Intelligent medical information management system

Abstract

The invention discloses an intelligent medical information management system, and relates to the technical field of medical information processing, and the system comprises a medical image sensing module which collects a dynamic image of an operation area and transmits the dynamic image according to a standard protocol; the low-delay semantic segmentation module is internally provided with an AI adaptive scene optimization sub-module, identifies a scene and adjusts parameters, and segments areas such as a focus by improving U-Net; the surgical navigation linkage module superposes results, adapts a visual angle and triggers an alarm; the data synchronous storage module encrypts stored data and supports tracing and medical record association; the equipment adaptation module is compatible with mainstream equipment, automatically matches a protocol and is deployed in a lightweight manner, and all the modules cooperate through a special Ethernet; the low-delay semantic segmentation module with the built-in AI self-adaptive scene optimization sub-module is matched with the improved U-Net architecture dynamic parameter adjustment segmentation, equipment compatibility, real-time navigation assistance and data safety tracing are achieved through cooperation of multiple modules, the problem that a traditional system is short is solved, and various surgical scenes and clinical requirements are met.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, specifically to a smart medical information management system. Background Technology

[0002] In the field of surgical diagnosis and treatment, precision and intelligence have become core development trends. Among them, surgical navigation technology based on medical image semantic segmentation, with its visualization of key areas such as lesions, blood vessels, and nerves, has become an important support for improving the accuracy of surgical operations and ensuring surgical safety. With the continuous development of smart medical technology, the application of deep learning algorithms in the field of image segmentation is becoming increasingly mature, various neural network architectures are being continuously optimized, and the digitalization and standardization of medical equipment are gradually improving, providing the basic technical conditions for real-time intraoperative image acquisition, segmentation, and navigation coordination. Currently, the core requirements for surgical navigation technology in clinical practice have shifted from basic image presentation to accurate segmentation that adapts to different surgical types and copes with complex intraoperative scenarios, as well as efficient coordination of the entire process of image acquisition, navigation assistance, and data management. This has also become a key direction for technological research and development in this field.

[0003] Existing surgical navigation systems fall short of meeting the aforementioned core clinical needs, exhibiting numerous key technological shortcomings. In image segmentation, current solutions generally employ fixed parameter settings, failing to dynamically adjust segmentation parameters based on different surgical types (e.g., neurosurgery, abdominal surgery) and complex scenarios such as intraoperative bleeding and tissue adhesions. This results in segmentation accuracy in key areas failing to match actual surgical requirements, hindering the provision of reliable image references. Regarding module collaboration, image acquisition, navigation assistance, and data storage are largely designed independently, lacking effective linkage mechanisms. This not only results in insufficient flexibility in perspective adaptation, making it difficult to align with surgeons' intraoperative operational perspectives, but also lacks a unified security encryption and rapid traceability scheme for data storage, preventing the formation of a complete closed loop of surgical assistance data. Finally, in terms of equipment compatibility, traditional systems have limited compatibility with mainstream brand intraoperative imaging equipment, cumbersome deployment processes, and stringent hardware requirements, making them difficult to adapt to the existing operating room environments and hardware infrastructure of hospitals at all levels. This limits their clinical application and prevents them from fully realizing the auxiliary value of smart healthcare technology. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a smart medical information management system. This system integrates five major modules: medical image sensing, low-latency semantic segmentation, surgical navigation linkage, data synchronization and storage, and device adaptation. It achieves efficient collaboration among these modules using a dedicated medical Ethernet bus. The core of the system lies in the AI ​​adaptive scene optimization submodule built into the low-latency semantic segmentation module. This submodule identifies the surgical type and intraoperative scenario through a dual mechanism of pre-operative doctor presets and image feature-assisted confirmation. Combined with feature fusion weights, confidence scores, and intersection-over-union (IoU) calculations, it dynamically adjusts segmentation parameters and, with the improved U-Net architecture, achieves accurate semantic segmentation of lesions, blood vessels, and nerve regions. The device adaptation module overcomes compatibility issues with multiple brand devices, supporting rapid deployment. The surgical navigation linkage module provides real-time perspective adaptation and risk alarms. The data synchronization and storage module enables encrypted data storage and traceability and can be linked to electronic medical records. The system requires no complex manual intervention, can adapt to diverse surgical scenarios, and standardizes the management of diagnostic and treatment data.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a smart medical information management system, which includes a medical image sensing module, a low-latency semantic segmentation module, a surgical navigation linkage module, a data synchronization and storage module, and a device adaptation module, wherein each module is connected to communicate via a medical dedicated Ethernet bus;

[0006] Medical image sensing module: used to acquire dynamic image data of the surgical area in real time. The image data includes pixel information, device parameters, timestamps and surgical site identifiers, and is transmitted using the medical image transmission standard protocol.

[0007] Low-latency semantic segmentation module: It has a built-in AI adaptive scene optimization submodule, which can identify the type of surgery and the intraoperative scene. Based on the recognition results, it dynamically adjusts the feature fusion weight, noise filtering intensity and segmentation threshold parameters. It achieves semantic segmentation of lesions, blood vessels and nerve regions by improving the U-Net architecture.

[0008] Surgical navigation linkage module: The semantic segmentation results are superimposed on the navigation interface in real time. By integrating the perspective tracking device, it dynamically adapts to the doctor's operating perspective and synchronously displays the spatial distance and scene status information between the lesion and blood vessels and nerves. When the distance is lower than the preset threshold, an abnormal sound and light alarm is triggered.

[0009] Data synchronization and storage module: encrypts and stores intraoperative images, segmentation results, scene recognition logs and parameter adjustment logs, uses a multi-dimensional indexing mechanism to achieve data traceability, and associates with the patient's electronic medical record through the medical record association standard protocol;

[0010] Equipment adapter module: Compatible with the communication protocols of mainstream brand intraoperative imaging equipment, supports automatic protocol matching, provides multiple physical interfaces and wireless connection methods, reserves expansion communication interfaces, and features lightweight installation and automated deployment.

[0011] Furthermore, the scene recognition unit of the AI ​​adaptive scene optimization submodule adopts the MobileNetV2 model, and identifies the surgical type and intraoperative scene through a dual mechanism of doctor's preoperative preset and image feature-assisted confirmation, with an accuracy rate of no less than 98% and a recognition delay of no more than 10 milliseconds.

[0012] Furthermore, the recognition confidence level of the scene recognition unit satisfies the formula: ,in, Here, n represents the confidence level, and n represents the number of image features. The score for matching the i-th feature. Let i be the weight coefficient of the i-th feature. The recognition result is confirmed to be valid when it reaches 95%.

[0013] Furthermore, the parameter dynamic adjustment unit of the AI ​​adaptive scene optimization submodule uses the following formula for dynamically adjusting the feature fusion weights: ,in, For the final feature fusion weights, This is the weighting coefficient for surgical type. As the basic weight for surgical type, Adjust the weights for the intraoperative scenario.

[0014] Furthermore, the weighting coefficient for the surgical type ranges from 0.4 to 0.6. During neurosurgery, the basic weight of the nerve region is higher than that of the blood vessel and lesion region. In bleeding scenarios, the noise filtering intensity is increased by 20% compared to the basic value.

[0015] Furthermore, the segmentation performance evaluation of the low-latency semantic segmentation module satisfies the formula: ,in, For the intersection-union ratio (IUGR) evaluation value, A represents the region segmented by the model, and B represents the region labeled by the doctor as the true region. When it falls below 85%, a secondary parameter fine-tuning is triggered.

[0016] Furthermore, the improved U-Net architecture includes a MobileNetV3 lightweight encoder, a multi-scale feature fusion layer, and a decoder. The multi-scale feature fusion layer has three parallel convolutional kernel branches, and the features are fused by weighted summation calculated by dynamically adjusting the feature fusion weights according to the formula.

[0017] Furthermore, the surgical navigation linkage module uses an RGB three-color annotation system to distinguish lesions, blood vessels, and nerve areas. The annotation line width is adjustable, and the transparency dynamically adapts to the image brightness. The view tracking device is an infrared view tracking device.

[0018] Furthermore, the data synchronization and storage module adopts dual protection of hardware encryption and software encryption. The hardware encryption adopts the AES-256 algorithm, and the multidimensional index includes patient identification, operation time, data type, and scene type. The medical record association standard protocol is the HL7 V3.0 protocol, and the medical image transmission standard protocol is the DICOM 3.0 protocol.

[0019] Furthermore, the low-latency semantic segmentation module also includes an incremental learning unit, which automatically triggers model fine-tuning after accumulating a preset number of clinical segmentation error data, and synchronously updates the scene feature library of the scene recognition model.

[0020] Compared with existing technologies, this intelligent medical information management system has the following advantages:

[0021] I. This invention achieves semantic segmentation of lesions, blood vessels, and nerve regions by incorporating an AI adaptive scene optimization submodule into the low-latency semantic segmentation module, combined with an improved U-Net architecture. This submodule can automatically identify the surgical type and intraoperative scene, dynamically adjusting core parameters such as feature fusion weights, noise filtering intensity, and segmentation thresholds based on the identification results. The three parallel convolutional kernel branches of the multi-scale feature fusion layer, through a weighted sum calculated using a dynamic feature fusion weight adjustment formula, enhance feature extraction performance in complex scenes. Compared to traditional segmentation schemes using fixed parameters, this solution can be specifically adapted to different types of surgery, such as neurosurgery and abdominal surgery, as well as special scenarios such as bleeding and tissue adhesion, achieving personalized segmentation processing. This ensures the accuracy and scene adaptability of the segmentation results, solving the problem that traditional segmentation techniques are difficult to adapt to diverse needs in complex surgical scenarios, and providing more reliable image data support for surgical operations.

[0022] II. This invention constructs a closed-loop surgical assistance system through the collaborative design of a medical image sensing module, a surgical navigation linkage module, and a data synchronization and storage module. The medical image sensing module collects data according to standard protocols, ensuring data standardization and integrity; the surgical navigation linkage module integrates infrared perspective tracking equipment, coupled with RGB three-color annotation and audible and visual alarm functions, to achieve real-time surgical assistance; the data synchronization and storage module adopts a dual encryption and multi-dimensional index design to complete secure data storage, rapid traceability, and association with electronic medical records; at the same time, the device adaptation module is compatible with mainstream brand intraoperative imaging equipment, provides multiple interface options, and has lightweight installation and automated deployment characteristics. This design solves the problems of limited device compatibility, lack of flexibility in navigation assistance, and fragmented data management in traditional systems, improves the safety and convenience of surgical operations, supports the standardized management of diagnostic and treatment data, and adapts to the operating room environment and actual clinical needs of different hospitals.

[0023] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from an examination of the following, or may be learned from the practice of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0025] Figure 1 A flowchart illustrating the relationships between system modules;

[0026] Figure 2 Workflow diagram for the low-latency semantic segmentation module;

[0027] Figure 3 This is a flowchart of the surgical navigation linkage module. Detailed Implementation

[0028] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0029] Example

[0030] This embodiment discloses a specific implementation of a smart medical information management system, specifically adapted for surgical navigation scenarios. The system first achieves compatibility with existing intraoperative imaging equipment in the hospital through a device adaptation module. Then, a medical image sensing module collects dynamic image data conforming to industry standards. A low-latency semantic segmentation module, combined with an AI adaptive scene optimization submodule and an improved U-Net architecture, achieves accurate segmentation of key regions. A surgical navigation linkage module provides real-time surgical assistance based on the segmentation results. A data synchronization and storage module ensures data security and traceability throughout the process. All modules work collaboratively, adapting to different surgical types and complex intraoperative scenarios while meeting actual clinical needs, providing comprehensive technical support for the precise implementation of surgical procedures. The collaborative workflow between modules is as follows: Figure 1 As shown.

[0031] The specific implementation method is as follows:

[0032] Implementation preparation

[0033] During the system deployment phase, the device adaptation module is the first to be activated, focusing on resolving compatibility issues between intraoperative imaging devices from different brands. This module incorporates communication protocol libraries for mainstream intraoperative imaging devices, covering various proprietary and industry-standard protocols. Once an intraoperative imaging device is connected, the module uses its built-in protocol recognition algorithm to quickly detect the device's communication protocol type, completing protocol matching and interface adaptation without manual configuration. This effectively avoids connection problems caused by differences in device brand and model. Simultaneously, the device adaptation module provides multiple physical interfaces and wireless connection methods, flexibly adapting to different operating room cabling environments. Reserved expansion communication interfaces can also meet the needs of future technology upgrades or the addition of new devices. The system is deployed using a lightweight installation package, with a fully automated installation process that requires no on-site technical personnel, enabling rapid system deployment and establishing a stable hardware connection foundation for the subsequent collaborative operation of various modules.

[0034] The surgeon presets the surgical type in the system according to the surgical plan and initializes the system's basic parameters, including the basic annotation style of the surgical navigation linkage module and the storage path of the data synchronization storage module. After initialization, the system automatically starts a self-check program to check the communication status of each module, the stability of device connections, and the readiness of the algorithm to ensure smooth data transmission between modules, normal device operation, and the ability to start the algorithm at any time, thus preventing problems such as equipment failure, data transmission interruption, or algorithm malfunction during surgery from the source.

[0035] Module running process

[0036] Medical image sensing module operation

[0037] After device adaptation is complete and system self-test passes, the medical image sensing module officially starts, beginning real-time acquisition of dynamic image data of the surgical area. The data acquired by this module includes pixel information, device parameters, timestamps, and surgical site identifiers. Pixel information is the core visual foundation for subsequent semantic segmentation; device parameters ensure the standardization and consistency of data acquisition; timestamps establish temporal relationships between data; and surgical site identifiers provide auxiliary basis for the scene recognition unit to accurately determine the surgical scene. Data acquisition strictly follows medical image transmission standard protocols, ensuring that the acquired data format conforms to industry standards, facilitating direct parsing and processing by the low-latency semantic segmentation module. During acquisition, the module maintains continuous and complete data transmission, preventing data loss or transmission delays, providing a stable and reliable input data source for the low-latency semantic segmentation module, ensuring that segmentation processing can keep pace with the surgical progress in real time without affecting the intraoperative operation rhythm.

[0038] Low-latency semantic segmentation module running

[0039] The low-latency semantic segmentation module is the core data processing unit of the system. After receiving the data transmitted by the medical image sensing module, it immediately starts the processing flow. Its built-in AI adaptive scene optimization sub-module is the key to achieving accurate segmentation in different scenarios.

[0040] The scene recognition unit utilizes the MobileNetV2 model, which combines lightweight design with high efficiency. It effectively reduces computational load while maintaining recognition accuracy, perfectly meeting the demands of real-time intraoperative processing. Scene recognition is achieved through a dual-mechanism approach: the surgeon's pre-defined surgical type provides a foundational direction for recognition, preventing overly broad coverage; intraoperative image feature-assisted confirmation refines the scene by extracting key features from the images, ensuring the recognition results accurately reflect the actual intraoperative situation. To prevent recognition errors from causing subsequent parameter adjustments mistakes, the reliability of the scene recognition results needs to be verified. Therefore, a scene recognition confidence score calculation is introduced, with the specific formula as follows: ,in, The scene recognition confidence score is given by n, which is the number of features extracted from the acquired images. These features are key information that can reflect the characteristics of the scene, such as the proportion of blood pixels and the blurring of tissue edges. The scene matching score for the i-th feature is used to quantify the degree of fit between a single feature and the target scene; Let be the weight coefficient of the i-th feature. This coefficient is set according to the importance of different features to scene recognition; features with a greater impact on scene recognition have higher weight coefficients. When the set standard is met, the recognition result is confirmed to be valid, and the subsequent parameter adjustment process is then triggered; if If the standard is not met, the module will re-extract the image features and re-identify until a valid result is obtained.

[0041] Once the scene recognition result is determined, the parameter dynamic adjustment unit starts working. Because the anatomical structures differ significantly between different surgical types, and the intraoperative scene may also change dynamically as the surgery progresses, fixed segmentation parameters cannot meet the accurate segmentation requirements of all scenes. Therefore, it is necessary to dynamically adjust the segmentation parameters based on the scene recognition result. Adjusting the feature fusion weights is the core step, and the calculation formula used is as follows: ,in, These are the final feature fusion weights, used to guide the feature fusion process of the multi-scale feature fusion layer; This is the weighting coefficient for surgery type, used to balance the impact of surgery type and intraoperative scenario on feature fusion; The basic weighting for surgical type is set according to the focus of attention on lesion, blood vessel and nerve areas based on different surgical types. For example, in neurosurgery, the basic weighting of nerve areas will be higher than that of blood vessel and lesion areas. The weights are adjusted for the intraoperative scene, determined based on the characteristics of the current scene. For example, in a bleeding scene, the weights of features that can penetrate blood occlusion are increased. In addition to feature fusion weights, the parameter dynamic adjustment unit also adjusts the noise filtering intensity and segmentation threshold based on the scene recognition results: for bleeding scenes, the noise filtering intensity is enhanced, and an adaptive threshold is used instead of a fixed threshold to deal with image interference caused by blood occlusion; for tissue adhesion scenes, the weights of deep semantic feature extraction are increased, and the boundary division of the adhesion area is optimized to ensure clear and accurate segmentation boundaries.

[0042] The adjusted segmentation parameters were applied to the segmentation process of the improved U-Net architecture. The improved U-Net architecture consists of three parts: a MobileNetV3 lightweight encoder, a multi-scale feature fusion layer, and a decoder. The MobileNetV3 encoder reduces model parameters through depthwise separable convolution, significantly improving processing speed without compromising feature extraction capabilities, perfectly meeting the core requirement of low latency during surgery. The multi-scale feature fusion layer uses three parallel convolutional kernel branches to extract shallow detail features, mid-level structural features, and deep semantic features from the image, respectively, and then calculates the feature fusion weights using the aforementioned formula. The features extracted from the three branches are weighted and summed to achieve effective fusion of multi-scale features. This ensures that the segmentation result retains detailed information of key regions while accurately identifying the semantic attributes of the regions. The decoder then maps the fused features to a segmentation mask corresponding to the original image size through an upsampling process, ultimately achieving accurate semantic segmentation of lesions, blood vessels, and nerve regions. The complete processing flow of this module from data reception to output segmentation results is as follows: Figure 2 As shown.

[0043] To quantify the segmentation effect and ensure that the segmentation accuracy meets the requirements of clinical surgery, an evaluation formula for crossover ratio (CR) is introduced: In this formula, IoU is the Intersection over Union (IoU) evaluation value, A is the segmented region output by the model, and B is the ground truth region labeled by the doctor. This formula intuitively reflects the degree of overlap between the segmented result and the ground truth region and is a core indicator for evaluating the accuracy of semantic segmentation. When the IoU is lower than the set standard, it indicates that the current segmentation parameters are not adequately adapted to the current scenario. The system will automatically trigger secondary parameter fine-tuning, adjusting the feature fusion weights, noise filtering intensity, or segmentation threshold slightly until the segmentation effect meets clinical requirements. In addition, the low-latency semantic segmentation module also includes an incremental learning unit that continuously collects segmentation error data from clinical surgeries. When the accumulated data reaches a preset amount, it automatically triggers model fine-tuning, synchronously updating the scene feature library of the scene recognition model, and continuously improving the system's adaptability to rare surgical types and special scenarios.

[0044] Surgical navigation linkage module operation

[0045] Upon receiving the segmentation mask output by the low-latency semantic segmentation module, the surgical navigation linkage module immediately initiates real-time surgical assistance. This module employs an RGB three-color annotation system to distinguish different key areas, using specific colors to identify lesions, blood vessels, and nerve regions. This allows surgeons to quickly and intuitively identify key surgical areas, reducing visual judgment time. The annotation line width can be flexibly adjusted according to the surgical field of view requirements, and the transparency dynamically changes with image brightness, ensuring clear visibility of the annotation information without obscuring important details in the original image, thus avoiding affecting the surgeon's overall assessment of the surgical area.

[0046] The integrated infrared perspective tracking device in the module captures the doctor's head position and line of sight in real time. Through a coordinate transformation algorithm, it dynamically adjusts the image display angle and annotation position on the navigation interface, allowing the doctor to clearly view the relative position of lesions, blood vessels, and nerves from any operating angle, avoiding operational errors caused by perspective deviations. Simultaneously, the module calculates the spatial distance between the lesion and blood vessels / nerves in real time and displays this information synchronously on the navigation interface. When the distance falls below a preset threshold, the system automatically triggers an abnormal audible and visual alarm, promptly alerting the doctor to avoid risks and ensuring surgical safety. The doctor can adjust the display state of the navigation interface through simple operations, such as zooming in and out of the image or moving the display area, to meet the auxiliary needs of different surgical stages. The real-time auxiliary decision-making and alarm triggering process of this module is as follows: Figure 3 As shown.

[0047] Data synchronization storage module running

[0048] Throughout the system's operation, the data synchronization and storage module continuously works, comprehensively storing all critical data, including intraoperative images, segmentation results, scene recognition logs, and parameter adjustment logs. Medical data involves patient privacy and treatment safety; therefore, the module employs a dual protection mechanism of hardware and software encryption. Hardware encryption uses specific encryption algorithms to perform low-level encryption on the data, while software encryption further strengthens data security. This dual protection effectively prevents data leakage or tampering.

[0049] The data is managed using a multi-dimensional indexing mechanism. The index includes patient identification, operation time, data type, and scenario type. Through this indexing system, medical staff or administrators can quickly locate and retrieve target data, meeting the data traceability needs in scenarios such as postoperative review, teaching and training, and medical quality control. Simultaneously, the data is linked to the patient's electronic medical record through a medical record association standard protocol, integrating various types of data generated during surgery into the patient's complete diagnosis and treatment data chain. This achieves standardized and integrated management of diagnosis and treatment data, providing reliable data support for subsequent diagnosis and treatment analysis, case studies, and improvement of medical quality.

[0050] In summary, this embodiment, through the orderly and collaborative operation of its various modules, fully realizes the application of a smart medical information management system in surgical navigation scenarios. The device adaptation module resolves communication issues between intraoperative imaging equipment from different brands, ensuring system compatibility with existing hospital hardware and reducing equipment upgrade costs. The medical image sensing module provides standardized, high-quality dynamic image data input, laying the data foundation for subsequent segmentation and navigation. The low-latency semantic segmentation module, leveraging the AI ​​adaptive scene optimization submodule and an improved U-Net architecture, achieves accurate low-latency segmentation for different surgical types and complex intraoperative scenarios through scientific formula calculations and dynamic parameter adjustments. The surgical navigation linkage module provides real-time and reliable assistance for surgical operations with intuitive annotations, flexible perspective adaptation, and timely alarm prompts, reducing surgical risks. The data synchronization and storage module, through dual encryption and multi-dimensional index design, ensures secure storage and rapid traceability of medical data, improving the diagnostic and treatment data management system. All modules work collaboratively through standardized communication protocols; the rational application of formulas ensures the scientific nature of parameter adjustments and segmentation evaluation; and dynamic parameter optimization allows the system to adapt to diverse surgical scenarios without complex manual intervention, effectively solving the technical problems of traditional surgical navigation systems.

[0051] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart medical information management system, characterized in that, The system includes a medical image sensing module, a low-latency semantic segmentation module, a surgical navigation linkage module, a data synchronization and storage module, and a device adaptation module. All modules are connected via a medical-specific Ethernet bus. Medical image sensing module: used to acquire dynamic image data of the surgical area in real time. The image data includes pixel information, device parameters, timestamps and surgical site identifiers, and is transmitted using the medical image transmission standard protocol. Low-latency semantic segmentation module: It has a built-in AI adaptive scene optimization submodule, which can identify the type of surgery and the intraoperative scene. Based on the recognition results, it dynamically adjusts the feature fusion weight, noise filtering intensity and segmentation threshold parameters. It achieves semantic segmentation of lesions, blood vessels and nerve regions by improving the U-Net architecture. Surgical navigation linkage module: The semantic segmentation results are superimposed on the navigation interface in real time. By integrating the perspective tracking device, it dynamically adapts to the doctor's operating perspective and synchronously displays the spatial distance and scene status information between the lesion and blood vessels and nerves. When the distance is lower than the preset threshold, an abnormal sound and light alarm is triggered. Data synchronization and storage module: encrypts and stores intraoperative images, segmentation results, scene recognition logs and parameter adjustment logs, uses a multi-dimensional indexing mechanism to achieve data traceability, and associates with the patient's electronic medical record through the medical record association standard protocol; Equipment adapter module: Compatible with the communication protocols of mainstream brand intraoperative imaging equipment, supports automatic protocol matching, provides multiple physical interfaces and wireless connection methods, reserves expansion communication interfaces, and features lightweight installation and automated deployment.

2. The intelligent medical information management system according to claim 1, characterized in that, The scene recognition unit of the AI ​​adaptive scene optimization submodule adopts the MobileNetV2 model. It identifies the surgical type and intraoperative scene through a dual mechanism of doctor's preoperative preset and image feature-assisted confirmation. The recognition accuracy is not less than 98% and the recognition delay is not more than 10 milliseconds.

3. The intelligent medical information management system according to claim 2, characterized in that, The recognition confidence level of the scene recognition unit satisfies the formula: ,in, Here, n represents the confidence level, and n represents the number of image features. The score for matching the i-th feature. Let i be the weight coefficient of the i-th feature. The recognition result is confirmed to be valid when it reaches 95%.

4. The intelligent medical information management system according to claim 1, characterized in that, The parameter dynamic adjustment unit of the AI ​​adaptive scene optimization submodule uses the following formula for dynamically adjusting the feature fusion weights: ,in, For the final feature fusion weights, This is the weighting coefficient for surgical type. As the basic weight for surgical type, Adjust the weights for the intraoperative scenario.

5. The intelligent medical information management system according to claim 4, characterized in that, The weighting coefficient for the surgical type ranges from 0.4 to 0.

6. During neurosurgery, the basic weight of the nerve region is higher than that of the blood vessel and lesion region. In bleeding scenarios, the noise filtering intensity is increased by 20% compared to the basic value.

6. The intelligent medical information management system according to claim 1, characterized in that, The segmentation performance evaluation of the low-latency semantic segmentation module satisfies the following formula: ,in, For the intersection-union ratio (IUGR) evaluation value, A represents the model segmentation result region, and B represents the doctor-annotated true region. When it falls below 85%, a secondary parameter fine-tuning is triggered.

7. The intelligent medical information management system according to claim 1, characterized in that, The improved U-Net architecture includes a MobileNetV3 lightweight encoder, a multi-scale feature fusion layer, and a decoder. The multi-scale feature fusion layer has three parallel convolutional kernel branches, and the features are fused by weighted summation calculated by dynamically adjusting the feature fusion weights according to the formula.

8. The intelligent medical information management system according to claim 1, characterized in that, The surgical navigation linkage module uses an RGB three-color annotation system to distinguish lesions, blood vessels, and nerve areas. The annotation line width is adjustable and the transparency dynamically adapts to the image brightness. The view tracking device is an infrared view tracking device.

9. The intelligent medical information management system according to claim 1, characterized in that, The data synchronization and storage module adopts dual protection of hardware encryption and software encryption. The hardware encryption adopts the AES-256 algorithm. The multidimensional index includes patient identification, operation time, data type, and scene type. The medical record association standard protocol is the HL7 V3.0 protocol, and the medical image transmission standard protocol is the DICOM 3.0 protocol.

10. A smart medical information management system according to claim 1, characterized in that, The low-latency semantic segmentation module also includes an incremental learning unit, which automatically triggers model fine-tuning after accumulating a preset number of clinical segmentation error data, and synchronously updates the scene feature library of the scene recognition model.

Images