randomly initialize weights and thresholds; Data acquisition and preprocessing: Robot sensors collect tumor microenvironment and drug concentration information and preprocess it, such as normalization; Forward propagation calculation: Input the preprocessed data into the neural network, and obtain the output layer prediction value, such as tumor cell apoptosis rate and tumor volume change trend, through the hidden layer nonlinear activation function processing; Back propagation training: Based on the error between the output layer prediction value and the actual observation value, use the back propagation algorithm to calculate the gradient and update the weights and thresholds, and repeat the data acquisition and calculation training process until the network converges or reaches the preset number of training times; Drug delivery regulation: When the tumor environment changes, the neural network predicts the tumor response under different drug release strategies based on the training model, adjusts the drug release rate, and realizes intelligent regulation.

[0078] Multi-body dynamics simulation algorithm application logic and steps: Model establishment: construct a multi-body dynamics model of the robot, drug carrier and liver tissue, determine the rigid body properties, connection relationship and interaction force; parameter setting and initialization: set the model parameters according to the actual situation, initialize the state variables of the robot and drug carrier; simulation calculation: use the numerical integration method to solve the differential equations of the multi-body dynamics model, simulate the interaction process, calculate the rigid body force, update the speed and position in each time step, and consider the drug release dynamics; result analysis and optimization: analyze the simulation results to obtain information such as the robot's motion trajectory, drug release distribution and the impact on surrounding tissues, and optimize the robot structure design and drug release mechanism accordingly to ensure accurate drug delivery and stable and reliable release, such as selecting the optimal thruster design scheme.

[0079] 3. Tumor Monitoring and Status Assessment Module Bayesian Deep Convolutional Generative Adversarial Network (BD-CGAN) application logic and steps: Network construction: Construct a BD-CGAN network. The generator is based on a deep convolutional neural network and uses Bayesian reasoning to model the network parameters. Training data preparation: Collect and preprocess high-quality ultrasound image samples of liver tumors. Training process: The generator receives random noise vectors to generate tumor images, and the discriminator distinguishes between true and false images. The parameters of the two are adjusted through adversarial training. The goal of the generator is to generate images that are difficult to distinguish between true and false, and the goal of the discriminator is to make accurate judgments. The parameters are updated using the back propagation algorithm based on the loss function to improve the quality of the generated images. Image generation and application: In liver tumor ultrasound image monitoring, the trained BD-CGAN uses a small number of samples to generate more high-quality images, enhance image quality, reduce interference, and assist in diagnosis and treatment evaluation.

[0080] Variational inference algorithm application logic and steps: Probabilistic model construction: construct a probabilistic model for multimodal high-dimensional data of tumor monitoring, assuming that the observed data is generated by latent variables and establish a joint probability distribution; variational inference: introduce variational distribution to approximate the posterior probability, determine the variational distribution parameters by minimizing the KL divergence of the two, and iteratively update the parameters using the variational method and EM algorithm until convergence; data analysis and application: map the data to a low-dimensional latent space and perform operations such as clustering, classification and anomaly detection to mine the association between different data modes, identify information such as tumor subtypes, grades and metastasis risks, and comprehensively and accurately assess the tumor status.

[0081] Edge detection and feature fusion algorithm application logic and steps: Edge detection: Use gradient-based edge detection operators (such as Sobel and Canny operators) to detect edges of tumor images. The edge pixels are determined by calculating the gradient amplitude and direction. For images with different resolutions, the operator parameters can be adjusted or a multi-scale method can be used.

[0082] Feature extraction: After extracting edge features, further features such as the internal texture of the tumor (such as using the grayscale co-occurrence matrix method) and density (such as calculated based on grayscale value distribution or image segmentation) are obtained.

[0083] Feature fusion: Edge features are fused with internal features by methods such as splicing or weighted summation. The fused feature vector more comprehensively describes the tumor characteristics and is used for accurate diagnosis and treatment effect evaluation. For example, the changes in tumor characteristics before and after treatment can be compared to determine regression and treatment response.

[0084] 4. Multimodal Data Fusion Module Adaptive density clustering algorithm application logic and steps: Data preprocessing: Receive data from different sensors and clean it to remove errors and outliers.

[0085] Local density calculation: Calculate the local density around each data point, which can be based on distance or density estimation methods.

[0086] Determination of cluster centers: Data points with high local density and interconnectedness are divided into clusters. The point with the maximum local density can be selected as the cluster center or determined by other methods.

[0087] Cluster allocation and adjustment: Assign other data points to the nearest cluster. For low-density isolated points or noise data, adaptively adjust clustering parameters, assign them to the nearest cluster or mark them as abnormal, effectively integrate multimodal data information, and improve data fusion accuracy and reliability.

[0088] Reinforcement Learning and Graph Convolutional Network (RL-GCN) application logic and steps: Reinforcement learning part: define the agent (data fusion system), environment (multimodal data space), action (data fusion strategy selection) and reward function (indicator for evaluating the accuracy and usefulness of fused data). The agent selects strategies based on actions in the environment and learns the optimal data fusion strategy based on reward feedback.

[0089] Graph convolutional network: The liver anatomical structure information is constructed as a graph structure, where nodes represent organ tissues or regions, edges represent spatial proximity or functional connections, and sensor data is associated with corresponding graph nodes. Through the graph convolutional network information transmission mechanism, data spatial correlation is used for fusion processing. For example, when fusing liver vascular ultrasound data with tumor tissue metabolite concentration data, the data weight and fusion method are adaptively adjusted according to the spatial position relationship between blood vessels and tumor tissues, providing accurate information support for tumor diagnosis and treatment.

[0090] Application logic and steps of simulated annealing and multi-granularity convolutional network: Simulated annealing algorithm: During the data fusion process, as the iteration proceeds, poor fusion results are accepted with a certain probability, jumping out of the local optimum and exploring a better fusion solution; Multi-granularity convolutional network: For liver tumor CT image data, convolution operations are performed at the local pixel level, regional block level and overall image level at the same time to extract image features at different levels. Then, the simulated annealing algorithm is used to optimize the fusion weights of different granularity features to obtain a more comprehensive and representative fusion feature vector, thereby improving the accuracy of tumor diagnosis and treatment.

[0091] 5. Real-time self-learning decision module Temporal Generative Adversarial Network (TGAN) application logic and steps: Data preparation: Collect historical monitoring data such as the patient's past tumor marker change curves and ultrasound image sequences; Network construction and training: Construct a TGAN network, the generator generates future tumor development based on historical data, the discriminator judges the difference between the generated data and the real data, and adjusts the parameters through adversarial training to enable the generator to learn the timing and uncertainty of tumor development.

[0092] Prediction and decision-making: Based on the trained network, the patient's historical data is used to predict changes in tumor size and whether it has metastasized in the next week. Based on the prediction results, the treatment strategy is adjusted in real time, such as changing drug dosage, adjusting radiotherapy intensity, or planning surgical plans, to improve the timeliness and effectiveness of treatment.

[0093] Multifractal analysis algorithm application logic and steps: Data collection: Collect physiological signals of liver tumors (such as heart rate variability, blood pressure fluctuations, etc.) and tumor-related indicators (such as tumor volume change rate, tumor marker concentration fluctuations, etc.).

[0094] Feature analysis: Calculate the signal's fractal dimension, singular spectrum and other multi-fractal parameters to reveal the self-similarity and complexity of the signal at different time and space scales. For example, in tumor treatment, intermittent changes in tumor growth and changes in signal characteristics after treatment intervention are found. This information helps determine the tumor development trend and treatment effect, and provides a basis for subsequent treatment decisions.

[0095] Deep graph network reinforcement learning application logic and steps: Network construction: A deep graph network structure is constructed to represent the state space (including tumor size, location, pathological type, patient physical condition, etc.), action space (various treatment methods and their combinations), and transfer probability during liver tumor treatment.

[0096] Reinforcement learning process: The agent (decision-making system) explores different treatment strategies in the deep graph network based on the reinforcement learning algorithm, obtains rewards based on environmental feedback (such as tumor shrinkage, patient adverse reactions, etc.), and gradually learns the optimal treatment decision sequence. For example, in the face of complex liver tumor cases, considering tumor heterogeneity, individual differences of patients and dynamic changes in treatment, a personalized and real-time optimized treatment plan is formulated to improve treatment effects and reduce complications.

[0097] 6. Advanced Molecular Simulation and Drug Optimization Module Quantum Particle Swarm Optimization (QPSO) application logic and steps: Initialization: Combining quantum computing with particle swarm optimization algorithm, quantum bits are used to encode the molecular structural characteristics and chemical properties of drugs, such as bond length, bond angle, charge distribution, etc., to randomly initialize the position and speed of the particle swarm.

[0098] Particle update: Based on the particle swarm optimization algorithm framework, the particle position is updated by combining quantum gate operation and particle swarm information sharing mechanism. For example, for specific receptors on the surface of liver tumor cells, the drug molecule structure with high affinity binding is optimized by searching and designing, while considering metabolic stability and toxicity, to optimize the overall performance of drug molecules.

[0099] Fitness evaluation and optimization: Calculate the particle fitness value (such as drug targeting, activity and other indicators), update the individual optimal and global optimal positions based on the fitness, and continuously iterate and optimize until the termination conditions are met to obtain the drug molecular structure with the best targeting and activity, thereby improving the drug treatment effect.

[0100] Dynamic Markov decision process application logic and steps: Model construction: Construct a kinetic model for the delivery of drug molecules in the liver in vivo, consider the drug absorption, distribution, metabolism and excretion process as a Markov decision process, and determine the state variables (such as drug concentration in different tissues and organs, chemical form, etc.) and actions (such as drug delivery methods, including injection site, rate, dosage form selection, etc.).

[0101] Probability and reward calculation: Based on the pharmacokinetic model, the transition probability and reward function of taking different actions under different states are calculated, such as the accumulation of drugs in tumor tissues and their toxicity to normal tissues.

[0102] Decision-making and solving: Determine the optimal drug delivery route and strategy by solving the kinetic Markov decision process. For example, when designing a chemotherapy drug delivery program for liver tumors, the injection method, time interval and dosage adjustment strategy are determined based on liver blood perfusion, tumor vascular distribution and pharmacokinetic characteristics to achieve maximum drug accumulation at the tumor site and minimum side effects.

[0103] Molecular dynamics simulation algorithm application logic and steps: Model building: Establish a detailed molecular model, including drug molecules, liver tumor cell targets and surrounding biological molecules (such as proteins, nucleic acids, etc.).

[0104] Simulation calculation: simulate the movement trajectory of drug molecules in physiological environment, the process of binding with targets and the impact on intracellular signal pathways. For example, simulate the binding of anticancer drugs to specific enzyme proteins in tumor cells, analyze the protein conformation before and after binding, changes in active sites and the blocking effect on downstream signal transduction.

[0105] Application of results: Provide a microscopic theoretical basis for drug optimization design, help design more efficient and specific drug molecules, and help understand the drug action mechanism and potential adverse reactions, and support drug development and clinical application. Specific embodiment four: like Figure 1-2 As shown, the following is the attached Figure 2 The simulation diagram in the figure is explained in detail: From the attached Figure 2 In the figure, we can see the simulation results (y-axis) of different algorithms (LSTM, RL, GA, ACO, PSO) in the same data range (x-axis: 0 to 10). Each curve represents the output of an algorithm, and different colors in the figure correspond to different algorithms. The following is a detailed analysis of each curve: 1.LSTM (red curve): Trend: The output of LSTM shows a curve with large fluctuations, accompanied by smaller fluctuations.

[0107] Morphology: It is highly volatile, shows periodic changes (similar to a sine wave), and has a certain noise component (random fluctuations), which means that LSTM can capture certain pattern changes when simulating data, but it is unstable.

[0108] Possible cause: LSTM (Long Short-Term Memory) is an algorithm for modeling time series data. In this application, it may be used to predict a system with periodic changes, but due to the noise in the data itself or the lack of training data, it exhibits this kind of fluctuation.

[0109] 2.RL (green curve): Trend: The RL curve shows a certain sinusoidal waveform, but the amplitude is small and relatively stable.

[0110] Morphology: Compared with LSTM, the curve of RL fluctuates less and shows a more gentle fluctuation. RL (reinforcement learning) algorithms are often used for decision optimization, which may reflect a more stable and adaptable strategy, aiming to reduce large fluctuations and achieve a certain optimization effect.

[0111] Possible reasons: The characteristic of RL is to optimize strategies through rewards and penalties. Especially in the decision-making process, it usually tends to avoid extreme fluctuations and presents a smooth curve.

[0112] 3.GA (blue curve): Trend: The curve of GA shows a gradually increasing trend with a large fluctuation range.

[0113] Shape: The curve looks like a gradually increasing trend. As the x value increases, the y value gradually increases, but it is also accompanied by large fluctuations. This may indicate that the genetic algorithm (GA) may have multiple fluctuations in the local optimal solution during the optimization process, showing an increasing trend in some areas and a certain oscillation in other areas.

[0114] Possible reasons: Genetic algorithms are optimization methods based on population search, which usually undergo multiple generations of evolution and produce certain local fluctuations, especially in the early stages of exploring the space.

[0115] 4.ACO (magenta curve): Trend: The ACO curve shows a gradually increasing trend, and the fluctuation is relatively large.

[0116] Shape: The curve is similar to GA, but the increment is more dramatic. The Ant Colony Algorithm (ACO) searches for the optimal solution by simulating the foraging process of ants. The curve reflects the large fluctuations in the path-finding process. Especially when solving complex problems, ACO often experiences large search fluctuations.

[0117] Possible cause: When the ant colony algorithm is exploring solutions, over time, ants will adjust their paths based on the concentration of pheromones, which may cause large oscillations, especially in the early stages of the algorithm.

[0118] 5. PSO (cyan curve): Trend: The PSO curve shows a decreasing trend, and the fluctuation is relatively stable.

[0119] Morphology: PSO (particle swarm optimization) algorithm is usually used for optimization problems in continuous space, showing a relatively stable downward trend. The fluctuation amplitude of the curve is small, indicating that the particle swarm has made slight adjustments near the global optimal solution and tends to be stable.

[0120] Possible reasons: The search strategy of the particle swarm optimization algorithm enables particles to quickly converge to a better solution during the search process, thus showing a relatively stable downward trend and less drastic fluctuations.

[0121] Summarize: Volatility and stability: LSTM and GA show large volatility, indicating that these algorithms may conduct more intense exploration in the early stages or in certain areas. In contrast, RL and PSO show a relatively stable trend, indicating that they tend to avoid excessive fluctuations and strive for stability during the optimization process.

[0122] Increasing and decreasing trends: The ACO and GA curves show a gradually increasing trend, indicating that in some optimization problems, they may present a pattern of continuous optimization; while PSO shows a decreasing trend, indicating that it gradually reduces the error when seeking the optimal solution.

[0123] Applicability analysis: For problems that require exploration or optimization in unstable environments (such as tasks simulated by LSTM and GA), LSTM and GA may perform more effectively; for tasks that require stability and gradual optimization (such as RL and PSO), RL and PSO may perform better.

[0124] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprising a reference structure" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0125] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A self-driving microrobot for targeted liver tumor treatment and real-time monitoring system, including a monitoring main system, characterized in that: The monitoring main system includes a self-driving navigation and positioning module, an intelligent drug delivery module, a tumor monitoring and status assessment module, a multimodal data fusion module, a real-time self-learning decision module, an advanced molecular simulation and drug optimization module, a secure communication module, a power grid environment adaptation module, and a fault detection and recovery module, wherein: The self-driving navigation and positioning module includes a quantum genetic algorithm, a fuzzy constraint optimization algorithm and an adaptive simulated annealing algorithm, which are used to achieve accurate navigation and obstacle avoidance of the robot in the complex environment of the liver. The formula of the fuzzy constraint optimization algorithm is as follows: Fuzzy constraint processing: ； in, is the constraint function that represents the constraints of the optimization problem. is the fuzzification factor, is the threshold value; Objective function optimization: ； in, are the coefficients of the objective function, is the weight of the fuzzy constraint, represents the fuzzy form of the constraint function, is the objective function; The intelligent drug delivery module includes an adaptive particle swarm optimization algorithm based on Lévy flight, a nonlinear feedback neural network, and a multibody dynamics simulation algorithm to optimize the dosage and strategy of drug release; The tumor monitoring and status assessment module includes a Bayesian deep convolutional generative adversarial network, a variational inference algorithm, and an edge detection and feature fusion algorithm, which are used for synthesis, repair, enhancement, and real-time assessment of tumor images; The multimodal data fusion module includes an adaptive density clustering algorithm, reinforcement learning and graph convolutional networks, and simulated annealing and multi-granularity convolutional networks, which are used to efficiently fuse data from different sensors and improve the accuracy of tumor diagnosis and treatment; The real-time self-learning decision module includes a multi-fractal analysis algorithm, deep graph network reinforcement learning, and a temporal generative adversarial network for real-time optimization of treatment decisions; The advanced molecular simulation and drug optimization module includes a quantum particle swarm optimization algorithm, a kinetic Markov decision process and a molecular dynamics simulation algorithm, which are used for the design of drug molecules and the optimization of delivery pathways.

2. The self-driving microrobot for targeted liver tumor treatment and real-time monitoring system according to claim 1 is characterized in that: The quantum genetic algorithm in the self-driving navigation and positioning module accelerates path search through quantum computing, thereby improving the accuracy and efficiency of navigation path planning. The quantum genetic algorithm performs encoding operations based on the superposition state and entanglement characteristics of quantum bits. In the path planning process, the internal environment of the liver is abstracted into a three-dimensional spatial network, in which nodes represent vascular branch points and tissue gap key points, and edges represent accessible paths between nodes. The quantum genetic algorithm updates chromosomes through quantum rotating gate operations. The fuzzy constraint optimization algorithm models uncertain factors through fuzzy set theory, in which uncertain factors include elastic deformation of vascular walls, uneven tissue density, and volatility of blood flow. In the application process, the elastic deformation of vascular walls, uneven tissue density, and volatility of blood flow are defined as fuzzy variables, respectively, to construct fuzzy constraints. In path planning, the robot's kinematic feasibility, safe distance from vascular walls and surrounding tissues, and estimated time to reach the tumor are used as optimization goals. The fuzzy membership function is used to quantitatively evaluate the degree of satisfaction of each constraint. When the robot navigates in the blood vessel, the path planning strategy is dynamically adjusted according to the information fed back by the sensor in real time. The quantum genetic algorithm is as follows: Qubit representation: ； in, and is the ground state of the qubit, and is a complex number that satisfies the normalization condition ; Fitness evaluation: In each generation, according to the fitness function Evaluate the pros and cons of each quantum individual: ； in, It is decision variables, is its corresponding weight, represents the probability amplitude of the two ground states in the quantum superposition state, which determines the state of the quantum bit, is the fitness function, which is used to evaluate the quality of the current solution.

3. The self-driving microrobot for targeted liver tumor treatment and real-time monitoring system according to claim 1 is characterized in that: The Lévy flight adaptive particle swarm optimization algorithm in the intelligent drug delivery module is used to accelerate the search for the optimal dose and rate of drug release and improve the therapeutic effect of the drug. The Lévy flight adaptive particle swarm optimization algorithm gives the particle swarm the ability to make long-distance jumps in the search space, breaking through the limitation of the traditional particle swarm optimization algorithm that is prone to fall into local optimality. During drug delivery, the Lévy flight adaptive particle swarm optimization algorithm uses the release dose and rate of the drug carried by the robot as the optimization target based on three factors: the diffusion model of the drug in the liver tissue, the uptake rate of the drug by tumor cells, and the metabolic kinetics of the drug. When in use, by simulating the diffusion process of drug molecules in the liver blood vessels and tissue gaps, the Lévy flight adaptive particle swarm optimization algorithm quickly searches for the optimal drug release strategy under the tumor size, location and pathological state, accelerates the search for the optimal dose and rate of drug release, improves the therapeutic effect of the drug, and reduces the toxic side effects on normal liver tissue.

4. The self-driving microrobot for targeted liver tumor treatment and real-time monitoring system according to claim 1 is characterized in that: The Bayesian deep convolutional generative adversarial network in the tumor monitoring and state assessment module is used to generate tumor images that are close to reality, enhance image quality and reduce interference factors. The Bayesian deep convolutional generative adversarial network consists of two parts: a generator and a discriminator. The generator is based on a deep convolutional neural network structure and uses Bayesian reasoning to model the uncertainty of network parameters. During the training process, the generator receives a random noise vector as input and attempts to generate a tumor image that is close to reality. The discriminator distinguishes and judges the generated image from the real tumor image. Through the adversarial training of the two, the quality of the image generated by the generator is continuously improved. In actual operation, in the ultrasound image monitoring of liver tumors, the Bayesian deep convolutional generative adversarial network uses high-quality ultrasound image samples for training.

5. The self-driving microrobot for targeted liver tumor treatment and real-time monitoring system according to claim 1 is characterized in that: The reinforcement learning and graph convolution network in the multimodal data fusion module are adaptively adjusted according to the spatial relationship of different sensor data to improve the accuracy of data fusion. The adaptive density clustering algorithm identifies local density changes in data from ultrasonic sensors, optical sensors, and electrochemical sensors to determine cluster centers. In actual operation, when fusing ultrasonic imaging data of liver tumors and tumor marker concentration data detected by electrochemical sensors, the local density around each data point is first analyzed, and data points with high density and interconnectedness are divided into a cluster. For isolated points with low density and noise data, clustering parameters are adaptively adjusted to assign them to the nearest cluster or mark them as abnormal data.

6. The self-driving microrobot for targeted liver tumor treatment and real-time monitoring system according to claim 1, characterized in that: The temporal generative adversarial network in the real-time self-learning decision module predicts the future evolution of the tumor, dynamically adjusts the treatment plan, models and predicts the development process of liver tumors, and predicts the size change of the tumor in the next week and whether metastasis will occur based on the patient's tumor marker change curve and ultrasound image sequence in the past month.

7. The self-driving microrobot for liver tumor targeted treatment and real-time monitoring system according to claim 1 is characterized in that: The quantum particle swarm optimization algorithm in the advanced molecular simulation and drug optimization module optimizes the design of drug molecules through quantum computing, thereby improving the targeting and therapeutic effects of drugs. The advanced molecular simulation and drug optimization module combines the characteristics of quantum computing with the particle swarm optimization algorithm. In terms of drug molecule design, quantum bits are used to encode the structural characteristics and chemical properties of drug molecules, including the bond length, bond angle and charge distribution of the molecules. Through quantum gate operations and the information sharing mechanism of particle swarms, the drug molecular structure with the best targeting and activity is searched in the chemical space. For specific receptors on the surface of liver tumor cells, the quantum particle swarm optimization algorithm can optimize the design of drug molecular structures that bind to them with high affinity, improve the targeting of drugs, and at the same time consider the metabolic stability and toxicity of drug molecules, optimize the overall performance of drug molecules, and improve the therapeutic effects of drugs.

8. The self-driving microrobot for liver tumor targeted treatment and real-time monitoring system according to claim 1 is characterized in that: The secure communication module uses the Advanced Encryption Standard encryption algorithm and elliptic curve cryptography to encrypt the data transmitted between the robot and the external monitoring device, ensuring the confidentiality of the data during transmission and preventing the patient's privacy information and treatment monitoring data from being stolen. At the same time, the message authentication code technology is used to verify the integrity of the data. During each data transmission, the sender uses the key to generate a message authentication code and attaches it to the data. The receiver uses the same key to verify the received data. If the message authentication code does not match, it indicates that the data may have been tampered with. The receiver will refuse to receive the data and notify the sender to resend it, thereby ensuring the integrity and reliability of the data and ensuring the safe and stable operation of the entire treatment monitoring process.

9. The self-driving microrobot for liver tumor targeted treatment and real-time monitoring system according to claim 1, characterized in that: The grid environment adaptation module is equipped with a voltage regulating device to monitor the fluctuation of the grid voltage in real time. When the grid voltage is higher than a threshold or lower than a threshold, the voltage is adjusted to the range required for the normal operation of the robot through an automatic transformer.

10. The self-driving microrobot for liver tumor targeted treatment and real-time monitoring system according to claim 1, characterized in that: The fault detection and recovery module monitors the system operating status in real time, detects potential faults, and automatically restores system functions when a fault occurs.

Images