Chronic wound assessment and diagnosis system based on intelligent image recognition

Abstract

The invention belongs to the technical field of wound assessment, and discloses a chronic wound assessment and diagnosis system based on intelligent image recognition, which comprises an image acquisition module, a region recognition module, a three-dimensional reconstruction module, an assessment module and a treatment scheme recommendation module, the image acquisition module can acquire multi-source heterogeneous chronic wound image data through a unified interface protocol; and the region identification module can segment a wound region from the processed chronic wound image data through a U-Net3 network algorithm and an analysis model, generate an edge weight map, classify necrotic tissues and analyze the trend of the restored sinus tract. The wound area is accurately segmented through the deep learning network, the wound features are intelligently recognized in combination with the image, the accuracy and reliability of evaluation are improved, the dynamic three-dimensional point cloud model is further constructed by applying the binocular vision principle, the spatial form, depth, area, volume, sneaking, sinus tract and flaccidity duct information of the wound is visually displayed, and the evaluation accuracy and reliability are improved. And the subsequent wound assessment effect can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wound assessment, and particularly relates to a chronic wound assessment and diagnosis system based on image intelligent identification. BACKGROUND

[0002] Chronic wounds such as diabetic foot ulcers, lower extremity venous ulcers and pressure injuries are common refractory wounds in clinical practice. Such wounds are often hindered in the healing process due to factors such as local blood circulation disorders, neuropathy or persistent pressure. The treatment cycle is long, often taking weeks or even months, during which the patient needs to frequently visit the doctor for dressing changes, bearing the dual pressure of physical and psychological stress; and is prone to cause serious complications such as infection, further threatening the life and health of the patient and seriously affecting the quality of life of the patient.

[0003] At present, the diagnosis of chronic wounds mainly relies on the personal experience of doctors. However, due to the differences between different doctors in clinical practice, knowledge reserve and judgment standard, the consistency of the assessment results is difficult to guarantee, which adversely affects the accuracy of chronic wound assessment, and therefore a chronic wound assessment and diagnosis system based on image intelligent identification is needed.

[0004] However, most of the current chronic wound assessment and diagnosis systems can only identify the wound boundary, and cannot analyze the spatial form, depth, area, volume, subcutaneous tissue, sinus and atrophic canal information and granulation activity of the wound, which has limitations for the diagnosis and treatment of some complex wounds, and reduces the accuracy of chronic wound assessment. SUMMARY

[0005] The application provides a chronic wound assessment and diagnosis system based on image intelligent identification, which aims to solve the problem that the prior art can only identify the wound boundary and cannot analyze the spatial form, depth, area, volume, subcutaneous tissue, sinus and atrophic canal information and granulation activity of the wound, which has limitations for the diagnosis and treatment of some complex wounds and reduces the accuracy of chronic wound assessment.

[0006] A chronic wound assessment and diagnosis system based on image intelligent identification, comprising an image acquisition module, a region identification module, a three-dimensional reconstruction module, an assessment module and a treatment scheme recommendation module.

[0007] The image acquisition module can obtain multi-source heterogeneous chronic wound image data through a unified interface protocol, and perform denoising and color standardization processing on the chronic wound image data; the image acquisition module comprises a standardization submodule and a form analysis submodule.

[0008] The region recognition module can segment the wound region from the processed chronic wound image data, generate an edge weight map, and classify the necrotic tissue and analyze the spatial morphology, depth, area, volume, subcutaneous tissue, sinus, and atrophy information through a U-Net3 network algorithm and an analysis model;

[0009] The three-dimensional reconstruction module can construct an interactive three-dimensional clinical model based on the reconstructed sinus direction analysis, clearly display the position and shape of the sinus, and mark the sinus information in combination with a clock position marker; the three-dimensional reconstruction module includes a construction submodule and a marker submodule.

[0010] The evaluation module can convert the RGB image to the HSV space, judge the capillary density index characteristics at different time nodes by quantifying the area proportion of the red and pink regions, and realize continuous monitoring of the granulation maturity.

[0011] The treatment scheme recommendation module can match the multi-source data of the necrotic tissue classification, sinus direction, and granulation maturity of the patient's wound with the standardized treatment scheme, determine and output the final diagnosis scheme.

[0012] Further, the standardization submodule is used to standardize the operation process of the smartphone digital camera through standardization instructions.

[0013] Further, the form analysis submodule can extract all the text content of the paper or electronic evaluation sheet of the doctor's diagnosis through the OCR technology, specially process the check boxes and underlines, and perform semantic analysis to convert the free text into structured data.

[0014] Further, the analysis model linearly transforms the RGB pixel value of the chronic wound image data to the XYZ intermediate color domain, then applies a nonlinear correction formula to map to the CIELAB space, separates the three independent channels of brightness, green-red axis, and blue-yellow axis, and obtains the Lab image; and based on the edge weight map and the Lab image, the K-means clustering is applied on the L channel of the Lab image to preliminarily mark the necrotic candidate area.

[0015] Further, the sinus direction analysis is realized based on the binocular vision technology, can collect 15 groups of checkerboard images at different angles according to camera calibration, then extract the corner points and solve the camera intrinsic parameters, distortion coefficients, and binocular extrinsic parameters by using OpenCV, and then align the left and right views by using stereo correction; then the SGBM algorithm is used to generate a disparity map, which is smoothed by WLS filtering and optimized by sub-pixel, and then the image disparity is converted into depth value Z based on the calibration parameters, and the conversion formula is as follows:

[0016]

[0017] where f xB is the baseline distance of binocular external parameters, and d is the parallax value of the pixel point.

[0018] Further, the sinus tract direction analysis further comprises extracting a three-dimensional coordinate of the wound area to generate a three-dimensional point cloud model in combination with an edge weight map, calculating a matrix C of the sinus tract direction according to the screened sinus tract boundary contour, and the calculation formula is as follows:

[0019]

[0020] wherein, is the point cloud centroid, (X k , Y k , Z k ) is the three-dimensional coordinate, P′ k is the centralized coordinate, and N is the total number of data points contained in the three-dimensional point cloud.

[0021] Further, the construction submodule can construct an interactive three-dimensional clinical model based on an energy minimization semi-global block matching strategy, establish a dense correspondence relationship between left and right views to generate a disparity map, convert two-dimensional pixel displacement values into millimeter-level depth data in combination with the calibrated baseline length and focal length parameters, and use a back projection algorithm to fit discrete point clouds into a continuous surface model.

[0022] Further, the marking submodule establishes a clock coordinate system with the wound geometric center as the origin, automatically labels the spatial direction of the sinus tract, and maps the depth value to a heat map.

[0023] Further, the evaluation module is constructed using a hybrid architecture of CNN and LSTM, which is used to realize the deep fusion of spatial feature extraction and time series modeling.

[0024] Further, the specific content of the region identification module further comprises hidden track and fistula detection and wound area and volume quantification.

[0025] The hidden track detection: taking the clock direction as the reference, the horizontal distance and depth between the wound edge and the healthy tissue attachment point are detected on the three-dimensional point cloud data, and when the horizontal distance is greater than 2mm, it is determined that there is a hidden track.

[0026] The fistula detection: searching for an abnormal channel connecting two independent wound areas or body surface openings in the three-dimensional point cloud model; after identification, it is marked as a fistula, and the tunnel length, path direction and opening diameter are quantified.

[0027] The wound area and volume are quantified: the volume estimation is realized in combination with a three-dimensional point cloud model, based on a segmentation mask and a standardized reference in the image, the two-dimensional projection area of the wound surface is obtained by counting the total number of pixels in the mask, the Alpha shape algorithm is used to automatically identify the complex contour boundary, and the interference of artifacts such as hair is excluded; based on the triangulation method, the irregular surface is decomposed into micro-element patches, and the three-dimensional volume data is accumulated.

[0028] Compared with the prior art, the present application has at least the following beneficial effects:

[0029] Based on further analysis and research on the problems of the prior art, the wound image of the patient is input through the image acquisition module, the wound area is accurately segmented through the deep learning network, and the wound features are intelligently identified in combination with the image, the accuracy and reliability of the evaluation are improved, and the dynamic three-dimensional point cloud model is further constructed by using the binocular vision principle, the spatial form, depth, area, volume, hidden path, sinus and fistula information of the wound are intuitively displayed, and the subsequent wound evaluation effect is improved;

[0030] The granulation activity is also observed through the evaluation module, so as to dynamically monitor the wound healing process, make the workflow of the whole system highly automated and intelligent, greatly reduce the demand for manual intervention, and improve the work efficiency and diagnostic accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 A module diagram of a chronic wound evaluation and diagnosis system based on image intelligent identification is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and embodiments.

[0033] As Figure 1 shown, the present application provides a chronic wound evaluation and diagnosis system based on image intelligent identification, which comprises an image acquisition module, a region identification module, a three-dimensional reconstruction module, an evaluation module and a treatment scheme recommendation module.

[0034] The image acquisition module can obtain multi-source heterogeneous chronic wound image data through a unified interface protocol, and perform denoising and color standardization processing on the chronic wound image data, enhance the contrast and clarity of the image, and provide a reliable basis for subsequent analysis, and the multi-source heterogeneous chronic wound image data includes image data shot by a smart phone, a digital camera and a professional medical imaging instrument.

[0035] The image acquisition module further includes a standardization submodule for regulating the operation process of the smartphone digital camera through standardization instructions. For example, the user is required to keep a fixed focal length and vertical angle for shooting. The standardization instructions can prompt the user to adjust the posture and lighting conditions in real time through the interactive interface, which can minimize the imaging differences caused by human factors.

[0036] In addition, the image acquisition module also supports metadata acquisition, obtaining device model and shooting timestamp, which is convenient for tracing the image source and correcting potential systematic bias during preprocessing, ensuring the universality and comparability of image data.

[0037] The denoising process first uses the median filtering algorithm to replace the center point value with the gray scale sorting in the pixel neighborhood to remove salt and pepper noise; then superimposes Gaussian filtering, which weights and averages the neighborhood pixels based on the two-dimensional normal distribution function to smooth the sudden peaks in high-frequency textures. Through the hybrid filtering scheme, both the wound boundary profile and the artifacts caused by sensor thermal noise or transmission errors can be preserved.

[0038] When the metadata acquisition obtains the device model as a smartphone, the image is processed with enhanced low-frequency component weight to solve the problem of uneven brightness caused by the high-sensitivity mode of the smartphone camera, to stabilize the overall brightness field.

[0039] Color standardization is used to correct the color difference caused by different devices. First, a standard reference area with known color features is selected in the image to construct a conversion model from device-dependent color space to standard color space. Then, a color profile matching algorithm is used to perform pixel-by-pixel color normalization on the wound image, ensuring that the same tissue type appears consistent in color under different device shooting, thereby eliminating color deviations caused by factors such as light source color temperature and camera white balance settings, laying a reliable color foundation for subsequent accurate wound region segmentation and tissue classification.

[0040] In addition, to ensure the accuracy of subsequent two-dimensional and three-dimensional measurements, the image acquisition module requires a standardized reference object (such as a ruler with specific color markings) to be placed next to the wound during shooting. This reference object will serve as an absolute basis for converting pixel size to physical units (millimeters) in subsequent modules, laying the foundation for accurate calculation of wound area and volume.

[0041] The image acquisition module also includes a form analysis submodule that can extract all the text content of the paper or electronic evaluation sheet of the doctor's diagnosis through OCR technology, while specially processing the check boxes (analyzing the filling state) and underlines (recognizing handwritten content). Semantic analysis is also performed to ensure that free text is converted into structured data, providing standard input for subsequent analysis, allowing the system to integrate multiple heterogeneous information sources for comprehensive analysis. At the same time, historical standardized treatment plans are input to assist in treatment plan recommendation.

[0042] The region recognition module can segment the wound area from the processed chronic wound image data through the U-Net3 network algorithm and analysis model, generate an edge weight map, and perform classification of necrotic tissue and analysis of spatial morphology, depth, area, volume, subcutaneous tissue, sinus, and atrophic pipe information. The classification of necrotic tissue includes four types: grayish yellow slough, attached yellow slough, soft black scab, and hard black scab. The specific content is as follows:

[0043] 1) Wound area segmentation

[0044] Taking the U-Net3 network algorithm as the skeleton, an EAM module is inserted at the encoder-decoder jump connection to form an analysis model, which contains two parallel branches:

[0045] Edge feature extraction branch: use the Sobel operator to calculate the gradient amplitude and generate an edge probability map (0-1).

[0046] Attention weight generation branch: map the edge probability map to attention weight (0-1) through 1x1 convolution, and multiply it with the feature map element by element. The final output is the attention weighted feature map, which enhances the semantic information of the wound boundary (such as immersion and edge rolling).

[0047] Collect 1200 labeled wound images (600 cases of diabetic foot ulcers, 400 cases of pressure ulcers, and 200 cases of venous ulcers), divide them into training, validation, and test sets in the ratio of 8:1:1, and train the analysis model. Provide the analysis model with a variety of samples that are clinically representative, ensuring that it can learn the common features of different wound types and accurately assess its generalization performance. After training, the U-Net3 network algorithm adaptively thresholds the processed chronic wound image data to generate an edge weight map.

[0048] 2) Necrotic tissue classification

[0049] The analysis model linearly transforms the RGB pixel values of the chronic wound image data to the XYZ intermediate color gamut, and then applies a non-linear correction formula to map it to the CIELAB space, separating the lightness (L), green-red axis (a), and blue-yellow axis (b*) into three independent channels to obtain the Lab image.

[0050] Based on the edge weight map and the Lab image, apply K-means clustering (k=3) on the L channel of the Lab image to preliminarily mark the low-light area (L<40) as a necrotic candidate area, excluding healthy tissue interference.

[0051] Based on the analysis model, all suspected necrotic areas are located, and multi-dimensional feature quantification is used to distinguish them:

[0052] In terms of texture smoothness, the contrast index was extracted using the gray level co-occurrence matrix (GLCM). Grayish yellow slough showed low contrast (<10) due to its uniform texture, while hard black scab showed high contrast (>50) due to its rough granular structure.

[0053] Edge sharpness was calculated by Canny detection to determine the proportion of boundary pixels. Adherent yellow slough showed low edge proportion (<0.2) due to infiltration and fusion with healthy tissue, while soft black scab showed high proportion (>0.5) due to its sharp boundary. Surface wetness was determined based on the mean value of the b channel (μ_b) in the Lab color space. Wet areas showed μ_b > 10 due to yellow pigment deposition, providing a reliable basis for the automated classification of the four types of necrotic tissue.

[0054] 3) Undermining and fistula detection

[0055] Undermining detection: From the wound edge to the healthy skin direction, the algorithm detects the horizontal distance and depth between the wound edge and the healthy tissue attachment point on the three-dimensional point cloud data. When the horizontal distance is >2mm, it is determined that there is undermining, and its position, horizontal extension distance (cm), and depth (cm) are recorded.

[0056] Fistula detection: In the three-dimensional point cloud model, search for abnormal channels connecting two independent wound areas or body surface openings. Once identified, it will be marked as a fistula, and its tunnel length (cm), path direction, and opening diameter will be quantified.

[0057] 4) Sinus tract analysis

[0058] Three-dimensional measurement of wound image data based on binocular vision technology: First, camera calibration is performed by collecting 15 sets of checkerboard images at different angles. The corner points are extracted using OpenCV, and the camera intrinsic parameters, distortion coefficients, and binocular extrinsic parameters are solved. Then, stereo rectification is used to align the left and right views. Next, the SGBM algorithm is used to generate a disparity map, which is smoothed by WLS filtering and optimized to sub-pixels. Based on the calibration parameters, the disparity is converted to depth value Z, and the conversion formula is as follows:

[0059]

[0060] where f is the focal length obtained by calibration, B is the baseline distance of the binocular extrinsic parameter, and d is the disparity value of the pixel point. x

[0061] Then, the three-dimensional coordinates of the wound area are extracted based on the edge weight map to generate a three-dimensional point cloud model. The area with a depth difference >2mm is detected as the depth difference threshold, which is used to select the sinus tract boundary profile. Then, the matrix C of the sinus tract direction is calculated to accurately determine the three-dimensional structure of the sinus tract. The calculation formula is as follows:

[0062] ​

[0063] wherein, is the point cloud centroid, (X k , Y k , Z k ) is the three-dimensional coordinate, P' is the centered coordinate, and N is the total number of data points contained in the three-dimensional point cloud. k

[0064] 5) Wound area and volume quantification

[0065] Volume estimation is achieved in combination with the three-dimensional point cloud model. Based on the segmentation mask and the standardized reference in the image, the actual area corresponding to a unit pixel is calculated, and the total number of pixels in the mask is counted to obtain the two-dimensional projection area of the wound surface. Alpha shape algorithm is used to automatically identify the complex contour boundary and exclude the interference of hair and other artifacts. Based on the triangular subdivision method, the irregular surface is decomposed into micro-element patches, and the three-dimensional volume data is accumulated. This index can dynamically reflect the wound contraction trend and provide quantitative basis for debridement range.

[0066] The three-dimensional reconstruction module can construct an interactive three-dimensional clinical model based on the analysis of the restored sinus direction, clearly display the position and shape of the sinus, and mark the sinus information in combination with the clock position marker, so as to facilitate the display through the visual mode and help the subsequent doctors to make more accurate surgical plans and nursing plans.

[0067] The three-dimensional reconstruction module includes a construction submodule and a marking submodule. The construction submodule is used to restore the sinus direction and construct an interactive three-dimensional clinical model, which clearly presents the position and shape of the sinus. The marking submodule can mark the sinus direction as a clock time according to the spatial coordinates of the sinus geometric body, taking the clock position as the reference. The specific content is as follows:

[0068] 1) Construction submodule

[0069] The interactive three-dimensional clinical model is constructed based on the energy minimization semi-global block matching strategy (SGBM), and a dense correspondence relationship is established between the left and right views to generate a disparity map. In combination with the calibrated baseline length and focal length parameters, the two-dimensional pixel displacement value is converted into millimeter-level depth data, and the reverse projection algorithm is used to fit the discrete point cloud into a continuous surface model, accurately restoring the concave-convex topological structure of the wound surface.

[0070] 2) Marking submodule

[0071] ​A clock coordinate system is established with the geometric center of the wound as the origin (12 o'clock always pointing to the patient's head side). The spatial direction of the sinus tract is automatically labeled (e.g., 3 o'clock direction), and the depth value is mapped to a heat map: red (depth > 2 cm), yellow (1-2 cm), blue (<1 cm). Key measurement points (e.g., maximum depth position) are marked on the grid surface. Rotation, scaling, and sectioning operations under the WebGL engine are also supported.

[0072] The evaluation module can convert RGB images to HSV space, judge the capillary density index characteristics at different time nodes by quantifying the area proportion of red and pink regions, and realize continuous monitoring of granulation maturity.

[0073] First, the RGB image is converted to HSV (hue, saturation, and brightness) space. By setting hue thresholds (e.g., 0-10° and 340-360° for red, 10-30° for pink), combining saturation and brightness to filter background interference, and quantifying the area proportion of red and pink pigmentation regions.

[0074] When the capillary blood vessels in the newly formed granulation tissue are abundant, the proportion of red regions increases significantly, while fibrotic or necrotic tissue shows a shift in hue or a decrease in saturation. The area proportion of red and pink pigmentation regions can directly reflect the capillary density.

[0075] At the same time, the evaluation module is built using a hybrid architecture of CNN and LSTM, which realizes the deep integration of spatial feature extraction and time series modeling. The CNN branch abstracts local lesion features through convolution and pooling layers to generate multi-channel feature maps. The LSTM branch concatenates the patient's historical visit data (e.g., weekly wound images) to construct a time series evolution curve of granulation maturity. The dual-channel design allows the model to grasp the spatial distribution characteristics of a single measurement and track the trend of tissue changes during treatment.

[0076] The evaluation module can continuously monitor the slope change of the granulation growth curve. Once it detects that the activity index of three consecutive observation points decreases at a rate exceeding the preset threshold (e.g., daily decay >5%), it immediately activates a multi-level warning system.

[0077] For example: the primary alarm prompts to strengthen local care observation, the intermediate warning suggests adjusting the type of dressing, and in emergency situations, it directly pushes the debridement surgery intervention plan.

[0078] The treatment recommendation module can match the multi-source data of the patient's wound necrotic tissue classification, sinus tract direction, and granulation maturity with standardized treatment plans, determine and output the final diagnosis plan, realize intelligent recommendation of treatment plans, and ensure the scientificity and individualization of treatment decisions.

[0079] In the above-mentioned chronic wound evaluation and diagnosis system based on image intelligent recognition, the image acquisition module is used to input the wound image of the patient, the deep learning network is used to accurately segment the wound area, and the image intelligent recognition is combined with the wound characteristics to improve the accuracy and reliability of the evaluation, and further use the binocular vision principle to construct a dynamic three-dimensional point cloud model to intuitively display the spatial form, depth, area, volume, hidden path, sinus and fistula information of the wound, thereby improving the subsequent wound evaluation effect;

[0080] The granulation activity is also observed through the evaluation module, so as to achieve dynamic monitoring of the wound healing process. The workflow of the whole system is highly automated and intelligent, greatly reducing the need for manual intervention, improving work efficiency and diagnosis accuracy.

[0081] The technical features of the above embodiments can be combined arbitrarily, and in order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictory, it should be considered as the scope of the present disclosure.

Claims

1. A chronic wound assessment and diagnosis system based on image intelligent recognition, characterized in that, It includes an image acquisition module, a region recognition module, a 3D reconstruction module, an evaluation module, and a treatment plan recommendation module; The image acquisition module can acquire multi-source heterogeneous chronic wound image data through a unified interface protocol, and perform noise reduction and color standardization processing on the chronic wound image data; the image acquisition module includes a standardization sub-module and a form parsing sub-module; The region identification module can segment wound regions from processed chronic wound image data using the U-Net3 network algorithm and analysis model, generate edge weight maps, and perform classification and reconstruction of necrotic tissue, including spatial morphology, depth, area, volume, undermining, sinus tracts, and fistula information analysis. The three-dimensional reconstruction module can construct an interactive three-dimensional clinical model based on the analysis of the restored sinus tract orientation, clearly displaying the location and shape of the sinus tract, and at the same time, combining clock orientation marks to indicate the sinus tract information; the three-dimensional reconstruction module includes a construction sub-module and a marking sub-module; The evaluation module can convert RGB images to HSV space, and determine the capillary density index characteristics at different time points by quantifying the area ratio of red and pink regions, thereby achieving continuous monitoring of granulation tissue maturity. The treatment plan recommendation module can match multi-source data such as the classification of necrotic tissue in the patient's wound, the direction of the sinus tract, and the maturity of granulation tissue with standardized treatment plans to determine and output the final diagnostic plan.

2. The chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The standardization submodule is used to standardize the operation process of smartphone digital cameras through standardized instructions.

3. The chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The form parsing submodule can extract all text content from the doctor's diagnosis paper or electronic assessment form using OCR technology. It also performs special processing on checkboxes and underlines, and performs semantic parsing to convert free text into structured data.

4. The chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The analysis model linearly transforms the RGB pixel values ​​of chronic wound image data to the XYZ intermediate color gamut, and then applies a nonlinear correction formula to map them to the CIELAB space, separating three independent channels: brightness, green-red axis, and blue-yellow axis, to obtain the Lab image. Based on the edge weight map and the Lab image, K-means clustering is applied to the L channel of the Lab image to initially label it as a necrosis candidate region.

5. The chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The sinus tract orientation analysis is based on binocular vision technology. It acquires 15 sets of checkerboard images from different angles according to camera calibration, then uses OpenCV to extract corner points and solve for camera intrinsic parameters, distortion coefficients, and binocular extrinsic parameters. Subsequently, stereo correction is used to align the left and right view rows. Next, the SGBM algorithm is used to generate a disparity map. After WLS filtering, smoothing, and sub-pixel optimization, the image disparity is converted into a depth value Z based on the calibration parameters. The conversion formula is as follows: Among them, f x The focal length obtained from the calibration is B, the baseline distance of the binocular extrinsic parameters is d, and the disparity value of the pixel is d.

6. The chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 5, characterized in that, The sinus tract orientation analysis also includes extracting the three-dimensional coordinates of the wound area by combining the edge weight map to generate a three-dimensional point cloud model. Based on the selected sinus tract boundary contour, the sinus tract orientation matrix C is calculated using the following formula: in, Let X be the centroid of the point cloud. k Y k Z k P' is a three-dimensional coordinate system. k The coordinates are centered, and N is the total number of data points in the 3D point cloud.

7. The chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The construction submodule can build an interactive 3D clinical model based on a semi-global block matching strategy with energy minimization, establish a dense correspondence between the left and right views to generate a disparity map, combine the calibrated baseline length and focal length parameters, convert the 2D pixel displacement values ​​into millimeter-level depth data, and use a back projection algorithm to fit the discrete point cloud into a continuous surface model.

8. The chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The marking submodule establishes a clock coordinate system with the geometric center of the wound as the origin, automatically marks the spatial direction of the sinus tract, and maps the depth value into a heat map.

9. A chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The evaluation module is constructed using a hybrid architecture of CNN and LSTM to achieve deep integration of spatial feature extraction and time series modeling.

10. A chronic wound assessment and diagnosis system based on image intelligent recognition according to claim 1, characterized in that, The specific content of the region identification module also includes burrowing and fistula detection and wound area and volume quantification; The stealth detection method involves using the clock face as a reference to probe from the edge of the wound toward the healthy skin; on the three-dimensional point cloud data, the horizontal distance and depth between the edge of the wound and the attachment point of the healthy tissue are detected, and when the horizontal distance is >2mm, it is determined that there is stealth. The fistula detection involves searching for abnormal channels connecting two independent wound areas or surface openings in a 3D point cloud model; once identified, these channels are marked as fistulas, and their tunnel length, path direction, and opening diameter are quantified. The wound area and volume quantization: Volume estimation is achieved by combining a three-dimensional point cloud model. Based on the segmentation mask and standardized reference objects in the image, the two-dimensional projected area of ​​the wound surface is obtained by counting the total number of pixels in the mask. The Alpha shape algorithm is used to automatically identify complex contour boundaries and eliminate interference from artifacts such as hair. The irregular surface is decomposed into micro-element patches based on the triangulation method, and the three-dimensional volume data is obtained by accumulating them.

Images