A method and system for segmenting a burn wound area

Abstract

The application provides a burn wound area segmentation method and system, constructs an anatomy database, obtains a to-be-segmented image, adopts a lightweight CNN classifier to extract a feature vector, determines a healing stage of the burn wound according to the feature vector, adjusts boundary parameters according to the healing stage, inputs the to-be-segmented image into an improved U-Net encoder network, and outputs a segmentation mask and a boundary probability graph; determines an interactive region according to the boundary probability graph, and performs interaction in the interactive region by referring to the segmentation mask, generates a smooth boundary according to the interactive data, performs post-processing optimization of the smooth boundary under a priori constraint, obtains a target boundary, and completes burn wound area segmentation. Specifically, the priori knowledge provides a basic constraint for segmentation, the interactive operation corrects a complex region in a targeted manner, and the post-processing realizes global consistency optimization, forming a closed-loop processing flow of 'constraint-correction-optimization'.

Description

Technical Field

[0001] This invention belongs to the field of burn wound area segmentation technology, and specifically relates to a method and system for burn wound area segmentation. Background Technology

[0002] In the field of burn medicine, accurately identifying and segmenting burn wounds that have damaged the dermis but still have dermal remnants and do not involve muscle, bone, nerve, or tendon tissues is of great significance for clinical diagnosis, treatment planning, and prognosis assessment.

[0003] Most current burn wound segmentation models are trained on specific datasets, making the collection of training sets particularly important. It's worth noting that wounds that have damaged the dermis but still retain some dermal tissue often have unclear boundaries with normal tissue, exhibiting a gradual transition. Furthermore, the wound's shape is often irregular, posing significant challenges to annotation. Differences in the judgment of wounds by experienced physicians can lead to low consistency in annotation results, affecting the quality of the training dataset and consequently reducing the model's segmentation accuracy. Summary of the Invention

[0004] Based on this, the present invention provides a method and system for segmenting burn wound areas, which aims to improve the segmentation accuracy of fuzzy and irregular boundaries while ensuring segmentation efficiency through the deep integration of prior knowledge and interactive operation.

[0005] A first aspect of the present invention provides a method for segmenting burn wound areas, the method comprising:

[0006] An anatomical database is constructed, which includes skin thickness, dermal residual probability distribution, and typical boundary morphology vectors for different parts of the human body.

[0007] The image to be segmented is obtained, and a lightweight CNN classifier is used to extract feature vectors. The feature vectors include the proportion of red and swollen area, the mean gray value of the exudate area, the wound texture entropy, and the boundary blur.

[0008] Based on the feature vector, the healing stage of the burn wound is determined, and the boundary parameters are adjusted according to the healing stage.

[0009] The image to be segmented is input into the improved U-Net encoder network, which outputs a segmentation mask and a boundary probability map. The boundary parameters are used as external parameter inputs to the improved U-Net encoder network to dynamically adjust the feature extraction strategy.

[0010] Based on the boundary probability map, the interaction region is determined, and with reference to the segmentation mask, interaction is performed within the interaction region. Based on the interaction data, a smooth boundary is generated.

[0011] The smooth boundary is optimized by post-processing with fusion prior constraints to obtain the target boundary, thereby completing the segmentation of the burn wound area.

[0012] Furthermore, the improved U-Net encoder network is based on ResNet-34 as the basic skeleton. By introducing a multi-scale feature fusion module and dilated convolutional layers, it enhances the ability to capture burn wound features. Among them, an anatomical feature mapping function is embedded in the third layer of the improved U-Net encoder network.

[0013] Furthermore, the expression for the anatomical feature mapping function is:

[0014] ;

[0015] Among them, w t w p w b These are the weighting coefficients. The image pixel coordinates are Skin thickness, The image pixel coordinates are The probability that dermis remains after a burn, B m The component in the matched typical boundary shape vector that is most similar to the current image. This is the function output value, used to adjust the feature extraction weights of the corresponding pixel.

[0016] Furthermore, in the step of determining the healing stage of the burn wound based on the feature vector, and adjusting the boundary parameters based on the healing stage, the expression for determining the healing stage of the burn wound is as follows:

[0017] ;

[0018] Where σ is the Softmax function, W is a 3×4 weight matrix, b is the bias vector, and the output S∈{0,1,2} corresponds to the acute phase, inflammatory phase, and repair phase, respectively. v1 represents the percentage of redness and swelling area, v2 represents the average gray value of the exudate area, v3 represents the wound texture entropy, and v4 represents the boundary ambiguity.

[0019] Furthermore, the step of determining the interaction region based on the boundary probability map, performing interaction within the interaction region with reference to the segmentation mask, and generating a smooth boundary based on the interaction data includes:

[0020] Based on the boundary probability map, the uncertainty index of each boundary pixel is calculated, and boundary pixels with uncertainty indices greater than a first threshold are marked. The marked boundary pixels are combined to obtain the interaction region.

[0021] Control points are manually marked within the interactive area, and B-spline curves are used to fit the control points to obtain the first smooth boundary.

[0022] The graffiti area is marked manually within the interactive area, the feature mean of the pixels within the graffiti area is calculated, and a template area with a feature deviation less than a second threshold is retrieved from the prior database based on the feature mean.

[0023] Calculate the matching degree between the graffiti area and the template area, and determine whether the matching degree is greater than a third threshold;

[0024] If the matching degree is determined to be greater than the third threshold, the boundary transition parameters of the corresponding template region are called, and a second smooth boundary is generated based on the boundary transition parameters.

[0025] Furthermore, the step of performing post-processing optimization on the smooth boundary by fusing prior constraints to obtain the target boundary includes:

[0026] Morphological operations are used to smooth the first and second smooth boundaries, and the degree of agreement between the smoothed boundaries and anatomical priors is calculated.

[0027] Determine whether the degree of fit is greater than the fourth threshold;

[0028] If the degree of similarity is determined to be greater than the fourth threshold, then the boundary is adjusted by combining the initial segmentation boundary probability map and the constraints of the interaction markers, and the target boundary is obtained.

[0029] Furthermore, the step of calling the boundary transition parameters of the corresponding template region and generating a second smooth boundary based on the boundary transition parameters includes:

[0030] After each interaction, the corrected boundary features are recorded, the prior database is updated, and personalized prior knowledge is formed.

[0031] A second aspect of this invention provides a burn wound region segmentation system for implementing the burn wound region segmentation method provided in the first aspect, the system comprising:

[0032] A construction module is used to construct an anatomical database, which includes skin thickness, dermal residual probability distribution, and typical boundary morphology vectors for various parts of the human body.

[0033] The extraction module is used to acquire the image to be segmented. It uses a lightweight CNN classifier to extract feature vectors, which include the proportion of red and swollen area, the mean gray value of the oozing area, the wound texture entropy, and the boundary blur.

[0034] An adjustment module is used to determine the healing stage of the burn wound based on the feature vector, and to adjust the boundary parameters based on the healing stage.

[0035] The input module is used to input the image to be segmented into the improved U-Net encoder network and output a segmentation mask and a boundary probability map. The boundary parameters are used as external parameter inputs of the improved U-Net encoder network to dynamically adjust the feature extraction strategy.

[0036] An interaction module is used to determine the interaction region based on the boundary probability map, and to perform interaction within the interaction region with reference to the segmentation mask, and to generate a smooth boundary based on the interaction data.

[0037] The post-processing optimization module is used to perform post-processing optimization on the smooth boundary by fusing prior constraints to obtain the target boundary, thereby completing the segmentation of the burn wound area.

[0038] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the burn wound region segmentation method provided in the first aspect.

[0039] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the burn wound area segmentation method provided in the first aspect.

[0040] This invention provides a method and system for segmenting burn wound regions. The method involves constructing an anatomical database to acquire images to be segmented, using a lightweight CNN classifier to extract feature vectors, determining the healing stage of the burn wound based on these feature vectors, and adjusting boundary parameters accordingly. The images to be segmented are then input into an improved U-Net encoder network, which outputs a segmentation mask and a boundary probability map. Based on the boundary probability map, interactive regions are determined, and interactions are performed within these regions with reference to the segmentation mask. Smooth boundaries are generated based on the interaction data. Post-processing optimization, incorporating prior constraints, is applied to the smooth boundaries to obtain the target boundaries, thus completing the burn wound region segmentation. Specifically, prior knowledge provides basic constraints for segmentation, interactive operations specifically correct complex regions, and post-processing achieves global consistency optimization, forming a closed-loop processing flow of "constraint-correction-optimization." Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the implementation of a burn wound area segmentation method according to Embodiment 1 of the present invention.

[0042] Figure 2This is a structural block diagram of a burn wound area segmentation system provided in Embodiment 2 of the present invention;

[0043] Figure 3 This is a structural block diagram of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0044] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0045] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0047] Example 1

[0048] Please see Figure 1 , Figure 1 The flowchart of a burn wound region segmentation method provided in Embodiment 1 of the present invention is shown. The burn wound region segmentation method specifically includes steps S01 to S06.

[0049] Step S01: Construct an anatomical database, which includes skin thickness, dermal residual probability distribution, and typical boundary morphology vectors corresponding to various parts of the human body.

[0050] In this embodiment of the invention, the anatomical database may contain three-dimensional anatomical parameters of key human body parts (such as the scalp, face, trunk, elbow, knee joint, etc.). Each part stores three types of core data: skin thickness, dermal residual probability distribution, and typical boundary morphology vector. The typical boundary morphology vector is... , These are morphological parameters such as boundary curvature and serration.

[0051] Step S02: Obtain the image to be segmented, and use a lightweight CNN classifier to extract feature vectors. The feature vectors include the proportion of red and swollen area, the average gray value of the oozing area, the wound texture entropy, and the boundary blur.

[0052] Specifically, the lightweight CNN classifier is a small neural network based on the MobileNet architecture, designed separately, containing 3 convolutional layers, 2 pooling layers, and 1 fully connected layer, for classification tasks.

[0053] Step S03: Determine the healing stage of the burn wound based on the feature vector, and adjust the boundary parameters based on the healing stage.

[0054] It should be noted that the expression for determining the healing stage of a burn wound is as follows:

[0055] ;

[0056] Where σ is the Softmax function, W is a 3×4 weight matrix, b is the bias vector, and the output S∈{0,1,2} corresponds to the acute phase, inflammatory phase, and repair phase, respectively. v1 represents the percentage of redness and swelling area, v2 represents the average gray value of the exudate area, v3 represents the wound texture entropy, and v4 represents the boundary ambiguity.

[0057] Subsequently, based on the corresponding stage, the corresponding boundary parameters are invoked. The boundary parameters include at least the boundary search radius and the ambiguity threshold. For example, in the acute phase (S=0): the boundary search radius R=15 pixels and the ambiguity threshold Th=0.6; in the inflammatory phase (S=1): R=10 pixels and Th=0.5; and in the repair phase (S=2): R=5 pixels and Th=0.4.

[0058] Step S04: Input the image to be segmented into the improved U-Net encoder network, and output a segmentation mask and a boundary probability map. The boundary parameters are used as external parameter inputs to the improved U-Net encoder network to dynamically adjust the feature extraction strategy.

[0059] In this embodiment of the invention, the improved U-Net encoder network is based on ResNet-34 as its basic skeleton. By introducing a multi-scale feature fusion module and dilated convolutional layers, it enhances the ability to capture burn wound features. Specifically, an anatomical feature mapping function is embedded in the third layer of the improved U-Net encoder network. The expression of the anatomical feature mapping function is as follows:

[0060] ;

[0061] Among them, w t w p w bThese are the weighting coefficients. The image pixel coordinates are Skin thickness, The image pixel coordinates are The probability that dermis remains after a burn, B m The component in the matched typical boundary shape vector that is most similar to the current image. This is the function output value, used to adjust the feature extraction weights of the corresponding pixel.

[0062] Specifically, the improved U-Net encoder network includes an input layer, a first stage, a second stage, a third stage, a fourth stage, and an output layer. The input layer receives RGB burn images of 512×512×3 pixels, which are then normalized (pixel values ​​are normalized to the [0,1] range) before being input into the network. The first stage consists of three layers: the first layer is a 7×7 convolutional layer (stride 2, padding=3), with 64 output channels. Combined with batch normalization and the ReLU activation function, its main function is to initially extract the edge and color features of the image; the second layer is a 3×7 convolutional layer (stride 2, padding=3). A 3x3 max-pooling layer (stride 2, padding=1) compresses the output size to 256×256×64, reducing the feature map size while retaining key information. The third layer includes the first residual block (composed of two 3×3 convolutional layers, maintaining 64 channels), using skip connections to alleviate the vanishing gradient problem and enhance feature propagation. The second stage consists of four layers, with layers 4-6 containing the corresponding second- to fourth residual blocks. Each residual block contains two 3×3 convolutional layers, increasing the number of channels from 64 to 128 (dimensionality adjustment is achieved through a 1×1 convolutional layer), thus scaling the output feature map size. The size is 128×128×128. The 7th layer is an embedded dilated convolutional layer (dilation rate = 2), which expands the receptive field without reducing resolution and captures a wider range of contextual information (such as the spatial relationship between the wound and surrounding tissue). The third stage consists of 5 layers. Layers 8-12 include the corresponding 5th-9th residual blocks, increasing the number of channels from 128 to 256. The output feature map size is 64×64×256. The 13th layer is a multi-scale feature fusion layer, which adds the output of this stage to the feature map of Stage 2 (after 1×1 convolutional dimensionality reduction) element-wise, fusing the features. The fourth stage consists of six layers. Layers 14-18 include residual blocks 10-14, increasing the number of channels from 256 to 512. The output feature map size is 32×32×512. Layer 19 is a global average pooling layer (aiding feature extraction), outputting a 1×1×512 global feature vector to assist the classifier's decision-making during the healing stage. The final encoder outputs a 32×32×512 high-dimensional feature map containing deep semantic features of the burn wound (such as tissue damage type, boundary morphology, etc.), providing a basis for the decoder's segmentation operation.

[0063] Boundary parameters are used as external parameter inputs to the improved U-Net encoder network to dynamically adjust the feature extraction strategy. Understandably, after the classifier determines S, the encoder calls the boundary model parameters corresponding to the healing stage (such as the search radius R and the ambiguity threshold Th). These parameters, by modifying the weights of the dilated convolutional layers in the second stage of the encoder (the dilation rate adaptively adjusts with R) and the boundary penalty coefficient of the loss function (Th determines the loss weight in the boundary region), achieve directional constraints on the feature extraction process. Additionally, some features from the encoder's intermediate layers are extracted and input into the classifier, enhancing the classifier's feature representation capabilities. Specifically, the feature map (256×256×64) of the encoder's third layer is global average pooled to output a 64-dimensional feature vector, which is fused with the 4-dimensional feature vector V extracted by the classifier itself to form an enhanced feature vector. This method achieves functional collaboration and complementarity between the classifier and encoder.

[0064] Step S05: Determine the interaction region based on the boundary probability map, and perform interaction within the interaction region with reference to the segmentation mask. Generate a smooth boundary based on the interaction data.

[0065] Specifically, based on the boundary probability map, the uncertainty index of each boundary pixel is calculated, and boundary pixels with uncertainty indices greater than a first threshold are marked. The marked boundary pixels are then combined to obtain the interaction region, i.e., the high-uncertainty region. The formula for calculating the uncertainty index is:

[0066] ;

[0067] Where α is the weight, The deviation of the current pixel from the anatomical prior (calculated using Euclidean distance). Here, H represents the boundary probability map, and H is the information entropy function.

[0068] Control points are manually marked within the interactive area, and B-spline curves are used to fit these control points to obtain a first smooth boundary. Understandably, clicking on key control points (such as points clearly belonging to wounds or normal tissue) on the ambiguous boundary allows the system to automatically generate a boundary curve passing through the control points, incorporating anatomical priors (such as the texture continuity of normal skin in that area). Specifically, the user marks k control points. ( l i =0 indicates normal tissue. l i =1 represents the wound area), and the calculation formula for fitting the control points using a B-spline curve is as follows:

[0069] ;

[0070] in, for p B-spline basis functions, Anatomical constraints are added during the fitting process; for example, the curve curvature must not exceed 1.5 times the typical boundary curvature of the region.

[0071] The graffiti area is marked manually within the interactive area. The feature mean of the pixels within the graffiti area is calculated. Based on the feature mean, a template area with a feature deviation less than a second threshold is called from the prior database. It can be understood that a rough graffiti is made for a large area of ​​blurred area. The prior of the healing stage corresponding to the area (such as the redness and swelling band feature of the wound edge during the inflammatory period) is called to match the graffiti area with the surrounding features and generate a smooth transition boundary.

[0072] Calculate the matching degree between the graffiti area and the template area, and determine whether the matching degree is greater than a third threshold. The formula for calculating the matching degree is:

[0073] ;

[0074] in, The pixel features of the graffiti area G represents the graffiti area, and T represents the template area.

[0075] If the matching degree is determined to be greater than the third threshold, the boundary transition parameters of the corresponding template region are called, and a second smooth boundary is generated based on the boundary transition parameters.

[0076] Furthermore, after each interaction, the corrected boundary features are recorded, the prior database is updated, and personalized prior knowledge is formed. Specifically, this can be represented as:

[0077] ;

[0078] in, The original prior probability, The actual label (0 or 1) is given to the user, and β is the weight of historical data to ensure a balance between the stability and up-to-dateness of prior knowledge.

[0079] Step S06: Perform post-processing optimization on the smooth boundary by fusing prior constraints to obtain the target boundary, thereby completing the segmentation of the burn wound area.

[0080] Specifically, morphological operations are used to smooth the first and second smooth boundaries, and the degree of agreement between the smoothed boundaries and anatomical priors is calculated. It should be noted that 3×3 structuring elements are used for opening and closing operations, removing areas smaller than A. min For a noise region of 20 pixels, the formula for calculating the boundary smoothness is:

[0081] ;

[0082] Where L is the total number of boundary pixels, c i Let be the curvature of the i-th pixel. Given the average curvature, the formula for calculating the fit is:

[0083] ;

[0084] Where N is the number of morphological parameters, b i These are the actual boundary parameters. These are anatomical prior parameters;

[0085] Determine whether the degree of fit is greater than the fourth threshold;

[0086] If the degree of fit is determined to be greater than the fourth threshold, then, combining the initial segmentation boundary probability map and the constraints of the interaction markers, an energy minimization algorithm is used to adjust the boundary to obtain the target boundary. In this embodiment of the invention, the energy minimization algorithm is expressed as:

[0087] ;

[0088] in, For data items (based on the initial probability map). E is the smoothing term (considering the consistency of adjacent pixel labels, i.e., the constraint of interactive labeling), λ is the weight parameter, p represents a single pixel in the image, and q represents the pixel adjacent to pixel p. Fine-tuning of the boundary is achieved by minimizing E.

[0089] The final output can include segmentation masks and boundary confidence heatmaps, with annotations on the correction basis for each boundary segment (such as "anatomical prior correction" and "user interaction confirmation"), providing interpretability support for clinical diagnosis.

[0090] In summary, the burn wound region segmentation method in the above embodiments of the present invention involves constructing an anatomical database to acquire the image to be segmented, employing a lightweight CNN classifier to extract feature vectors, determining the healing stage of the burn wound based on the feature vectors, and adjusting boundary parameters accordingly. The image to be segmented is input into an improved U-Net encoder network, which outputs a segmentation mask and a boundary probability map. Based on the boundary probability map, interactive regions are determined, and interactions are performed within these regions with reference to the segmentation mask. Smooth boundaries are generated based on the interaction data. Post-processing optimization, which integrates prior constraints, is applied to the smooth boundaries to obtain the target boundaries, thus completing the burn wound region segmentation. Specifically, prior knowledge provides basic constraints for segmentation, interactive operations specifically correct complex regions, and post-processing achieves global consistency optimization, forming a closed-loop processing flow of "constraint-correction-optimization."

[0091] Example 2

[0092] Please see Figure 2 , Figure 2 This is a structural block diagram of a burn wound region segmentation system provided in Embodiment 2 of the present invention. The burn wound region segmentation system 200 includes: a construction module 21, an extraction module 22, an adjustment module 23, an input module 24, an interaction module 25, and a post-processing optimization module 26, wherein:

[0093] Module 21 is used to construct an anatomical database, which includes skin thickness, dermal residual probability distribution and typical boundary morphology vectors corresponding to various parts of the human body.

[0094] Extraction module 22 is used to acquire the image to be segmented. It uses a lightweight CNN classifier to extract feature vectors, which include the proportion of red and swollen area, the mean gray value of the oozing area, the wound texture entropy, and the boundary blur.

[0095] Adjustment module 23 is used to determine the healing stage of the burn wound based on the feature vector, and adjust the boundary parameters based on the healing stage to determine the expression for the healing stage of the burn wound as follows:

[0096] ;

[0097] Where σ is the Softmax function, W is a 3×4 weight matrix, b is the bias vector, and the output S∈{0,1,2} corresponds to the acute phase, inflammatory phase, and repair phase, respectively. v1 is the percentage of red and swollen area, v2 is the average gray value of the exudate area, v3 is the wound texture entropy, and v4 is the boundary ambiguity.

[0098] Input module 24 is used to input the image to be segmented into the improved U-Net encoder network and output a segmentation mask and boundary probability map. The boundary parameters are used as external parameter inputs to the improved U-Net encoder network to dynamically adjust the feature extraction strategy. The improved U-Net encoder network is based on ResNet-34 as its basic skeleton and enhances its ability to capture burn wound features by introducing a multi-scale feature fusion module and dilated convolutional layers. An anatomical feature mapping function is embedded in the third layer of the improved U-Net encoder network, and the expression of the anatomical feature mapping function is:

[0099] ;

[0100] Among them, w t w p w b These are the weighting coefficients. The image pixel coordinates are Skin thickness, The image pixel coordinates are The probability that dermis remains after a burn, B m The component in the matched typical boundary shape vector that is most similar to the current image. The output value of the function is used to adjust the feature extraction weights of the corresponding pixel;

[0101] The interaction module 25 is used to determine the interaction region based on the boundary probability map, and to perform interaction within the interaction region with reference to the segmentation mask, and to generate a smooth boundary based on the interaction data.

[0102] The post-processing optimization module 26 is used to perform post-processing optimization on the smooth boundary by fusing prior constraints to obtain the target boundary, so as to complete the segmentation of the burn wound area.

[0103] Furthermore, in other embodiments of the present invention, the interaction module 25 includes:

[0104] The first calculation unit is used to calculate the uncertainty index of each boundary pixel according to the boundary probability map, mark the boundary pixels with uncertainty index greater than a first threshold, and combine the marked boundary pixels to obtain the interaction area.

[0105] The fitting unit is used to manually mark control points within the interactive area and fit the control points using a B-spline curve to obtain a first smooth boundary.

[0106] The second calculation unit is used to manually mark the graffiti area in the interactive area, calculate the feature mean of the pixels in the graffiti area, and call the template area with feature deviation less than the second threshold in the prior database based on the feature mean.

[0107] The first judgment unit is used to calculate the matching degree between the graffiti area and the template area, and to determine whether the matching degree is greater than a third threshold.

[0108] The calling unit is used to call the boundary transition parameters of the corresponding template region if the matching degree is determined to be greater than the third threshold, and generate a second smooth boundary according to the boundary transition parameters.

[0109] Furthermore, in other embodiments of the present invention, the post-processing optimization module 26 includes:

[0110] The smoothing unit is used to perform morphological operations to smooth the first smooth boundary and the second smooth boundary, and to calculate the degree of agreement between the smoothed boundary and the anatomical prior.

[0111] The second judgment unit is used to determine whether the degree of matching is greater than the fourth threshold.

[0112] The adjustment unit is used to adjust the boundary by combining the initial segmentation boundary probability map and the constraints of the interaction markers if the degree of similarity is greater than the fourth threshold, so as to obtain the target boundary.

[0113] Furthermore, in other embodiments of the present invention, the interaction module 25 further includes:

[0114] The update unit is used to record the corrected boundary features after each interaction, update the prior database, and form personalized prior knowledge.

[0115] Example 3

[0116] In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3 The image shows an electronic device according to Embodiment 3 of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, it implements the burn wound area segmentation method as described above.

[0117] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0118] The memory 20 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, SmartMediaCard (SMC), SecureDigital (SD) card, FlashCard, etc., equipped on the electronic device. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0119] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0120] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the burn wound area segmentation method described above.

[0121] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0122] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0123] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0124] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0125] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for segmenting burn wound areas, characterized in that, The method includes: An anatomical database is constructed, which includes skin thickness, dermal residual probability distribution, and typical boundary morphology vectors for different parts of the human body. The image to be segmented is obtained, and a lightweight CNN classifier is used to extract feature vectors. The feature vectors include the proportion of red and swollen area, the mean gray value of the exudate area, the wound texture entropy, and the boundary blur. Based on the aforementioned feature vector, the expression for the healing stage of the burn wound is determined as follows: ； Where σ is the Softmax function, W is a 3×4 weight matrix, b is the bias vector, and the output S∈{0,1,2} corresponds to the acute phase, inflammatory phase, and repair phase, respectively. v1 is the percentage of redness and swelling area, v2 is the average gray value of the exudate area, v3 is the wound texture entropy, v4 is the boundary ambiguity, and the boundary parameters are adjusted according to the healing stage. The image to be segmented is input into the improved U-Net encoder network, which outputs a segmentation mask and a boundary probability map. The boundary parameters are used as external parameter inputs to the improved U-Net encoder network to dynamically adjust the feature extraction strategy. Based on the boundary probability map, the interaction region is determined, and with reference to the segmentation mask, interaction is performed within the interaction region. Based on the interaction data, a smooth boundary is generated, including: Based on the boundary probability map, the uncertainty index of each boundary pixel is calculated, and boundary pixels with uncertainty indices greater than a first threshold are marked. The marked boundary pixels are combined to obtain the interaction region. Control points are manually marked within the interactive area, and B-spline curves are used to fit the control points to obtain the first smooth boundary. The graffiti area is marked manually within the interactive area, the feature mean of the pixels within the graffiti area is calculated, and a template area with a feature deviation less than a second threshold is retrieved from the prior database based on the feature mean. Calculate the matching degree between the graffiti area and the template area, and determine whether the matching degree is greater than a third threshold; If the matching degree is determined to be greater than the third threshold, the boundary transition parameters of the corresponding template region are called, and a second smooth boundary is generated based on the boundary transition parameters. The smooth boundary is optimized by post-processing with prior constraints. The degree of agreement between the smoothed boundary and the anatomical prior is calculated. If the degree of agreement is greater than the fourth threshold, the boundary is adjusted by energy minimization algorithm in combination with the boundary probability map of the initial segmentation and the constraints of the interaction markers to obtain the target boundary and complete the segmentation of the burn wound area. The improved U-Net encoder network is based on ResNet-34. By introducing a multi-scale feature fusion module and dilated convolutional layers, it enhances the ability to capture burn wound features. Among them, an anatomical feature mapping function is embedded in the third layer of the improved U-Net encoder network. The expression for the anatomical feature mapping function is: ； Among them, w t w p w b These are the weighting coefficients. The image pixel coordinates are Skin thickness, The image pixel coordinates are The probability that dermis remains after a burn, B m The component in the matched typical boundary shape vector that is most similar to the current image. This is the function output value, used to adjust the feature extraction weights of the corresponding pixel.

2. The method for segmenting burn wound areas according to claim 1, characterized in that, The step of performing post-processing optimization on the smooth boundary by fusing prior constraints to obtain the target boundary further includes: Morphological operations are used to smooth the first smooth boundary and the second smooth boundary.

3. The method for segmenting burn wound areas according to claim 2, characterized in that, The step of calling the boundary transition parameters of the corresponding template region and generating the second smooth boundary based on the boundary transition parameters includes: After each interaction, the corrected boundary features are recorded, the prior database is updated, and personalized prior knowledge is formed.

4. A burn wound area segmentation system, characterized in that, The system for implementing the burn wound area segmentation method according to any one of claims 1-3, the system comprising: A construction module is used to construct an anatomical database, which includes skin thickness, dermal residual probability distribution, and typical boundary morphology vectors for various parts of the human body. The extraction module is used to acquire the image to be segmented. It uses a lightweight CNN classifier to extract feature vectors, which include the proportion of red and swollen area, the mean gray value of the oozing area, the wound texture entropy, and the boundary blur. An adjustment module is used to determine the healing stage of the burn wound based on the feature vector, and to adjust the boundary parameters based on the healing stage. The input module is used to input the image to be segmented into the improved U-Net encoder network and output a segmentation mask and a boundary probability map. The boundary parameters are used as external parameter inputs of the improved U-Net encoder network to dynamically adjust the feature extraction strategy. An interaction module is used to determine the interaction region based on the boundary probability map, and to perform interaction within the interaction region with reference to the segmentation mask, and to generate a smooth boundary based on the interaction data. The post-processing optimization module is used to perform post-processing optimization on the smooth boundary by fusing prior constraints to obtain the target boundary, thereby completing the segmentation of the burn wound area.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the burn wound region segmentation method as described in any one of claims 1-3.

6. An electronic device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the program, implements the burn wound area segmentation method as described in any one of claims 1-3.

Images