An electronic medical record generation system based on burn wound images

Abstract

The application provides a kind of electronic medical record generation system based on burn wound image, it is related to image data processing technical field. By obtaining target wound image, and the image is anchored in the coordinate system in the level of human anatomy, obtain the body surface positioning result;Subsequently, the boundary form, damage grade and spatial measurement of the lesion are extracted and calculated from the target wound image. By integrating the parameters extracted in the above different technical dimensions into the human template, an electronic burn map containing anatomical location and lesion morphology is generated. Finally, a standardized electronic medical record with global spatial reference and objective quantitative characteristics is generated according to the electronic burn map. This scheme realizes the automatic conversion from the original image to the structured medical record, provides a unified data benchmark for subsequent diagnosis and treatment, and reduces the subjective bias introduced by manual assessment.

Description

Technical Field

[0001] This application relates to the field of image data processing technology, and in particular to an electronic medical record generation system based on burn wound images. Background Technology

[0002] In clinical practice, burn specialists need to determine the location, extent, depth, and area of ​​patient wounds, and conduct continuous follow-up. The percentage of total burn surface area directly determines the formulation of fluid resuscitation, intensive care, nutritional support, and overall treatment strategies, while burn depth is the basis for determining whether surgery is necessary and the timing of debridement and skin grafting. Therefore, standardized patient records are the prerequisite and foundation for achieving standardized treatment.

[0003] In actual clinical practice, images of the injured area are typically taken to obtain information about the wound, and medical staff then write written medical records based on their observations of these images. However, this method of archiving visual images and textual descriptions separately leads to heterogeneity and fragmentation of medical record data in terms of spatial reference and quantitative assessment, creating a cognitive burden for clinicians. Summary of the Invention

[0004] This application provides an electronic medical record generation system based on burn wound images, which is used to write these structured data into electronic medical records, realizing the transformation of photo information into standardized medical record content and facilitating the standardization of subsequent treatment.

[0005] In a first aspect, this application provides an electronic medical record generation system based on burn wound images, comprising: an image acquisition module for acquiring a target wound image and corresponding auxiliary information; the auxiliary information includes physical reference features, imaging features, and metadata; a body surface location recognition module for locating the target wound image in a preset human body template; a wound analysis module for performing boundary segmentation and depth level classification on the target wound image to generate wound analysis results; a quantization calculation module for performing calculations based on the wound analysis results, body surface location results, and auxiliary information to obtain quantization results; a template mapping module for mapping the wound analysis results and quantization results to a human body template according to the body surface location results to generate an electronic burn image; and a medical record generation module for generating and outputting an electronic medical record based on the electronic burn image.

[0006] By employing the aforementioned technical solution, an image acquisition module acquires an image of the target wound, and a body surface location recognition module anchors this image in a coordinate system at the anatomical level to obtain the body surface localization result. Subsequently, the wound analysis module and the quantitative calculation module work together to extract and calculate quantitative results such as the boundary morphology, damage level, and spatial measurement of the lesion from the target wound image. A template mapping module integrates the parameters extracted from the different technical dimensions into a human body template, generating an electronic burn map containing anatomical location and lesion morphology. Finally, the medical record generation module generates a standardized electronic medical record with global spatial reference and objective quantitative characteristics based on this electronic burn map. This solution achieves automatic conversion from raw images to structured medical records, providing a unified data benchmark for subsequent diagnosis and treatment, and reducing subjective bias introduced by manual assessment.

[0007] In conjunction with some embodiments of the first aspect, in some embodiments, the body surface location identification module specifically includes: a dense mapping submodule, used to scan the target wound image and generate a physical topology map representing the human body structure; compare the physical topology map with constant sequences of different body surface partitions to obtain the target constant sequence and body surface partition identifier of the target body surface partition to which the physical topology map belongs; retrieve the corresponding two-dimensional unfolded surface based on the body surface partition identifier; and perform coordinate operations by mapping the physical topology map and the target constant sequence onto the two-dimensional unfolded surface to obtain the template parameter coordinates; an anchor point generation submodule, used to extract and calculate the intermediate parameter coordinates of the target wound image in the observation space system based on physical reference features; and a positioning fusion submodule, used to establish the coordinate correspondence between the target wound image and the human body template based on the body surface partition identifier, template parameter coordinates, and intermediate parameter coordinates, and merge the output as the body surface positioning result.

[0008] By employing the aforementioned technical solutions, the dense mapping submodule performs global inference at a macroscopic level, and the output template parameter coordinates resolve the macroscopic attribution problem of the target wound image within the human body template. On the other hand, the anchor point generation submodule analyzes physical reference features and outputs intermediate parameter coordinates with physical references, providing a physical scale calibration benchmark for the target wound image within the current observation field of view. Finally, the localization and fusion submodule performs geometrical spatial alignment and fusion of these two sets of coordinate data obtained from macroscopic and microscopic dimensions respectively, constructing a body surface localization result that combines global anatomical attribution with local physical references. This solves the problem that close-up photos cannot be effectively addressed on the standard model due to the loss of global references.

[0009] In conjunction with some embodiments of the first aspect, in some embodiments, the dense mapping submodule is further configured to: perform continuity analysis on the physical topology map when it is determined that template parameter coordinates cannot be output or the output confidence scalar is lower than a preset safety threshold, identify feature-deficient distribution areas where feature faults exceed the threshold; construct a failure mask in the physical topology map to cover and isolate feature-deficient distribution areas, and update the physical topology map; determine continuous reference images adjacent to the target wound image on the spatial shooting trajectory based on physical reference features and metadata in the auxiliary information; extract the connected physical topology map corresponding to the continuous reference images, and Edge feature matching and fusion are performed on the physical topology map and the connected physical topology map to generate a joint physical topology map. The joint physical topology map is compared with the constant sequences of different continuum surface partitions to obtain the target continuous constant sequence and the associated continuum surface partition identifiers. The continuum surface partition is composed of at least two surface partitions with boundary connections. Based on the associated continuum surface partition identifiers, the corresponding two-dimensional unfolded surfaces are spliced ​​in the background space to obtain a joint two-dimensional unfolded surface. The joint physical topology map and the target continuous constant sequence are mapped to the joint two-dimensional unfolded surface to perform coordinate operations and solve for the template parameter coordinates.

[0010] By employing the aforementioned technical solution, when faced with scenarios where severe carbonization leads to significant loss of target wound image features, the dense mapping submodule identifies and isolates areas of feature loss through continuity analysis, preventing invalid data from interfering with subsequent calculations. Subsequently, while preserving valid information from the target wound image, the dense mapping submodule generates a joint physical topology map that expands the physical perception field by retrieving continuous reference images adjacent to it on the spatial imaging trajectory. Finally, the joint physical topology map is compared with cross-regional continuum surface partitions. With the structural constraints of the extensional partitions, the template parameter coordinates lost in the carbonized region are solved on the stitched joint two-dimensional unfolded surface, enabling localization even when the visual features of a single image are impaired.

[0011] In conjunction with some embodiments of the first aspect, in some embodiments, the body surface location recognition module further includes: an abnormal trigger replacement submodule, used to obtain the confidence vector corresponding to the template parameter coordinates output by the dense mapping submodule; when the confidence vector is lower than a preset confidence threshold, receiving external input coordinates and updating the template parameter coordinates using the external input coordinates.

[0012] By adopting the above technical solution, the abnormal trigger replacement submodule of the body surface location recognition module can obtain the confidence information of the automatic localization results in real time. When the confidence of the automatically generated template parameter coordinates falls below the preset confidence threshold due to the complexity of the wound, the low-quality data will be paused, and external input coordinates will be received to update it. This design, while retaining the efficiency of automated processing, introduces human experience as the final decision-making channel, thereby ensuring the high clinical usability of the output localization coordinates under complex wound conditions.

[0013] In conjunction with some embodiments of the first aspect, in some embodiments, the wound analysis module specifically includes: a global surface localization submodule, used to identify visible burn areas, non-burn areas, and unidentifiable areas based on the target wound image, and output a whole-body burn area mask; a fine analysis submodule, used to perform boundary segmentation and depth level classification on the whole-body burn area mask, and extract and output wound segmentation results and corresponding burn depth grading results; and a feature merging submodule, used to merge and encapsulate the whole-body burn area mask, wound segmentation results, and burn depth grading results as the wound analysis result output.

[0014] By adopting the above technical solution, the global localization submodule first outputs a mask of the whole-body burn area from a macroscopic perspective. This mask is used to delineate the approximate distribution range of lesions and initially eliminate non-evaluation areas. Based on this, the fine analysis submodule focuses its computational power on the interior of this mask, performing pixel-level boundary segmentation and depth-level classification. This coarse-to-fine layered processing architecture helps avoid misjudgments by the segmentation model in the global image due to background interference, and ensures that the final wound analysis results simultaneously possess macroscopic distribution information from the global view and microscopic boundary information for quantification calculations.

[0015] In conjunction with some embodiments of the first aspect, in some embodiments, the fine analysis submodule is also used to: generate a wound surface model based on the different wound images when multiple wound images of the same visible burn area are detected by metadata; and update the wound segmentation results based on the wound surface model.

[0016] By employing the aforementioned technical solution, when multiple images from different perspectives showing the same visible burn area are detected through metadata, the fine analysis submodule no longer treats these images as isolated data. Instead, it utilizes the parallax information between them to perform 3D surface reconstruction, generating a wound surface model that reflects the physical contours of the area. Subsequently, the original 2D wound segmentation results are updated and optimized based on this wound surface model. This approach resolves the issues of boundary calculation distortion and area estimation bias caused by perspective collapse in 2D photographs, enabling the segmentation results to better match the actual anatomical structure of the human body.

[0017] In conjunction with some embodiments of the first aspect, in some embodiments, the quantization calculation module specifically includes: a proportion aggregation calculation submodule, used to parse the preset area weight of the corresponding body surface region from the human body template; calculate the area proportion of the target wound image in the corresponding body surface region based on the wound analysis results; perform a weighted multiplication aggregation operation on the area weight and the area proportion to obtain the total burn body surface area proportion, and put it into the quantization result.

[0018] By adopting the above technical solution, the quantification calculation module, when estimating the total burn surface area ratio, no longer simply calculates the original proportion of lesion pixels to the total number of pixels in the image. Instead, through the proportion aggregation calculation submodule, it first parses the physiologically preset area weight coefficients for different body parts from the human body template. Subsequently, it performs a weighted multiplication aggregation operation on the wound area ratio of each part calculated by machine vision and these area weights. This calculation method, which combines pixel statistics with prior medical knowledge, allows the final total burn surface area ratio parameter to overcome the limitations of pure machine vision and has clinical reference value.

[0019] In conjunction with some embodiments of the first aspect, in some embodiments, the quantization calculation module further includes: a physical area calculation submodule, used to extract the scale conversion coefficient mapped by the scale marker information from the physical reference features; extract the equivalent pixel area corresponding to the wound segmentation region in the wound analysis result; perform an algebraic multiplication of the scale conversion coefficient and the equivalent pixel area to calculate the absolute area of ​​the wound, and put it into the quantization result.

[0020] By employing the aforementioned technical solution, the physical area calculation submodule extracts scale conversion coefficients mapped from the physical reference features of auxiliary information, establishing a mathematical mapping relationship between the image pixel space and the real physical space. Subsequently, this module algebraically multiplies the equivalent pixel area of ​​the wound segmentation region with these scale conversion coefficients to calculate the absolute physical area of ​​the wound. This absolute physical area index provides an objective physical measurement basis for clinical surgical planning and precise medication administration.

[0021] In conjunction with some embodiments of the first aspect, in some embodiments, a follow-up management module is also included. The follow-up management module is used to: acquire a first wound record generated at a first historical time node and a second wound record generated at a second current time node for the same patient; align and map the first wound record and the second wound record to the human body template coordinate system; and perform a time-varying comparison on the first wound record and the second wound record within the coordinate system, calculate and generate a comparison result output.

[0022] By adopting the above technical solution, the follow-up management module aligns and maps wound records from different time points to a unified human body template coordinate system, solving the problem of the inability to directly and quantitatively compare data before and after the initial observation due to varying imaging parameters in traditional follow-up. Within this unified coordinate system, time-varying dynamic comparisons can be performed between the first and second wound records, quantitatively calculating the expansion or contraction of the wound area and the worsening or improvement of its depth. This allows clinicians to gain insight into the dynamic evolution trend of the wound based on objective data, providing a basis for adjusting treatment plans.

[0023] In conjunction with some embodiments of the first aspect, in some embodiments, an image preprocessing module is further included, which is used to perform quality detection on the target wound image and select a first intermediate image that meets a preset standard; perform boundary cropping on the first intermediate image to obtain a second intermediate image; perform noise reduction, background interference suppression, brightness and color normalization on the second intermediate image to obtain a third intermediate image; perform data desensitization processing on sensitive identity information in the third intermediate image to obtain a fourth intermediate image; use physical reference features as a marker reference to perform spatial distortion correction on the fourth intermediate image to obtain a fifth intermediate image, and update the target wound image based on the fifth intermediate image.

[0024] By employing the aforementioned technical solutions, the image preprocessing module, through a series of operations such as quality inspection, cropping, denoising, and background suppression, helps filter out unavoidable physical noise and irrelevant information interference in complex clinical environments. Brightness and color standardization helps unify imaging benchmarks under different devices and lighting conditions. Furthermore, desensitization and distortion correction help ensure data compliance and spatial accuracy. This module's design improves the operational stability of downstream algorithms and contributes to enhancing the reliability of the final output of the entire electronic medical record generation system.

[0025] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0026] 1. The target wound image is acquired through an image acquisition module, and the image is anchored in a coordinate system at the anatomical level using a body surface location recognition module to obtain the body surface localization result. Subsequently, the wound analysis module and the quantitative calculation module work together to extract and calculate the boundary morphology, damage level, and spatial measurement of the lesion from the target wound image. The template mapping module integrates the parameters extracted from the above different technical dimensions into a human body template to generate an electronic burn map containing anatomical location and lesion morphology. Finally, the medical record generation module generates a standardized electronic medical record with global spatial reference and objective quantitative characteristics based on this electronic burn map. This solution realizes the automatic conversion from raw image to structured medical record, provides a unified data benchmark for subsequent diagnosis and treatment, and reduces the subjective bias introduced by manual assessment.

[0027] 2. The dense mapping submodule performs a global inference at the macroscopic level, and the output template parameter coordinates solve the macroscopic attribution problem of the target wound image in the human body template. On the other hand, the anchor point generation submodule analyzes the physical reference features and outputs intermediate parameter coordinates with physical reference, providing a physical scale calibration benchmark for the target wound image in the current observation field of view. Finally, the localization fusion submodule performs geometric spatial alignment and fusion of these two sets of coordinate data obtained from macroscopic and microscopic dimensions respectively, constructing a body surface localization result that combines global anatomical attribution and local physical reference, thereby solving the problem that local close-up photos cannot be effectively addressed on the standard model due to the loss of global reference.

[0028] 3. When faced with scenarios where severe carbonization leads to significant loss of target wound image features, the dense mapping submodule identifies and isolates areas of feature loss through continuity analysis, preventing invalid data from interfering with subsequent calculations. Subsequently, while preserving valid information from the target wound image, the dense mapping submodule generates a joint physical topology map that expands the physical perception field by retrieving continuous reference images adjacent to it on the spatial shooting trajectory. Finally, the joint physical topology map is compared with the cross-regional continuum surface partitions. With the structural constraints of the extensional partitions, the template parameter coordinates lost in the carbonized region are solved on the stitched joint two-dimensional unfolded surface, thus enabling localization even when the visual features of a single image are impaired. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of a module of an electronic medical record generation system based on burn wound images in an embodiment of this application.

[0030] Figure 2 This is a schematic diagram of a specific module of the body surface part recognition module in the embodiments of this application.

[0031] Figure 3 This is a schematic diagram of a specific module of the wound analysis module in the embodiments of this application.

[0032] Figure 4 This is a schematic diagram of the target wound image in an embodiment of this application.

[0033] Figure 5 This is a schematic diagram of an electronic burn map in an embodiment of this application.

[0034] Figure 6 yes Figure 5 A magnified view of a portion of the image.

[0035] Figure 7 This is a schematic diagram of the wound analysis results in an embodiment of this application.

[0036] Figure labels: 101, Image acquisition module; 102, Image preprocessing module; 103, Body surface location recognition module; 104, Wound analysis module; 105, Quantitative calculation module; 106, Template mapping module; 107, Doctor interaction module; 108, Medical record generation module; 109, Follow-up management module; 130, Dense mapping submodule; 131, Anchor point generation submodule; 132, Localization fusion submodule; 133, Abnormal trigger replacement submodule; 140, Global body surface localization submodule; 141, Fine analysis submodule; 142, Feature package merging submodule. Detailed Implementation

[0037] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0038] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0039] Please see Figure 1 , Figure 1 This is a schematic diagram of a module of an electronic medical record generation system based on burn wound images, as described in an embodiment of this application. (Refer to...) Figure 4 , Figure 4 This is a schematic diagram of the target wound image in an embodiment of this application.

[0040] Image acquisition module 101 is used to acquire images of the target wound and corresponding auxiliary information; wherein, the auxiliary information includes physical reference features, imaging features, and metadata.

[0041] Among them, the target wound image refers to the original digital image of the patient's burn area captured by the image acquisition device; the physical reference feature refers to the reference information used to correct the physical properties of the image; the imaging feature refers to the supplementary image data acquired by the sensor with a different imaging principle than visible light; and the metadata refers to text or numerical information that is unrelated to the image content and is used to describe the image management attributes.

[0042] In some embodiments, physical reference features may include preliminary site description, left and right side information, scale marker information, color calibration information, etc.; imaging features may include depth images, thermal images, etc.; metadata may include patient basic information, acquisition time, wound number, etc.

[0043] The body surface location recognition module 103 is configured to locate the target wound image in a preset human body template and find the corresponding body surface location result.

[0044] Among them, the human body template refers to a two-dimensional or three-dimensional human body model with standardized anatomical partitions and coordinate systems that is pre-installed in the system backend; the body surface positioning result refers to a set of parametric data used to describe the spatial position of the target wound image on the human body template.

[0045] In some embodiments, the body surface location recognition module 103 is specifically configured to receive a target wound image input from the front end and call a background computer vision algorithm to analyze the image. The task of this module is to solve the problem of the correspondence between images and body parts. It uses deep learning networks or image processing methods to find the anatomical location of a local image lacking spatial reference within the global coordinate system of the human body template. Its output body surface localization result serves as a "spatial addressing pointer," which is called by the downstream quantization calculation module 105 and template mapping module 106 to ensure that subsequent processing is performed on a unified anatomical reference.

[0046] In some specific embodiments, a dense mapping network based on deep learning can be used to perform pixel-level analysis on the target wound image and output the coordinates of each pixel on the unfolded surface of the human body template; subsequently, clustering and geometric center calculation are performed on these coordinate point sets to generate the body surface localization results, which is not limited here.

[0047] In some other specific embodiments, a skeleton key point recognition algorithm can be used first to extract macroscopic features such as limb contours and joint nodes in the image; then, these macroscopic features are rigidly registered with the skeleton of the human body template; finally, based on the registered macroscopic position, the surface positioning result of the wound area on the surface of the human body template is derived, which is not limited here.

[0048] In actual use, when processing close-up images that do not contain clear anatomical landmarks (such as joints, navels, or outlines), the body surface recognition module 103 may fail to locate the image due to relying solely on the internal texture of the image.

[0049] In some preferred embodiments, the body surface location recognition module 103 can receive and parse physical reference features in the auxiliary information. For example, textual information describing the preliminary location can be used as an initial region definition, or a stable local coordinate system can be constructed using reference markers captured along with the image, and then fused with the global coordinate system of the human body template to solve the aforementioned problem of positioning failure.

[0050] Please see Figure 2 , Figure 2 This is a schematic diagram of a specific module of the body surface part recognition module in the embodiments of this application.

[0051] In some preferred embodiments, the body surface location recognition module 103 includes:

[0052] The dense mapping submodule 130 is used to scan the target wound image and generate a physical topology map representing the human body structure; compare the physical topology map with the constant sequences of different body surface partitions to obtain the target constant sequence and body surface partition identifier of the target body surface partition to which the physical topology map belongs; and based on the body surface partition identifier, retrieve the corresponding two-dimensional unfolded surface, map the physical topology map and the target constant sequence onto the two-dimensional unfolded surface to perform coordinate operations, and solve for the template parameter coordinates.

[0053] Among them, the physical topology map refers to the high-dimensional feature map extracted from the image through a deep network that can reflect the geometric structure and texture continuity of the human body surface; the constant sequence refers to the discrete numerical values ​​or code set used to identify each standard anatomical region; and the two-dimensional unfolded surface refers to the mathematical unfolding of the three-dimensional surface of the human body template into a two-dimensional planar coordinate system.

[0054] In some embodiments, the dense mapping submodule 130 receives a target wound image and extracts visual features representing human anatomical structures using a deep convolutional network to form a physical topological map. Subsequently, this module matches or classifies this map against pre-existing model libraries containing typical feature templates (i.e., constant sequences) representing different body parts, thereby macroscopically determining the anatomical region to which the image belongs and outputting a body surface region identifier. Finally, it retrieves the corresponding two-dimensional unfolded surface based on this identifier, projects the map features onto it, and uses regression calculations to determine the relative position of the wound on the region plane, i.e., the template parameter coordinates.

[0055] In some specific embodiments, the dense mapping submodule 130 may employ an end-to-end fully convolutional network trained to output a multi-channel map of the same size as the input image, wherein one channel is a pixel-level part classification map (corresponding to body surface partition identifiers), and the other two channels are coordinate regression maps of that part (corresponding to template parameter coordinates), which are not limited here.

[0056] In some other specific embodiments, the dense mapping submodule 130 may first extract general features through a backbone network, and then send the features into two parallel branch networks. One branch is responsible for performing the classification task and outputting the body surface partition identifier, while the other branch dynamically loads the corresponding regression model according to the classification result to calculate the template parameter coordinates. This is not limited here.

[0057] Anchor point generation submodule 131 is used to extract and calculate the intermediate parameter coordinates of the target wound image in the observation space system based on the physical reference features.

[0058] Among them, intermediate parameter coordinates refer to the position parameters in a local coordinate system established with the physical reference object within the target wound image as the origin.

[0059] In some embodiments, the anchor point generation submodule 131 is primarily used to provide physical scale and positional references for the target wound image at the microscopic level. This submodule is configured to parse the physical reference features carried in the auxiliary information and identify the image entity corresponding to the feature from the pixel matrix of the target wound image, i.e., one or more physical markers (such as rulers or stickers) of known size and fixed shape. Upon successful identification, the submodule uses the geometric center or specific corner of this physical marker as the coordinate origin (i.e., spatial anchor point) of the local view, establishing a local observation space system defined by the physical reference object, independent of image content. Subsequently, the module calculates the vector displacement of pixels within the wound area relative to this origin, thereby generating physically meaningful intermediate parameter coordinates. The intermediate parameter coordinates output by this module provide a physical anchoring basis for subsequent macroscopic and microscopic coordinate fusion.

[0060] In some specific embodiments, Hough transform or edge detection can be performed on the target wound image to identify straight lines or circles as physical markers; then, the pixel / length ratio of the image is calculated based on the pixel size of these geometric primitives and the pre-stored real physical size; finally, intermediate parameter coordinates are generated based on this ratio and the position of the markers, which is not limited here.

[0061] In other specific embodiments, a deep learning-based object detection network can be used. This network is trained to recognize reference stickers of specific patterns. After detecting the sticker, the network outputs its predefined corner coordinates and confidence scores. This submodule uses these high-confidence corner coordinates as a reference to construct a local reference frame and calculate intermediate parameter coordinates, which is not limited here.

[0062] In actual use, physical markers may have their visual features damaged during the shooting process due to liquid immersion, reflection, or partial occlusion, making it impossible for the recognition algorithm to extract their complete geometric shape, thus causing deviations in the calculated intermediate parameter coordinates.

[0063] In some preferred embodiments, the submodule can be configured as a robust identification mechanism based on redundant features and geometric constraints. Under this mechanism, the system does not rely on a single marker or feature, but instead requires a composite marker containing multiple feature points (such as a checkerboard pattern with multiple corner points) to be placed around the wound. When some feature points are occluded, the submodule can still utilize the remaining visible feature points and, through the inherent, known geometric constraints between these points (e.g., they form a square in the real world), reverse-engineer the location of the occluded points. This approach, based on redundant information and geometric constraints, enhances the success rate of physical marker identification and the stability of intermediate parameter coordinates in complex clinical environments.

[0064] The positioning and fusion submodule 132 is used to establish the coordinate correspondence between the target wound image and the human template based on the body surface partition identifier, template parameter coordinates and intermediate parameter coordinates, and merge the output as the body surface positioning result.

[0065] In some embodiments, the localization fusion submodule 132 is configured as the coordinate system registration and unification center for the body surface part recognition process. Its core task is to solve the problem of fusing two sets of heterogeneous coordinate data generated by different algorithms and different reference systems. The data flow of this submodule is as follows: it simultaneously receives macroscopic localization information (i.e., body surface partition identifiers and template parameter coordinates) in the human template coordinate system from the dense mapping submodule 130, and microscopic localization information (i.e., intermediate parameter coordinates) in the local observation space from the anchor point generation submodule 131. This submodule transforms and aligns the microscopic intermediate parameter coordinates to the macroscopic human template coordinate system by performing geometric space transformation operations. Through this fusion process, the physical accuracy of the microscopic coordinates is used to correct any recognition biases that may exist in the macroscopic coordinates, and finally, a localization result in a unified coordinate system that combines anatomical attribution and physical anchoring is calculated and output, i.e., the body surface localization result.

[0066] In some specific embodiments, key feature points of the local image can be calculated based on intermediate parameter coordinates, and corresponding feature points can be found within the region determined by template parameter coordinates. Then, the mapping relationship between the two coordinate systems is established by solving the affine transformation matrix between the two sets of points. Finally, the coordinates of the local image are transformed as a whole using this matrix to complete the fusion. No limitation is made here.

[0067] In practical applications, the aforementioned surface localization methods rely on the extraction and matching of native physiological features of the human body surface. However, in specific clinical burn scenarios, when dealing with moderate to severe burns, the wound area often exhibits pathological changes such as tissue carbonization, eschar formation, or severe edema due to deep thermal damage. These pathological changes manifest at the image level as loss of high-frequency texture information, non-rigid deformation of the original geometric contours, and a break in the color space from normal tissue. This variation in input data leads to a "feature vacuum" in conventional dense mapping networks during feature extraction, resulting in unreliable outputs for subsequent classification and regression calculations due to a lack of effective input support.

[0068] Therefore, in some preferred embodiments, the dense mapping submodule 130 is further configured to: perform continuity analysis on the physical topology map when it is determined that the template parameter coordinates cannot be output or the output confidence scalar is lower than a preset safety threshold, identify the feature missing distribution area where the feature fault exceeds the threshold; construct a failure mask in the physical topology map to cover and isolate the feature missing distribution area, and update the physical topology map; determine the continuous reference image adjacent to the target wound image on the spatial shooting trajectory based on the physical reference features and metadata in the auxiliary information; extract the connected physical topology map corresponding to the continuous reference image, and update the physical topology map. Edge feature matching and fusion are performed on the physical topology map and the connected physical topology map to generate a joint physical topology map. The joint physical topology map is compared with the constant sequences of different continuum surface partitions to obtain the target continuous constant sequence and the associated continuum surface partition identifiers. The continuum surface partition is composed of at least two surface partitions with boundary connection relationships. Based on the associated continuum surface partition identifiers, the corresponding two-dimensional unfolded surfaces are spliced ​​in the background space to obtain a joint two-dimensional unfolded surface. The joint physical topology map and the target continuous constant sequence are mapped to the joint two-dimensional unfolded surface to perform coordinate operations and solve for the template parameter coordinates.

[0069] Among them, the feature missing distribution area refers to the region in the physical topology map that cannot be effectively mapped due to the damage to the original image features; the failure mask refers to the digital marker used to isolate the feature missing distribution area at the data level to prevent it from interfering with subsequent calculations; the continuous reference image refers to the adjacent image that is continuous with the target wound image in terms of shooting time sequence or spatial location; the joint physical topology map refers to the composite feature map with a wider field of view formed by splicing and fusing at least two independent physical topology maps.

[0070] In some embodiments, the dense mapping submodule 130 is configured to include a control flow for anomaly detection and compensation. The execution logic of this control flow is as follows: First, during the calculation of conventional coordinate solving, the submodule monitors the output status of the template parameter coordinates and its accompanying confidence scalar in real time. When a mapping anomaly is detected, such as a coordinate solution gap or a confidence scalar lower than a preset confidence threshold, the system does not terminate the current task but suspends the main calculation process and triggers a fault-tolerant compensation sequence.

[0071] This fault-tolerant compensation sequence performs surface feature continuity analysis on the physical topology map corresponding to the local region where mapping anomalies occur, in order to locate and identify the distribution area of ​​feature loss caused by problems with the image data source (e.g., tissue carbonization). Subsequently, a failure mask is constructed at the corresponding location in the physical topology map to isolate the region at the data level.

[0072] This submodule retrieves and calls up consecutive reference images adjacent to the current target wound image from the image dataset based on temporal or spatial correlation information in the metadata. By fusing the physical topology map of the target wound image (which already includes the failure mask) with the physical topology maps of the consecutive reference images using edge features, a joint physical topology map with an extended receptive field is generated.

[0073] This submodule uses this joint physical topology map as new input and feeds it back into the subsequent alignment and coordinate solving process. In this process, the undamaged anatomical features provided by the continuous reference images are used as an effective benchmark for coordinate derivation. Through spatial interpolation or geometric constraints, the coordinates of the missing template parameters in the original feature-deficient distribution area are solved in reverse, thereby completing the data-level repair and compensation for the initial mapping anomalies.

[0074] In some specific embodiments, the video image of the previous or next frame of the target wound image can be retrieved through temporal metadata as a continuous reference image; the pixel displacement field between the two frames is calculated using an optical flow estimation algorithm, and the physical topology map of the two frames is pixel-level aligned and stitched based on the displacement field; finally, the stitched joint physical topology map is put into the coordinate solving network, which is not limited here.

[0075] In other specific embodiments, when a series of photographs surrounding the wound are recorded, multi-view geometric registration can be performed using physical reference features (such as multiple marker points) to determine the spatial geometric relationship between adjacent photographs; based on this geometric relationship, the physical topology maps of the two photographs are fused in three-dimensional space to generate a joint physical topology map; then the joint physical topology map is projected back onto a two-dimensional unfolded surface for coordinate solving, which is not limited here.

[0076] In practical use, there may be a problem in ensuring that the stitched joint physical topology map correctly matches the continuous body surface partitions in the background. For example, when the target wound image is located in "lower left thigh" and the continuous reference image is located in "left knee joint", the stitched joint physical topology map spans two independent anatomical partitions. If the system still uses a constant sequence of a single partition for comparison, it may lead to matching failure.

[0077] In some preferred embodiments, the background constant sequence library of the dense mapping submodule 130 not only contains independent body surface partitions, but also pre-sets a series of continuous body surface partitions with physiological connections (such as the "lower left thigh - left knee" composite area). When the system generates a joint physical topology map, it will first call the constant sequences of these continuous body surface partitions for comparison. After a successful match, the system can not only obtain the cross-regional associated continuous body surface partition identifiers, but also synchronously retrieve the spliced ​​joint two-dimensional unfolded surface, thereby ensuring that coordinate solving can still be performed on the correct mathematical basis even with field of view expansion.

[0078] However, in more complex scenarios, especially in boundary areas such as the torso, a single body surface region may have physiological connections with multiple other regions (for example, the "abdomen" region is connected to both the "chest" and the "groin"). In this case, after the dense mapping submodule 130 generates a joint physical topology map, if matching is based solely on features, multiple reasonable continuous body surface regions may be matched from the constant sequence library, resulting in ambiguity in the matching results.

[0079] In some preferred embodiments, the system is configured to incorporate prior knowledge of the shooting behavior to constrain the matching range. Specifically, when determining continuous reference images, the dense mapping submodule 130 not only acquires the images themselves but also parses temporal information or spatial trajectory vectors related to the shooting behavior from metadata. When comparing the joint physical topology map with the continuous body surface partition constant sequence, the system uses this prior knowledge of the shooting behavior as a constraint to filter candidate continuous body surface partitions. For example, if the shooting trajectory recorded by the system moves from the torso to the lower limbs, then when matching the "abdomen" region, the system will automatically filter out the candidate partition "abdomen-chest" which connects upwards, and only retain the candidate partition "abdomen-groin" which conforms to the movement trend. In this way, the system transforms the operational intent into search constraints for the algorithm, compressing the search space for matching, thereby helping to achieve unambiguous cross-region localization in complex scenarios with multiple possible physiological connections.

[0080] In some preferred embodiments, the body surface location recognition module 103 further includes:

[0081] The exception trigger replacement submodule 133 is used to obtain the confidence vector corresponding to the template parameter coordinates output by the dense mapping submodule 130; when the confidence vector is lower than the preset confidence threshold, it receives external input coordinates and uses the external input coordinates to update the template parameter coordinates.

[0082] The confidence vector refers to the numerical sequence output synchronously by the deep network of the dense mapping submodule 130 during inference, used to quantify the reliability of its own prediction results; the external input coordinates refer to the coordinate data manually calibrated or input by users with operating permissions (such as clinicians) through the graphical interactive interface.

[0083] In some embodiments, the anomaly triggering replacement submodule 133 acts as a parallel monitoring unit of the body surface location recognition module 103. It does not directly participate in coordinate calculation but instead "listens" to the output of the dense mapping submodule 130 in real time. Its workflow is as follows: First, it captures the coordinates of each automatically generated template parameter and its accompanying confidence vector. Then, it compares the confidence vector with a preset threshold within the system. Once the confidence is found to be lower than the threshold, indicating that the machine has low confidence in the current automatic positioning result, the submodule triggers an interrupt or suspension signal and activates the user interface, waiting for and receiving external input coordinates representing human experience. Finally, the submodule overwrites or replaces the original low-quality machine coordinates with the external input coordinates and then releases the coordinates to downstream modules.

[0084] Please see Figure 3 and Figure 7 , Figure 3 This is a schematic diagram of a specific module of the wound analysis module in the embodiments of this application; Figure 7 This is a schematic diagram of the wound analysis results in an embodiment of this application.

[0085] The wound analysis module 104 is configured to perform boundary segmentation and depth level classification on the target wound image to generate wound analysis results.

[0086] Among them, the wound analysis results refer to structured data packages that include the geometric contour information of the wound area and the pathological damage level of each area.

[0087] In some embodiments, the wound analysis module 104 is specifically configured to perform pixel-level semantic understanding on the input target wound image. It contains at least two parallel processing branches: one is a boundary segmentation branch, which uses an image segmentation network to perform multivariate classification of pixels in the image into foreground (wound) and background (normal skin), outlining the edge contour of the irregular wound and forming a digital mask; the other is a depth-level classification branch, which analyzes visual features such as color, gloss, and texture of the wound area and determines the pathological attributes of the pixel areas within the mask according to the clinical grading standards for burns. Finally, this module encapsulates the geometric morphology data and pathological attribute data to generate wound analysis results, providing content for downstream quantification and visualization mapping.

[0088] In some specific embodiments, a cascaded network structure can be adopted. First, a coarse-grained segmentation network is used to locate the approximate wound area. Then, the area is input into a high-resolution fine-grained segmentation network for edge refinement. At the same time, a parallel classification network performs depth level discrimination on the refined area. This is not limited here.

[0089] In some other specific embodiments, a single deep network with multi-task learning can be used, with two different output heads set at the end of the network. One output head is responsible for generating a pixel-level segmentation mask, while the other output head directly outputs the depth-level classification probability distribution of the corresponding region, so that segmentation and classification tasks can be completed simultaneously in one forward propagation. This is not limited here.

[0090] In actual use, the wound analysis module 104 faces the challenge of the complexity of the visual features of the burn wound itself. For example, burn areas of different depths may be mixed on the same wound surface, and blood scabs, blisters, secretions, etc. can interfere with the extraction of visual features, leading to boundary segmentation and depth misjudgment.

[0091] In some preferred embodiments, imaging features can be used for multimodal fusion analysis. For example, temperature field data (reflecting blood supply) from thermal imaging images or three-dimensional structural data (reflecting tissue swelling or indentation) from depth images can be fused. Cross-validation of multidimensional physical information can penetrate visual surface interference, define functional tissue boundaries, and assess damage depth, thereby improving the objectivity of wound analysis results.

[0092] In some preferred embodiments, the wound analysis module 104 specifically includes:

[0093] The global surface localization submodule 140 is used to identify visible burn areas, non-burn areas and unidentifiable areas based on the target wound image, and output a whole-body burn area mask.

[0094] Among them, the whole-body burn area mask refers to a digital layer with the same size as the target wound image, used to mark the burn area, normal area and interference area at the pixel level.

[0095] The main function of the global surface localization submodule 140 is to perform macroscopic region division on the entire input image, initially delineate the burn area to be analyzed, and remove invalid areas such as background and occlusion in advance. This submodule provides focused input for subsequent fine analysis algorithms, improving analysis efficiency.

[0096] The fine analysis submodule 141 is used to perform boundary segmentation and depth level classification on the mask of the whole-body burn area, and extract the wound segmentation results and the corresponding burn depth classification results.

[0097] Among them, the wound segmentation result refers to the vector or pixel boundary data that can represent the irregular shape of the wound after delineating the edge of the mask inside the whole-body burn area; the burn depth grading result refers to the pathological damage level label assigned to each segmented wound sub-region according to clinical standards.

[0098] In some embodiments, it receives a filtered full-body burn region mask from an upstream submodule and concentrates computational resources on the region defined by the mask. Within this region, the submodule invokes a high-resolution deep learning network to perform two tasks: first, to perform pixel-level fine-grained boundary segmentation, extracting contours along the boundary between the wound and normal tissue, and outputting wound segmentation results that can be used for area calculation; second, for each segmented wound subregion, to perform multi-category burn depth classification by analyzing its color saturation, texture complexity, gloss reflection, and other detailed features, and output burn depth grading results.

[0099] In practical applications, burn wounds often present complex mixed morphologies. For example, in the same large area of ​​burns, there may be areas of superficial second-degree, deep second-degree, and third-degree burns at the same time, with blurred boundaries. It is difficult for segmentation networks to perform pixel-level segmentation and classification of multiple categories simultaneously.

[0100] In some preferred embodiments, a hierarchical iterative analysis strategy based on confidence feedback can be employed. The specific execution process of this strategy is as follows: The submodule first performs an initial multi-class depth-level segmentation and classification on the entire whole-body burn area mask. During this process, not only is a preliminary depth partition map output, but a classification confidence score is also generated for each pixel. Subsequently, the system identifies and marks pixel regions with low classification confidence scores (usually corresponding to blurred boundaries between different depth levels) as "undetermined regions" based on a preset confidence threshold. In the first round of analysis, the system temporarily suspends the final classification of these undetermined regions, prioritizing ensuring that the segmentation results of high-confidence regions (i.e., easily identifiable, typical regions at each depth level) are solidified. In the iterative phase, for each isolated "undetermined region," the system analyzes its spatially adjacent regions that have already undergone high-confidence classification. For example, if a region to be classified is adjacent to both a defined "shallow II" region and a "III" region, the system will reduce the original global multi-class classification problem targeting that region to a local binary classification problem containing only the two options of "shallow II" and "III". In this way, a complex multi-classification task is decomposed into multiple locally defined binary classification sub-tasks with clear boundaries and specific objectives, thereby improving the accuracy of segmentation and classification of each sub-region in mixed-type deep burn scenarios.

[0101] Feature merging submodule 142 is used to merge and encapsulate the whole-body burn area mask, wound segmentation results and burn depth grading results as the output of wound analysis results.

[0102] The feature package merging submodule 142 is responsible for data integration and standardized output in the system. It encapsulates various data from different analysis stages with different granularities in a structured manner. This submodule ensures that the wound analysis results output downstream are a complete and uniformly formatted data package, facilitating subsequent modules to call and parse them.

[0103] In some preferred embodiments, the fine analysis submodule 141 is further configured to: generate a wound surface model based on the multiple wound images from different perspectives when multiple wound images of the same visible burn area are detected through metadata; and update the wound segmentation results based on the wound surface model.

[0104] Among them, the wound surface model refers to the point cloud or mesh data that can represent the true three-dimensional concave and convex shape of the wound surface, which is recovered from a two-dimensional image sequence through multi-view geometric algorithms.

[0105] In some embodiments, the fine analysis submodule is configured as an enhancement processing system capable of extracting three-dimensional information from two-dimensional images. Its execution logic is as follows: When the system runs, it first determines whether the batch of images being processed belongs to a sequence of multi-angle photographs of the same wound by parsing metadata (such as wound number, shooting timestamp, etc.). Upon confirmation, the submodule no longer processes these images as isolated planes but activates its three-dimensional reconstruction function. It utilizes the parallax and feature correspondence between these images, employing algorithms such as structure-of-motion reconstruction or multi-view stereo matching, to calculate and generate a wound surface model that reflects the true physical contours of the wound area. Finally, it projects the wound segmentation result originally calculated on the two-dimensional plane onto this three-dimensional model and optimizes and corrects it using geometric constraints in three-dimensional space, thereby outputting a wound segmentation result that is closer to the actual anatomical structure and updated with three-dimensional information.

[0106] In some specific embodiments, a feature point detection algorithm can be used first to find and match corresponding points in multiple images; then, the relative pose of the camera and the sparse structure of the 3D point cloud can be estimated through epipolar geometric constraints; finally, a dense matching algorithm can be used to encrypt the sparse point cloud to generate the final wound surface model, which is not limited here.

[0107] In some other specific embodiments, a deep learning-based neural radiation field technique can be used to train a neural network to implicitly represent the entire three-dimensional scene by taking multiple wound images from different perspectives as input. Then, the wound surface model is rendered and extracted from this trained network, and the segmentation results are optimized accordingly. No limitation is made here.

[0108] The quantitative calculation module 105 is configured to perform calculations based on the wound analysis results, the body surface positioning results, and auxiliary information to obtain quantitative results.

[0109] Among them, quantitative results refer to a set of numerical indicators used to objectively describe the degree of wound, such as area and percentage.

[0110] The main task of the quantitative calculation module 105 is to transform the geometric figures (wound analysis results) and spatial locations (body surface localization results) output by the upstream modules into standardized numerical indicators that are meaningful for clinical decision-making. This module ensures the objectivity and reproducibility of wound assessment.

[0111] In some preferred embodiments, the quantization calculation module 105 specifically includes:

[0112] The percentage aggregation calculation submodule is used to parse the preset area weights of the corresponding body surface regions from the human body template; calculate the area percentage of the target wound image in the corresponding body surface region based on the wound analysis results; perform a weighted multiplication aggregation operation on the area weights and area percentages to obtain the total burn body surface area percentage, and put it into the quantification result.

[0113] Among them, area weight refers to a fixed coefficient that is preset for each anatomical region of the human body template according to the standard human body surface area estimation method, representing its percentage of the total surface area.

[0114] In some embodiments, the proportion aggregation calculation submodule is configured as a computational unit that fuses pixel statistics from machine vision with prior knowledge from clinical medicine. Specifically, the proportion aggregation calculation submodule obtains a preset area weight corresponding to the body surface localization result; based on this preset area weight and the area proportion of the wound in the corresponding body part of the target wound image, it performs a weighted multiplicative aggregation operation. This method combines anatomical prior area weights to weighted evaluate the image pixel area proportion, improving the objective reference value of the total burn body surface area proportion calculation result.

[0115] In some specific embodiments, a discrete partitioning calculation method can be used to perform a logical AND operation between the wound segmentation mask and the human body template partitioning map to calculate the pixel area of ​​the wound in each partition. Then, the pixel area in each partition is multiplied by its corresponding area weight and summed. This is not limited here.

[0116] In some other specific embodiments, the proportion aggregation calculation submodule may use the continuous surface integration method. On the three-dimensional human body template, the wound segmentation mask is transformed into a surface patch, and then the area integration is performed directly on the preset continuous weight function on the surface to obtain a total burn surface area proportion. This is not limited here.

[0117] In some preferred embodiments, the quantization calculation module 105 further includes:

[0118] The physical area calculation submodule is used to extract the scale conversion coefficient mapped by the scale marker information from the physical benchmark features; extract the equivalent pixel area corresponding to the wound segmentation region in the wound analysis results; multiply the scale conversion coefficient and the equivalent pixel area algebraically to calculate the absolute area of ​​the wound, and put it into the quantization result.

[0119] Among them, the scale conversion factor is a proportional constant used to establish the correspondence between image pixel distance and real-world physical length; the equivalent pixel area refers to the total number of pixels enclosed by the wound segmentation region on the two-dimensional image plane.

[0120] In some embodiments, the physical area calculation submodule is configured to provide a physically meaningful metric in addition to relative area proportions. Its internal data processing flow is as follows: First, the submodule is triggered to parse the physical reference features carried in the auxiliary information and extract the scale conversion coefficients calculated from the image information of scale markers (such as a reference object of known length). These coefficients are crucial for establishing the connection between the virtual pixel world and the real physical world. Subsequently, the submodule obtains a pixel-level wound segmentation mask from the upstream wound analysis results and calculates the equivalent pixel area covered by the mask. Finally, it performs an algebraic multiplication operation between the scale conversion coefficients and the equivalent pixel area, thereby converting the unitless number of pixels into the absolute wound area with a defined physical unit (such as square centimeters).

[0121] In some specific embodiments, the system automatically detects predefined checkerboard patterns or specific patterns in the image as scale markers, calculates the ratio of their pixel size in the image to the pre-stored physical size, generates scale conversion coefficients, and uses them for absolute area calculation; this is not limited here.

[0122] In some other specific embodiments, the physical area calculation submodule can be implemented by the operator manually dragging a virtual ruler on the interface when acquiring images, aligning its two ends with objects of known length in the image. The system then calculates the scale conversion factor based on the pixel length and standard length of the object input by the operator, and applies it to the calculation of the absolute area of ​​the wound. This is not limited here.

[0123] Please see Figure 5 and Figure 6 , Figure 5 This is a schematic diagram of the electronic burn map in the embodiments of this application; Figure 6 yes Figure 5 A magnified view of a portion of the image.

[0124] Template mapping module 106 is configured to map wound analysis results and quantification results to a human body template according to the body surface positioning results, thereby generating an electronic burn map.

[0125] Among them, electronic burn maps refer to standardized medical atlases that graphically reproduce the location, shape, depth, and quantitative data of wounds on a human body template.

[0126] The template mapping module 106 receives all analysis and computation data from upstream sources and graphically reconstructs it on a standardized two-dimensional or three-dimensional human body model. The electronic burn map generated by this module allows doctors to understand the distribution and severity of burns throughout the patient's body, improving the readability of medical records.

[0127] The medical record generation module 108 is configured to generate and output electronic medical records based on electronic burn images.

[0128] The medical record generation module 108, as the terminal of the data processing flow, is responsible for integrating and formatting all structured information processed and visualized by the system. Based on the preset medical record template, it automatically fills in the electronic burn image and related quantitative values, timestamps and other metadata to generate a standardized electronic medical record document that can be directly archived, realizing a closed loop from raw images to standardized clinical records.

[0129] In some preferred embodiments, a doctor interaction module 107 is also included, configured to generate and provide a graphical user interface. This interface receives correction or confirmation instructions from the operator regarding the system-generated wound boundaries, surface locations, depth grading, and area results, and updates or confirms the aforementioned data based on these instructions. By introducing this module, the final generated electronic medical record is based on results that have been manually verified, balancing automation efficiency with the rigor of clinical application.

[0130] In some preferred embodiments, a follow-up management module 109 is also included. The follow-up management module 109 is used to: acquire a first wound record generated at a first historical time node and a second wound record generated at a second current time node for the same patient; align the first wound record and the second wound record and map them to the human body template coordinate system; perform a time-varying comparison on the first wound record and the second wound record within the human body template coordinate system, and calculate and generate a comparison result output.

[0131] Among them, the first wound record and the second wound record refer to complete data packages containing electronic burn images and related quantitative results generated at different time points; the human body template coordinate system refers to the fixed global spatial coordinate system defined by the human body template.

[0132] In some embodiments, the follow-up management module 109 is configured as a functional unit to realize longitudinal comparison and dynamic assessment of wound progression. Its working logic is as follows: When follow-up comparison is required, the module first retrieves wound records from the database at least two different time points based on the patient's identifier and timestamp. Since the shooting angle and distance may differ each time, a key step performed by this module is to use the pre-calculated surface localization results in each wound record to map and align all data from these different source images into the same background human body template coordinate system. Under this unified metric, two originally incomparable records become standardized data layers that can be compared within the same coordinate system. Finally, the module performs difference analysis (such as logical subtraction and overlap calculation) on these two data layers within this coordinate system, thereby quantitatively calculating the comparison results such as the increase or decrease in wound extent and changes in depth.

[0133] In some specific embodiments, the two electronic burn images can be superimposed pixel by pixel in the human body template coordinate system, and a difference visualization map can be generated by calculating the color difference. The map highlights the areas of wound healing, expansion or no change with different colors, and calculates the percentage change in area of ​​each area simultaneously. This is not limited here.

[0134] In some other specific embodiments, the wound division boundaries in the first wound record and the second wound record can be displayed on a three-dimensional human body template to generate two three-dimensional contour lines of different colors. At the same time, the trend curves of the total burn area and the area of ​​each depth level changing over time can be displayed in the sidebar in the form of data charts, and quantitative comparison results such as healing rate or deterioration rate can be output. This is not limited here.

[0135] In some preferred embodiments, the system further includes an image preprocessing module 102, which performs quality detection on the target wound image to select a first intermediate image that meets preset standards; performs boundary cropping on the first intermediate image to obtain a second intermediate image; performs noise reduction, background interference suppression, brightness and color standardization on the second intermediate image to obtain a third intermediate image; performs data desensitization on sensitive identity information in the third intermediate image to obtain a fourth intermediate image; performs spatial distortion correction on the fourth intermediate image using physical reference features as a marker reference to obtain a fifth intermediate image; and updates the target wound image based on the fifth intermediate image.

[0136] The image preprocessing module 102 is the starting point of the system's data processing pipeline. It is responsible for performing a series of standardization and optimization operations on the raw, inconsistent-quality input images, including removing low-quality data, removing irrelevant background and noise, unifying optical properties, protecting patient privacy, and correcting geometric distortions. The image preprocessing module filters out redundant noise and unifies image benchmarks, providing a standardized data source for subsequent feature extraction of the target wound image, thereby ensuring the stability and security of the entire system.

[0137] To facilitate the explanation of the technical solution of the present invention, the process of generating electronic medical records based on burn wound images can be abstractly represented as the following function mapping relationship.

[0138] Let the first The burned wound images obtained in the second acquisition were The auxiliary information related to this collection is: Therefore, the overall processing procedure of the present invention can be expressed as follows: .

[0139] in, This represents the final structured electronic medical record result. This represents a composite mapping process consisting of multiple processing modules.

[0140] Furthermore, the composite mapping process can be decomposed into: .

[0141] in, The symbol represents compound operators. The order of operations in function composition is from right to left. Here, the symbol is used to concisely express a multi-step algorithm with a specific order. This represents the image preprocessing mapping. This represents the mapping between body surface region identification and template anchor point generation. This represents the wound analysis mapping. Represents quantization calculation mapping, This indicates the mapping between template mapping and electronic burn map generation. This indicates a doctor-interactive verification mapping. This represents the structured generation mapping of electronic medical records.

[0142] Original wound image Preprocessing is performed to obtain a standardized image: .

[0143] in, This represents image data that has undergone quality inspection, noise reduction, cropping, brightness and color normalization, background suppression, desensitization, and optional reference marker correction.

[0144] Based on the preprocessed image data, body surface region recognition and template anchor point generation are performed to obtain the body surface localization results: .

[0145] in, This indicates the surface localization result corresponding to the wound. Preferably, when using the dense mapping method of the human body surface, It can be represented as: .

[0146] in, Indicates the identification of body surface regions. and Indicates the template parameter coordinates. Indicates the confidence level of identification. When At that time, the manual confirmation process is triggered: .

[0147] in, To preset the reliability threshold, This indicates a manual correction operation. When manual correction is triggered, the surface localization results in subsequent mapping processes are... Alternative .

[0148] Furthermore, when reference markers are present in the image, local wound anchor point information can be generated: .

[0149] in, This represents the anchor point result of the local wound area in a standard human body template. Preferably, .

[0150] in, Indicates reference marker identification, This indicates the anchor point location confidence level. This represents the local parameter coordinates of the anchor point within the template area. The anchor point results are used to establish the correspondence between whole-body images, local wound images, and standard human body templates.

[0151] The wound was analyzed based on the preprocessed image data, surface localization results, and anchor point results: .

[0152] in, This indicates the results of wound analysis. Preferably, the wound analysis results can be divided into coarse analysis results across the whole body and fine analysis results for specific areas: .

[0153] in, This indicates the results of coarse-grained burns throughout the body. This indicates results for areas that are obscured or cannot be evaluated. This indicates the results of fine segmentation of the local wound. This indicates the burn depth grading results.

[0154] When considering only the detailed analysis of local wound areas, the results of local wound segmentation can be expressed as: .

[0155] The burn depth grading results can be expressed as: .

[0156] in, This represents the segmentation mapping of the wound region. This represents a mapping of burn depth grading.

[0157] When multiple views of the same wound exist, local surface restoration can be performed based on the local multi-view images to obtain a local wound surface representation: .

[0158] in, This represents a local wound surface model, a local mesh, or a local lesion surface area. This represents locally multi-view normalized image data. When locally multi-view images do not exist, let... .

[0159] Based on the wound analysis results, body surface localization results, auxiliary information, and optional local surface representations, the wound is quantitatively calculated: .

[0160] in, The quantitative results are represented. Preferably, the quantitative results include information such as the proportion of the template region, the proportion of the total burn surface area, and the absolute area.

[0161] When calculating the percentage of total burn surface area, it can be expressed as: .

[0162] in, Indicates the first human body Area weight of each template region This indicates the percentage of the area covered by the wound in the corresponding region.

[0163] When scale markers are present in the image, the absolute area of ​​the wound can be further calculated: .

[0164] in, Indicates the scale conversion factor. This represents the pixel area or equivalent area of ​​a segmented local wound.

[0165] When there is a local surface representation Furthermore, the surface area of ​​the wound on the local three-dimensional surface can be calculated: .

[0166] in, This represents the surface area of ​​the wound. This describes the area calculation process based on a local surface model.

[0167] The surface localization results, anchor point results, wound analysis results, and quantification results are mapped to a standard human body template to obtain an electronic burn map: .

[0168] in, The mapping process for representing electronic burn images is preferably a process of writing the local wound lesion area back into the standard human body template under the constraint of the corresponding template area.

[0169] The doctor verified the electronic burn image and analysis results, and obtained the final diagnosis: .

[0170] in, This indicates the correction information entered by the doctor. This represents the electronic burn image after it has been verified by a doctor.

[0171] Based on the doctor's verification results, a structured electronic medical record is automatically generated: .

[0172] in, This includes the location of the wound, the extent of the wound, the depth of the burn, area indicators, time information, and the correlation between the original image and the electronic burn map.

[0173] For the same patient at different time points and Wound record, can be represented as: , , .

[0174] The follow-up comparison results are defined as follows: .

[0175] in, This indicates the changes in the wound at different points in time, reflecting the dynamic evolution trend of the wound's extent, depth, and area. This indicates the comparison process.

[0176] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An electronic medical record generation system based on burn wound images, characterized in that, include: The image acquisition module is used to acquire images of the target wound and corresponding auxiliary information; the auxiliary information includes physical reference features, imaging features, and metadata; A body surface location recognition module is used to locate the target wound image in a preset human body template, specifically including: The dense mapping submodule is used to scan the target wound image and generate a physical topology map characterizing the human body structure; The physical topology map is compared with the constant sequences of different body surface partitions to obtain the target constant sequence and body surface partition identifier of the target body surface partition to which the physical topology map belongs. Based on the body surface partition identifier, the corresponding two-dimensional unfolded surface is retrieved; and the physical topology map and the target constant sequence are mapped to the two-dimensional unfolded surface to perform coordinate operations and solve for the template parameter coordinates. Anchor point generation submodule is used to extract and calculate the intermediate parameter coordinates of the target wound image in the observation space system based on the physical reference features; The positioning and fusion submodule is used to establish the coordinate correspondence between the target wound image and the human template based on the body surface partition identifier, the template parameter coordinates and the intermediate parameter coordinates, and merge the output as the body surface positioning result. The wound analysis module is used to perform boundary segmentation and depth level classification on the target wound image to generate wound analysis results. The quantitative calculation module is used to perform calculations based on the wound analysis results, the body surface positioning results, and the auxiliary information to obtain quantitative results. The template mapping module is used to map the wound analysis results and the quantification results to the human body template according to the body surface positioning results, so as to generate an electronic burn map. The medical record generation module is used to generate and output electronic medical records based on the electronic burn images.

2. The electronic medical record generation system based on burn wound images according to claim 1, characterized in that, The dense mapping submodule is also used for: If it is determined that the template parameter coordinates cannot be output or the output confidence scalar is lower than a preset safety threshold, the physical topology map is subjected to continuity analysis to identify feature missing distribution areas where the feature faults exceed the threshold. A failure mask is constructed in the physical topology map to cover and isolate the feature missing distribution area, and the physical topology map is updated. Based on the physical reference features and metadata in the auxiliary information, a series of continuous reference images adjacent to the target wound image on the spatial shooting trajectory are determined; Extract the connected physical topology map corresponding to the continuous reference image, and perform edge feature matching and fusion between the physical topology map and the connected physical topology map to generate a joint physical topology map; The joint physical topology map is compared with the constant sequences of different continuum surface partitions to obtain the target continuous constant sequence and the associated continuum surface partition identifier, wherein the continuum surface partition is composed of at least two surface partitions with boundary connection relationship. Based on the partition identifier of the associated continuum table, the corresponding two-dimensional unfolded surfaces are spliced ​​together in the background space to obtain a joint two-dimensional unfolded surface; The joint physical topology map and the target continuous constant sequence are mapped onto the joint two-dimensional unfolded surface, and coordinate operations are performed to obtain the template parameter coordinates.

3. The electronic medical record generation system based on burn wound images according to claim 1, characterized in that, The body surface location recognition module also includes: An exception-triggered replacement submodule is used to obtain the confidence vector corresponding to the template parameter coordinates output by the dense mapping submodule; If the confidence vector is lower than a preset confidence threshold, external input coordinates are received, and the template parameter coordinates are updated using the external input coordinates.

4. The electronic medical record generation system based on burn wound images according to claim 1, characterized in that, The wound analysis module specifically includes: The global surface localization submodule is used to identify visible burn areas, non-burn areas and unidentifiable areas based on the target wound image, and output a whole-body burn area mask. The fine analysis submodule is used to perform boundary segmentation and depth level classification on the mask of the whole-body burn area, and extract and output the wound segmentation results and the corresponding burn depth grading results. The feature merging submodule is used to merge and encapsulate the whole-body burn area mask, the wound segmentation result, and the burn depth grading result as the output of the wound analysis result.

5. The electronic medical record generation system based on burn wound images according to claim 4, characterized in that, The fine analysis submodule is also used for: In the case where multiple wound images of the same visible burn area from different perspectives are detected through the metadata, a wound surface model is generated based on the different wound images; The wound segmentation results are updated based on the wound surface model.

6. The electronic medical record generation system based on burn wound images according to claim 1, characterized in that, The quantization calculation module specifically includes: The proportion aggregation calculation submodule is used to parse the preset area weights of the corresponding body surface regions from the human body template; Based on the wound analysis results, the area ratio of the target wound image in the corresponding body surface region is calculated; The area weight and the area percentage are aggregated by a weighted multiplication operation to obtain the total burn surface area percentage, which is then included in the quantification result.

7. The electronic medical record generation system based on burn wound images according to claim 6, characterized in that, The quantization calculation module also includes: The physical area calculation submodule is used to extract the scale conversion coefficients mapped by the scale marker information from the physical reference features; Extract the equivalent pixel area corresponding to the segmented region of the wound within the wound analysis results; The scale conversion factor is algebraically multiplied by the equivalent pixel area to calculate the absolute area of ​​the wound, which is then incorporated into the quantization result.

8. The electronic medical record generation system based on burn wound images according to claim 1, characterized in that, It also includes a follow-up management module, which is used for: Acquire the first wound record generated at the first historical time point and the second wound record generated at the second current time point for the same patient; Align the first wound record with the second wound record and map them together into the human body template coordinate system; Within the coordinate system, a time-varying comparison is performed on the first wound record and the second wound record, and the comparison results are calculated and output.

9. The electronic medical record generation system based on burn wound images according to claim 1, characterized in that, The system also includes an image preprocessing module, used to perform quality detection on the target wound image and select a first intermediate image that meets preset standards; A second intermediate image is obtained by performing boundary cropping on the first intermediate image. The second intermediate image is subjected to denoising, background interference suppression, brightness and color normalization to obtain the third intermediate image; The sensitive identity information in the third intermediate image is subjected to data desensitization processing to obtain the fourth intermediate image; Using the physical reference features as a marker reference, spatial distortion correction is performed on the fourth intermediate image to obtain a fifth intermediate image, and the target wound image is updated based on the fifth intermediate image.

Images