Neurosurgical operation planning and risk early warning method and system based on multiple agents

Abstract

The invention discloses a neurosurgery operation planning and risk early warning method based on multiple agents. The neurosurgery operation planning and risk early warning method comprises the following steps that (1) preoperative images and intraoperative video streams / electrophysiological signals and expert knowledge data are collected through a data input module; (2) the digital agent management center schedules a preoperative modeling agent, completes skull base three-dimensional reconstruction, risk assessment and path planning, and outputs a VTK model and a path file; (3) processing a video stream in real time by the intra-operative sensing agent, and completing scene segmentation, dynamic three-dimensional modeling and precise registration with a preoperative model; (4) providing active decision support, risk early warning and voice interaction guidance by the operation guidance agent in combination with intra-operative data and an expert knowledge base; and (5) the integration module generates a real-time report, the doctor feeds back adjustment suggestions, and the system updates the knowledge base and optimizes the agent model. The invention further provides a neurosurgical operation planning and risk early warning system based on the multiple agents. The safety and effectiveness of the skull base neurosurgery operation are improved.

Description

Technical Field

[0001] This invention belongs to the field of medical artificial intelligence and neurosurgical navigation technology, and in particular relates to a method and system for neurosurgical planning and risk warning based on multi-agent, which is applicable to complex neurosurgical scenarios such as skull base tumors. Background Technology

[0002] Skull base neurosurgery, due to its complex anatomical structure (including dense cranial nerves, blood vessels, and irregular skull) and limited surgical space, demands extremely high precision in preoperative path planning, real-time intraoperative scene awareness, and proactive surgical guidance. Traditional neuronavigation suffers from three core problems: preoperative planning relies on experience, specifying paths solely based on 3D images without incorporating individualized skull base anatomical landmarks, resulting in insufficient consistency between the path and expert experience; it depends on preoperative static images and instrument positioning, failing to process dynamic information in the surgical video stream in real time and making it difficult to construct an intraoperative 3D scene; and it only provides instrument position feedback, lacking proactive decision support based on multi-source information, thus failing to meet the needs for real-time risk warning and operational guidance in complex surgeries.

[0003] With the development of artificial intelligence and multi-agent technology, decision-making and interaction achieved through agent collaboration have made it possible to provide more accurate planning and more effective risk warning for neurosurgery. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, this invention provides a method and system for neurosurgical planning and risk warning based on multi-agent systems. By working collaboratively with multiple agents, it integrates preoperative three-dimensional modeling of the skull base, intraoperative visual dynamic perception, and active surgical guidance functions, solving the problems of low accuracy, delayed perception, and passive guidance in traditional navigation systems, thereby improving the safety and effectiveness of skull base neurosurgery.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A multi-agent-based neurosurgical procedure planning and risk warning method includes the following steps:

[0007] (1) Multi-source data acquisition: Preoperative images (sMRI / dMRI / CT), intraoperative video streams / electrophysiological signals and expert knowledge data are acquired through the data input module;

[0008] (2) Preoperative modeling: The digital intelligent agent management center schedules the preoperative modeling intelligent agent to complete the three-dimensional reconstruction of the skull base, risk assessment and path planning, and outputs the VTK model and path file;

[0009] (3) Intraoperative perception: The intraoperative perception agent processes the video stream in real time, completes scene segmentation, dynamic 3D modeling and accurate registration with the preoperative model;

[0010] (4) Surgical guidance: The surgical guidance agent combines intraoperative data and expert knowledge base to provide proactive decision support, risk warning and voice interaction guidance;

[0011] (5) Information integration and feedback: The integration module generates real-time reports, doctors provide feedback and adjustment suggestions, the system updates the knowledge base and optimizes the intelligent agent model.

[0012] Furthermore, in (2), the processing procedure of the preoperative modeling agent is as follows:

[0013] 2.1) Multimodal reconstruction: Using FSL and MRtrix tools, sMRI / dMRI data were preprocessed (head motion correction, eddy current correction), and MNI standard neural atlas registration was performed to extract the ROI (region of interest, such as cranial nerves, blood vessels, and skull landmarks) at the skull base and generate a mask.

[0014] 2.2) Path planning: Based on the bony landmarks of the skull (such as the carotid artery carina and tuberculum sellae), individualized candidate bone windows are generated. A risk assessment model is constructed through a twin neural network (inputting the geometric / anatomical features of the path and outputting the risk coefficient). The size of the bone window and the path are dynamically adjusted to ensure that the planned path is consistent with the experts by ≥80% and the algorithm response time is ≤5s.

[0015] 2.3) Output results: Generate a 3D model of the skull base (including lesions, nerves, and blood vessels) in VTK format and a surgical path file, and synchronize it to the navigation system.

[0016] Furthermore, in (3), the processing procedure of the intraoperative sensing agent is as follows:

[0017] 3.1) Scene Segmentation: Based on fine-tuning of the SAM2 large model, real-time segmentation is performed on the surgical video stream to identify surgical instruments (such as bipolar electrocoagulation forceps) and key brain tissues (such as tumors and cranial nerves). The segmentation DICE coefficient is ≥85%, and the object recognition accuracy is ≥95%.

[0018] 3.2) 3D modeling: The depth information of the video stream is mined by the Transformer-Based stereo depth estimation module, and combined with surfel scene representation and camera pose estimation to construct a dynamic 3D scene during surgery;

[0019] 3.3) Real-time registration: The "probabilistic key point detection (GCNN network) + thin plate spline transformation (TPS)" technology is used to align the intraoperative 3D model with the preoperative VTK model, with a registration accuracy of ≥85% and ensuring that the anatomical structure mapping error is ≤2mm.

[0020] Furthermore, in step (4), the processing procedure of the surgical guidance agent is as follows:

[0021] 4.1) Decision support: Based on a large model in the vertical field of neurosurgery (Qwen series fine-tuning), and integrating an expert knowledge base (including knowledge from more than 200 surgical videos), it provides recommendations for surgical approaches and instrument operation suggestions, with a decision plan consistency of ≥80% with experts;

[0022] 4.2) Risk warning: Real-time monitoring of the distance between the device and key nerves (<1mm triggers Level 1 warning, <0.5mm triggers Level 2 warning), and neurophysiological signals (EMG abnormal discharge >100μV alarm), with a warning response delay ≤1000ms;

[0023] 4.3) Interactive guidance: Through the voice interaction module (Whisper model to realize voice-to-text conversion), it receives doctors' instructions and provides feedback on visual results (such as lesion size, instrument travel distance), and supports multi-threaded parallel response.

[0024] A multi-agent-based neurosurgical procedure planning and risk warning system includes:

[0025] The multimodal data input module is used to acquire preoperative images (sMRI / dMRI / CT), intraoperative video streams / electrophysiological signals, and expert knowledge data;

[0026] The core module for intelligent agent scheduling includes a preoperative modeling intelligent agent, an intraoperative perception intelligent agent, and a surgical guidance intelligent agent. The preoperative modeling intelligent agent, based on multimodal image data, realizes three-dimensional reconstruction of the facial and auditory nerves and skull base structures, as well as surgical path planning. The intraoperative perception intelligent agent processes intraoperative video streams and real-time data, completing surgical scene segmentation, three-dimensional modeling, and registration with the preoperative model. The surgical guidance intelligent agent, combining an expert knowledge base and intraoperative data, provides proactive decision support, facial and auditory nerve protection risk warnings, and operational guidance.

[0027] The multi-source information integration module is used to integrate the output of the intelligent agent and generate reports;

[0028] The dynamic update and feedback module is used to support the real-time writing of intraoperative data into the expert knowledge base (such as adding new surgical cases), receive feedback from doctors on adjustments to the path / decision, and use it for iterative optimization of the agent model (such as updating the parameters of the risk assessment model).

[0029] The beneficial effects of this invention are mainly reflected in:

[0030] 1. Precise planning enhances surgical safety: By integrating multimodal imaging data through preoperative modeling intelligence, combined with FSL, MRtrix tools and twin neural networks, high-fidelity reconstruction of the facial and auditory nerves and individualized risk quantification path planning are achieved. The planned path is consistent with the expert's ≥80%, and the instrument positioning accuracy is ≤2mm, which greatly reduces the risk of facial and auditory nerve damage and helps to preserve the patient's residual hearing and facial nerve function.

[0031] 2. Real-time perception and adaptation to intraoperative dynamics: The intraoperative perception agent is based on the SAM2 large model and Transformer-Based depth estimation technology to realize real-time segmentation and dynamic 3D modeling of the surgical scene. The GCNN+TPS registration technology ensures accurate alignment between the intraoperative and preoperative models, effectively dealing with intraoperative interference such as tissue deformation and surgical smoke.

[0032] 3. Proactive guidance optimizes the surgical experience: The surgical guidance agent integrates a large model of the vertical domain and an expert knowledge base to provide proactive decision support, multi-dimensional risk warning and voice interaction functions. It breaks through the limitations of passive feedback in traditional navigation, reduces the difficulty of operation for doctors, and improves the efficiency and accuracy of complex skull base surgery. Attached Figure Description

[0033] Figure 1 This is an architecture diagram of a multi-agent neurosurgical planning and risk warning system.

[0034] Figure 2 It is a flowchart for modeling intelligent agents.

[0035] Figure 3 This is a flowchart of a perceptual intelligent agent.

[0036] Figure 4 This is a flowchart of the surgical guidance intelligent agent. Detailed Implementation

[0037] The present invention will now be further described with reference to the accompanying drawings.

[0038] Reference Figures 1-4 A method for neurosurgical planning and risk warning based on multi-agent systems includes the following steps:

[0039] (1) Multi-source data acquisition: Preoperative images (sMRI / dMRI / CT), intraoperative video streams / electrophysiological signals and expert knowledge data are acquired through the data input module;

[0040] (2) Preoperative modeling: The digital intelligent agent management center schedules the preoperative modeling intelligent agent to complete the three-dimensional reconstruction of the skull base, risk assessment and path planning, and outputs the VTK model and path file;

[0041] In embodiment (2), the processing procedure of the preoperative modeling agent is as follows:

[0042] 2.1) Multimodal reconstruction: Using FSL and MRtrix tools, sMRI / dMRI data were preprocessed (head motion correction, eddy current correction), and MNI standard neural atlas registration was performed to extract the ROI (region of interest, such as cranial nerves, blood vessels, and skull landmarks) at the skull base and generate a mask.

[0043] 2.2) Path planning: Based on the bony landmarks of the skull (such as the carotid artery carina and tuberculum sellae), individualized candidate bone windows are generated. A risk assessment model is constructed through a twin neural network (inputting the geometric / anatomical features of the path and outputting the risk coefficient). The size of the bone window and the path are dynamically adjusted to ensure that the planned path is consistent with the experts by ≥80% and the algorithm response time is ≤5s.

[0044] 2.3) Output results: Generate a 3D model of the skull base (including lesions, nerves, and blood vessels) in VTK format and a surgical path file, and synchronize it to the navigation system.

[0045] (3) Intraoperative perception: The intraoperative perception agent processes the video stream in real time, completes scene segmentation, dynamic 3D modeling and accurate registration with the preoperative model;

[0046] In embodiment (3), the processing procedure of the intraoperative sensing agent is as follows:

[0047] 3.1) Scene Segmentation: Based on fine-tuning of the SAM2 large model, real-time segmentation is performed on the surgical video stream to identify surgical instruments (such as bipolar electrocoagulation forceps) and key brain tissues (such as tumors and cranial nerves). The segmentation DICE coefficient is ≥85%, and the object recognition accuracy is ≥95%.

[0048] 3.2) 3D modeling: The depth information of the video stream is mined by the Transformer-Based stereo depth estimation module, and combined with surfel scene representation and camera pose estimation to construct a dynamic 3D scene during surgery;

[0049] 3.3) Real-time registration: The "probabilistic key point detection (GCNN network) + thin plate spline transformation (TPS)" technology is used to align the intraoperative 3D model with the preoperative VTK model, with a registration accuracy of ≥85% and ensuring that the anatomical structure mapping error is ≤2mm.

[0050] (4) Surgical guidance: The surgical guidance agent combines intraoperative data and expert knowledge base to provide proactive decision support, risk warning and voice interaction guidance;

[0051] In embodiment (4), the processing procedure of the surgical guidance agent is as follows:

[0052] 4.1) Decision support: Based on a large model in the vertical field of neurosurgery (Qwen series fine-tuning), and integrating an expert knowledge base (including knowledge from more than 200 surgical videos), it provides recommendations for surgical approaches and instrument operation suggestions, with a decision plan consistency of ≥80% with experts;

[0053] 4.2) Risk warning: Real-time monitoring of the distance between the device and key nerves (<1mm triggers Level 1 warning, <0.5mm triggers Level 2 warning), and neurophysiological signals (EMG abnormal discharge >100μV alarm), with a warning response delay ≤1000ms;

[0054] 4.3) Interactive guidance: Through the voice interaction module (Whisper model to realize voice-to-text conversion), it receives doctors' instructions and provides feedback on visual results (such as lesion size, instrument travel distance), and supports multi-threaded parallel response.

[0055] (5) Information integration and feedback: The integration module generates real-time reports, doctors provide feedback and adjustment suggestions, the system updates the knowledge base and optimizes the intelligent agent model.

[0056] A multi-agent-based neurosurgical procedure planning and risk warning system includes:

[0057] The multimodal data input module is used to acquire preoperative images (sMRI / dMRI / CT), intraoperative video streams / electrophysiological signals, and expert knowledge data;

[0058] Preoperative imaging data in this embodiment, including sMRI (structural magnetic resonance imaging), dMRI (diffusion tensor imaging), and high-resolution CT temporal bone scan, are used for reconstruction of the facial and auditory nerves and skull base anatomy.

[0059] Intraoperative real-time data: surgical video stream (endoscopic / extrascopic), neurophysiological signals (such as facial nerve EMG, auditory brainstem response ABR), instrument positioning data (acquired by optical positioning system);

[0060] Expert knowledge database: Neurosurgical guidelines, expert experience in protecting the facial and auditory nerves, more than 20,000 Q&A entries and more than 20,000 instruction entries (building an expert knowledge base).

[0061] The core module for intelligent agent scheduling includes a preoperative modeling intelligent agent, an intraoperative perception intelligent agent, and a surgical guidance intelligent agent. The preoperative modeling intelligent agent, based on multimodal image data, realizes three-dimensional reconstruction of the facial and auditory nerves and skull base structures, as well as surgical path planning. The intraoperative perception intelligent agent processes intraoperative video streams and real-time data, completing surgical scene segmentation, three-dimensional modeling, and registration with the preoperative model. The surgical guidance intelligent agent, combining an expert knowledge base and intraoperative data, provides proactive decision support, facial and auditory nerve protection risk warnings, and operational guidance.

[0062] The multi-source information integration module is used to integrate the output of the intelligent agent and generate two types of reports;

[0063] Written report: includes preoperative pathway parameters, intraoperative risk thresholds, and key points of operation guidance (such as "maintain a distance of ≥1mm from the facial nerve when separating the tumor");

[0064] Visualized report: A 3D rendered view of the fused "preoperative model - intraoperative scene - surgical path" is displayed, with warning areas (within 1mm of the nerve) highlighted.

[0065] The dynamic update and feedback module is used to support the real-time writing of intraoperative data into the expert knowledge base (such as adding new surgical cases), receive feedback from doctors on adjustments to the path / decision, and use it for iterative optimization of the agent model (such as updating the parameters of the risk assessment model).

[0066] This embodiment uses "right acoustic neuroma resection via retrosigmoid approach" as an example to illustrate the implementation process of the present invention in detail:

[0067] Step (1) Multi-source data acquisition, as follows:

[0068] 1.1) Preoperative data: 3.0T MRI data of the patient were acquired (sMRI: slice thickness 0.8mm; dMRI: b-value = 1000s / mm). 2 The high-resolution CT temporal bone scan data (with 32 diffusion directions) (locating the opening of the internal auditory canal and the course of the facial nerve canal) is uploaded to the data input module.

[0069] 1.2) Expert data: Enter knowledge such as "Guidelines for the protection of facial and auditory nerves in acoustic neuroma surgery" and "Standards for identifying anatomical landmarks in the retrosigmoid approach", and supplement 20,000 Q&A data (such as "the positional relationship between the facial nerve and the auditory nerve in the internal auditory canal");

[0070] 1.3) Intraoperative equipment: Connect endoscope (to acquire surgical video stream), optical positioning system (to monitor the position of microsurgical instruments), EMG / ABR monitor, preset data interface, and enter the doctor's special requirements such as "patient has moderate sensorineural hearing loss before surgery and needs to preserve residual hearing".

[0071] Step (2) Preoperative modeling of the intelligent agent is as follows:

[0072] 2.1) Multimodal reconstruction: FSL 6.0 was used to preprocess the dMRI. The MRtrix tool was used to reconstruct the fiber bundles with the mask as the seed point (the facial nerve originates from the brainstem and extends to the internal auditory canal, and the auditory nerve extends from the internal auditory canal to the brainstem). A 3D model of the facial and auditory nerves in VTK format was generated, clearly marking the anatomical relationship of the facial nerve being located on the ventral side of the tumor and the auditory nerve being partially wrapped by the tumor.

[0073] 2.2) Path planning: Three candidate bone windows were generated based on bony landmarks at the posterior margin of the internal auditory canal and the anterior wall of the sigmoid sinus. These windows were input into a twin neural network (the training set contained surgical paths for 500 cases of acoustic neuroma) to calculate the risk coefficient of facial and auditory nerve injury. Among them, the "retrosigmoid approach" had the lowest risk (the shortest distance between the path and the facial nerve was 1.5 mm) and could preserve hearing-related structures to the greatest extent. It was determined to be the optimal path. The planned path was consistent with the surgeon's plan in 83% of cases, and the response time was 4.1 seconds.

[0074] 2.3) Output results: The VTK model and path file (starting point: posterior margin of the sigmoid sinus on the skull surface, ending point: surface of the tumor) are synchronized to the intraoperative navigation system, and the shortest distance between the tumor and the facial nerve is calculated to be 1.8 mm.

[0075] Step (3) The intraoperative sensing agent operates as follows:

[0076] 3.1) Scene segmentation: The SAM2 fine-tuning model segments the endoscopic video stream in real time, identifying bipolar electrocoagulation forceps (DICE coefficient 88%), acoustic neuroma (DICE coefficient 86%), and facial nerve (DICE coefficient 85%), with a processing time of 0.2s per frame, effectively filtering out surgical smoke interference;

[0077] 3.2) Three-dimensional modeling: The stereo depth estimation module calculates the depth of the instrument tip to be 6.2mm, and combines it with Surfel representation to construct an intraoperative three-dimensional scene, displaying the dynamic adhesion relationship between the tumor and the facial nerve in real time;

[0078] 3.3) Real-time registration: GCNN network detects corresponding key points during / before surgery (such as the floor of the internal auditory canal and the opening of the facial nerve canal), and TPS transformation is used to achieve model alignment. The registration accuracy is 89%, and the facial and auditory nerve mapping error is 1.7mm, ensuring accurate intraoperative anatomical positioning.

[0079] Step (4) The surgery guides the intelligent agent to work as follows:

[0080] 4.1) Decision support: Doctors use voice commands to "assess the safety of the current separation plan". The AI ​​agent calls up the knowledge base and provides feedback such as "It is recommended to use blunt dissection on the ventral side of the tumor, maintain a distance of ≥1mm from the facial nerve, and avoid electrocoagulation thermal damage". The decision plan is consistent with the expert's 82%;

[0081] 4.2) Risk warning: When the instrument approaches the facial nerve (0.9mm away), the system triggers a level one warning (yellow highlight + voice prompt "Approaching the facial nerve, please pay attention to operation"); when the intraoperative EMG detects an abnormal discharge of 130μV, the instrument navigation is paused and the system prompts "The facial nerve is stimulated, it is recommended to adjust the operation angle";

[0082] 4.3) Interactive guidance: When the doctor asks "distance between the auditory nerve and the instrument", the intelligent agent calculates the distance through the intraoperative model and provides voice feedback "current distance 1.2mm, within the safe range", while marking the distance data and the direction of the auditory nerve on the navigation interface.

[0083] Step (5) Dynamic updates and feedback are as follows:

[0084] After the surgery, the system automatically writes the surgical video (including annotations of the facial nerve adhesion separation operation) and path adjustment records into the knowledge base. The doctor reports that "the facial and auditory nerves were accurately located during the operation and the warning was timely." Based on this feedback, the system optimizes the regularization parameters of the TPS transform and the FA threshold of the risk assessment model to improve the navigation accuracy of subsequent acoustic neuroma surgeries.

[0085] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for neurosurgical planning and risk warning based on multi-agent systems, characterized in that, The method includes the following steps: (1) Multi-source data acquisition: Preoperative images (sMRI / dMRI / CT), intraoperative video streams / electrophysiological signals and expert knowledge data are acquired through the data input module; (2) Preoperative modeling: The digital intelligent agent management center schedules the preoperative modeling intelligent agent to complete the three-dimensional reconstruction of the skull base, risk assessment and path planning, and outputs the VTK model and path file; (3) Intraoperative perception: The intraoperative perception agent processes the video stream in real time, completes scene segmentation, dynamic 3D modeling and accurate registration with the preoperative model; (4) Surgical guidance: The surgical guidance agent combines intraoperative data and expert knowledge base to provide proactive decision support, risk warning and voice interaction guidance; (5) Information integration and feedback: The integration module generates real-time reports, doctors provide feedback and adjustment suggestions, the system updates the knowledge base and optimizes the intelligent agent model.

2. The method for neurosurgical planning and risk warning based on multi-agent technology as described in claim 1, characterized in that, In (2), the processing procedure of the preoperative modeling agent is as follows: 2.1) Multimodal reconstruction: FSL and MRtrix tools were used to preprocess sMRI / dMRI data, and MNI standard neural atlas registration was performed to extract the ROI of the skull base (and generate a mask); 2.2) Path planning: Individualized candidate bone windows are generated based on skull bony landmarks. A risk assessment model is constructed through a twin neural network to dynamically adjust the size and path of the bone windows, ensuring consistency between the planned path and expert opinions and the algorithm response time. 2.3) Output results: Generate a 3D model of the skull base and surgical path file in VTK format and synchronize it to the navigation system.

3. The method for neurosurgical planning and risk warning based on multi-agent technology as described in claim 1 or 2, characterized in that, In (3), the processing procedure of the intraoperative sensing agent is as follows: 3.1) Scene segmentation: Based on the fine-tuning of the SAM2 large model, the surgical video stream is segmented in real time to identify surgical instruments and key brain tissues, and the segmentation DICE coefficient and object recognition accuracy are set. 3.2) 3D modeling: The depth information of the video stream is mined by the Transformer-Based stereo depth estimation module, and combined with surfel scene representation and camera pose estimation to construct a dynamic 3D scene during surgery; 3.3) Real-time registration: Align the intraoperative 3D model with the preoperative VTK model, set the registration accuracy, and ensure that the anatomical structure mapping error is within the preset limit.

4. The method for neurosurgical planning and risk warning based on multi-agent technology as described in claim 1 or 2, characterized in that, In (4), the processing procedure of the surgical guidance agent is as follows: 4.1) Decision support: Based on a large model of the neurosurgery vertical field and integrating an expert knowledge base, it provides recommendations for surgical approaches and instrument operation suggestions. The consistency between the decision-making plan and the experts is greater than a preset threshold. 4.2) Risk warning: Real-time monitoring of the distance between the device and key nerves, and neurophysiological signals, with a set warning response delay; 4.3) Interactive guidance: The system receives doctor's instructions and provides visual feedback through the voice interaction module, supporting multi-threaded parallel response.

5. A system for implementing the multi-agent-based neurosurgical planning and risk warning method as described in claim 1, characterized in that, The system includes: The multimodal data input module is used to acquire preoperative images, intraoperative video streams / electrophysiological signals, and expert knowledge data; The core module for intelligent agent scheduling includes a preoperative modeling intelligent agent, an intraoperative perception intelligent agent, and a surgical guidance intelligent agent. The preoperative modeling intelligent agent, based on multimodal image data, realizes three-dimensional reconstruction of the facial and auditory nerves and skull base structures, as well as surgical path planning. The intraoperative perception intelligent agent processes intraoperative video streams and real-time data, completing surgical scene segmentation, three-dimensional modeling, and registration with the preoperative model. The surgical guidance intelligent agent, combining an expert knowledge base and intraoperative data, provides proactive decision support, facial and auditory nerve protection risk warnings, and operational guidance. The multi-source information integration module is used to integrate the output of the intelligent agent and generate reports; The dynamic update and feedback module supports real-time writing of intraoperative data into the expert knowledge base and receives feedback from doctors on adjustments to the path / decision, which is used for iterative optimization of the agent model.

Images