A portable wound intelligent measurement method and system based on Hough circle transformation and deep network fusion

Abstract

This invention discloses a portable intelligent wound measurement method and system based on the fusion of Hough circle transform and deep network, belonging to the field of artificial intelligence medical image processing. The method includes the following steps: acquiring wound images containing circular calibration objects and ensuring standardized image capture through a pose awareness mechanism; performing feature enhancement and preprocessing on the images; anchoring the calibration objects using an adaptive parameter Hough circle transform algorithm and obtaining pixel radii to establish a spatial scale; inputting the images into a pre-trained U-Net++ deep semantic segmentation network, extracting features using dense skip connections, and outputting a binary segmentation mask of the wound surface; finally, retrieving the true area and size based on the mask features and the scale. This invention uses a deep network to extract the wound surface and combines it with Hough circle transformation for scale fixation and pixel calculation, effectively improving the segmentation accuracy and measurement accuracy of complex wounds, achieving automated quantification of wound size and area, reducing manual measurement errors, and improving the efficiency of wound assessment and management.

Description

Technical Field

[0001] This invention relates to an automatic wound measurement method and system based on image processing and deep networks, belonging to the field of artificial intelligence medical image processing technology. Background Technology

[0002] The diagnosis of wound areas in burns and various chronic ulcers (such as diabetic foot and pressure ulcers) is a crucial step in injury assessment and treatment planning. Reliable and effective wound severity assessment is the foundation of clinical decision-making. Accurate measurement and tracking of patient wounds are particularly important, playing a key role in ensuring patients receive appropriate treatment and care. However, current wound measurement methods still have many limitations.

[0003] Currently, there are two main methods for wound measurement: manual measurement (such as using a ruler or transparent tracing film), which is complex and highly dependent on the doctor's skill and experience, often resulting in significant errors; and measurement using 3D modeling or dedicated structured light scanners, which offers higher accuracy but is expensive, lacks portability, and is unsuitable for home care or high-frequency bedside monitoring. Furthermore, some traditional computer vision measurement software based on smartphones often uses simple algorithms such as color thresholding or Canny edge detection to extract wounds, which are highly susceptible to interference from ambient lighting and complex skin textures, leading to severely fragmented wound extraction and highly unstable computational accuracy. Therefore, we urgently need an automated measurement method that can guarantee high-precision semantic segmentation while enabling convenient scale calibration and rapid measurement using mobile devices. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a portable intelligent wound measurement method and system based on the fusion of Hough circle transform and deep network. This method can accurately fix the scale using traditional algorithms and automatically identify complex wound areas by integrating a deep network, ensuring the accuracy of the calculation of the actual wound size.

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

[0006] In a first aspect, embodiments of the present invention provide a portable intelligent wound measurement method based on the fusion of Hough circular transform and deep networks, comprising the following steps:

[0007] The system collects original color images containing a standard circular calibration object and the wound, and ensures the standardization of the shooting posture through visual guidance and posture perception mechanisms.

[0008] The original color image is preprocessed with feature enhancement and normalization to obtain a standardized image suitable for model input;

[0009] The Hough circle transform algorithm with an adaptive parameter mechanism is used to perform a global scan of the processed image to anchor the circular calibration object and obtain the center coordinates and pixel radius of the effective calibration object.

[0010] The processed image is input into the pre-trained U-Net++ deep semantic segmentation network for multi-scale feature extraction and feature fusion, accurately locating the wound and outputting a pixel-level binary segmentation mask of the wound area.

[0011] A spatial mapping scale is established based on the pixel radius and actual physical size of the effective calibrator. Combined with the geometric features of the binarized segmentation mask, the actual physical size and area of ​​the wound are calculated.

[0012] As one possible implementation of this embodiment, the acquisition includes original color images of a standard circular calibration object and the wound, and ensures the standardization of the shooting pose through visual guidance and posture perception mechanisms, including:

[0013] Overlay a vertical auxiliary grid onto the mobile terminal's shooting interface;

[0014] Read data from the mobile terminal's built-in gyroscope to determine the angle between the lens's main optical axis and the plane where the wound is located. If the tilt angle is less than a preset threshold, shooting is allowed; otherwise, a real-time pose correction prompt is triggered on the interface.

[0015] As one possible implementation of this embodiment, the feature enhancement and normalization preprocessing of the original color image includes:

[0016] The image is converted into a grayscale matrix and then smoothed and denoised using an adaptive Gaussian kernel.

[0017] Adaptive histogram equalization is applied to the denoised image to enhance the contrast between the skin background and the wound tissue;

[0018] The image is cropped or scaled to a fixed resolution to meet the input dimension requirements of the deep learning network, and the pixel values ​​are normalized.

[0019] As one possible implementation of this embodiment, the method of using the Hough circle transform algorithm with an adaptive parameter mechanism to obtain the center coordinates and pixel radius of the effective calibration object includes:

[0020] The potential center of a circle is locked by accumulating gradients along the edge gradient direction in a two-dimensional plane using the Hough gradient method.

[0021] Dynamically adjust the accumulator threshold based on the overall image contrast.

[0022] Set edge continuity index and radius error tolerance, filter pseudo-circle interference, and output the pixel radius of the only valid calibration object.

[0023] As one possible implementation of this embodiment, the U-Net++ deep semantic segmentation network adopts nested and dense skip connection paths to fuse semantic feature maps of different depths in the encoder with the corresponding feature maps in the decoder; the network output outputs a probability map with the same resolution as the input image through the Sigmoid activation function, and after thresholding, generates a binary segmentation mask that only contains the wound target region and the background region.

[0024] As one possible implementation of this embodiment, the step of establishing a spatial mapping scale based on the pixel radius and actual physical size of the effective calibrator, and calculating the true physical size and area of ​​the wound in conjunction with the geometric features of the binarized segmentation mask, includes:

[0025] Calculate the fixed scale factor based on the actual physical diameter and pixel radius of the calibration object;

[0026] The total number of pixels in the target region of the wound in the binarized segmentation mask is counted, and the major axis pixels and minor axis pixels of the bounding rectangle of the wound contour are obtained.

[0027] Multiply the length and width pixel values ​​of the bounding rectangle of the wound by the scale factor to obtain the true length and true width of the wound.

[0028] The true area of ​​the wound is obtained by multiplying the total number of pixels in the target area by the square of the scale factor.

[0029] Secondly, the present invention provides a portable intelligent wound measurement system based on the fusion of Hough circular transform and deep network, comprising:

[0030] The standardized acquisition module is used to acquire original color images containing a standard circular calibration object and the wound, and ensures the standardization of the shooting posture through visual guidance and posture perception mechanisms.

[0031] The image preprocessing module is used to perform feature enhancement and normalization preprocessing on the original color image;

[0032] The scale anchoring module is used to obtain the center coordinates and pixel radius of the effective calibrator by employing the Hough circle transform algorithm with an adaptive parameter mechanism.

[0033] The wound depth segmentation module is used to extract features from the image using a pre-trained U-Net++ deep semantic segmentation network and output a pixel-level binary segmentation mask for the wound area.

[0034] The true size calculation module is used to calculate the true physical size of the wound based on the spatial mapping scale established by the pixel radius and the geometric features of the mask.

[0035] The technical solutions of the embodiments of the present invention can have the following beneficial effects:

[0036] This invention introduces an adaptive parameter Hough circle algorithm based on the Hough gradient method, achieving a highly robust automatic fixation function for scale ratios and significantly reducing computational costs. Simultaneously, it innovatively combines the U-Net++ deep semantic segmentation network to extract wound surfaces, bridging the semantic gap through dense skip connections, and accurately capturing wound boundaries in complex environments.

[0037] Using the portable wound measurement method of the present invention, the patient's wound only needs to be photographed through a smart terminal, without the need for expensive 3D equipment. Through automated preprocessing, mask generation and pixel counting calculation, the accuracy of the actual wound size calculation is greatly improved, providing strong support for improving the efficiency of diagnosis and treatment and remote monitoring. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating a portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network, according to an exemplary embodiment.

[0039] Figure 2 This is a schematic diagram illustrating a portable intelligent wound measurement system based on the fusion of Hough circular transform and deep network, according to an exemplary embodiment.

[0040] Figure 3 This is a block diagram illustrating a portable intelligent wound measurement system based on the fusion of Hough circular transform and deep network, according to an exemplary embodiment. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments: To clearly illustrate the technical features of this solution, the present invention will be described in detail below through specific embodiments and with reference to the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. The present invention omits descriptions of well-known components, processing techniques, and processes to avoid unnecessarily limiting the present invention.

[0042] A portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network, such as Figure 1 As shown, it includes the following steps:

[0043] The system collects original color images containing a standard circular calibration object and the wound, and ensures the standardization of the shooting posture through visual guidance and posture perception mechanisms.

[0044] The original color image is preprocessed with feature enhancement and normalization to obtain a standardized image suitable for model input;

[0045] The Hough circle transform algorithm with an adaptive parameter mechanism is used to perform a global scan of the processed image to anchor the circular calibration object and obtain the center coordinates and pixel radius of the effective calibration object.

[0046] The processed image is input into the pre-trained U-Net++ deep semantic segmentation network for multi-scale feature extraction and feature fusion, accurately locating the wound and outputting a pixel-level binary segmentation mask of the wound area.

[0047] like Figure 2 As shown in the figure, an embodiment of the present invention provides a portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network, which includes the following steps:

[0048] Step 1: Collect original color images containing a standard circular calibration object and the wound, and ensure the standardization of the shooting pose through visual guidance and posture perception mechanisms.

[0049] Before measurement, the operator places a calibration object (standard circular patch) with a known diameter next to the wound. In this embodiment, a standard circular calibration patch with an actual physical diameter of 2.0 cm is used. When using a mobile camera to capture images, to avoid area calculation errors caused by perspective distortion, the system embeds a posture perception module: it reads data from the mobile terminal's built-in gyroscope, determines the angle between the lens's principal optical axis and the plane where the wound is located, and allows shooting if the tilt angle is less than a preset threshold (e.g., 5 degrees); otherwise, it triggers a real-time posture correction prompt on the interface.

[0050] Step 2: Perform feature enhancement and normalization preprocessing on the original color image to obtain a standardized image suitable for model input.

[0051] After acquiring images, preprocessing is required.

[0052] First, the image is converted into a grayscale matrix, and then an adaptive Gaussian kernel is used for smoothing and noise reduction.

[0053] Adaptive histogram equalization was then performed to enhance the contrast between the skin background and the wound tissue, and to remove interference from uneven ambient lighting.

[0054] Finally, the image is cropped or scaled to a fixed resolution to meet the input dimension requirements of the subsequent U-Net++ deep learning network, and pixel values ​​are normalized.

[0055] Step 3: Using the Hough circle transform algorithm with an adaptive parameter mechanism, the processed image is scanned globally to anchor the circular calibration object, and the center coordinates and pixel radius of the effective calibration object are obtained.

[0056] Traditional 3D parameter space Hough transform computations are computationally intensive. This invention utilizes the Hough gradient method to accumulate and lock potential circle centers along the edge gradient direction in a 2D plane. To adapt to images of varying resolutions, the algorithm dynamically adjusts the accumulator threshold based on the overall image contrast; it sets edge continuity indices and radius error tolerances, filtering out pseudo-circle interference such as overlapping circles, thereby stably outputting the pixel radius of the unique valid calibration object even in complex backgrounds. The specific calculation process and core formulas are as follows:

[0057] First, the edge matrix of the image is extracted using the Canny operator. For any non-zero pixel on the edge... The Sobel operator is used to solve for its horizontal direction. and vertical direction The gradient components are then used to determine the gradient magnitude of the pixel at that point. With gradient direction angle :

[0058]

[0059]

[0060] Suppose a point is detected on the left edge of the calibration patch, and its image coordinates are... The derivative of the operator is obtained as follows: Substituting into the formula, the amplitude is calculated. gradient direction angle (That is, horizontally to the right, pointing towards the center of the circle).

[0061] Secondly, based on the geometric properties of a circle, the normal to any point on the circumference (i.e., the image gradient direction) must pass through the center of the circle. The system ensures that all edge points... Along its own gradient direction At the preset prior distance Within the specified range, spatial ray voting is performed on the two-dimensional accumulator array. Candidate center point. Satisfies the following parametric equations:

[0062]

[0063] After smoothing the two-dimensional accumulator, the system finds the global extreme point with the highest number of votes, thereby locking in the unique candidate center.

[0064] Using the above coordinate points and direction angle When the algorithm iterates to the distance variable When the pixel is used, the point will be the coordinate. , Cast a vote with a weight of 1. The accumulator performs global statistics and finds the coordinates. The point where the most votes converge is used to pinpoint the center coordinates. .

[0065] Finally, the center of the circle is located, and the system extracts a set of all edge pixels within the target area. Calculate their distance to the center of the circle. Euclidean distance :

[0066]

[0067] The system for all A one-dimensional frequency histogram is constructed, and the distance corresponding to the highest peak of the frequency distribution is taken as the pixel radius of the candidate calibrator. .

[0068] Furthermore, to eliminate interference from non-closed curves caused by factors such as water bubbles, edge continuity is introduced. Validation model:

[0069]

[0070] in, This represents the number of valid edge pixels contained at the peak of the histogram. When the value exceeds the threshold set by the system, the calibration is deemed valid and the final parameters are output.

[0071] Statistical center Calculate the distance to the surrounding edge points. The pixel had the highest frequency (total number of valid points counted). (Number). Substitute into the continuity formula to calculate: .because The verification passed. The center of the final stable output calibrator is... pixel radius .

[0072] Step 4: Input the processed image into the pre-trained U-Net++ deep semantic segmentation network to perform multi-scale feature extraction and feature fusion, accurately locate the wound, and output a pixel-level binary segmentation mask of the wound area.

[0073] For irregular and complex wounds, this embodiment employs the U-Net++ network. Compared to traditional CV methods such as Canny edge detection, this network seamlessly integrates shallow edge spatial features in the encoder with deep abstract semantic features in the decoder through its unique nested and dense skip connection paths, greatly improving the recognition accuracy for complex wound boundaries such as necrotic tissue and exudate. The network output uses a sigmoid activation function to output a probability map, which, after thresholding, generates a binary segmentation mask containing only the wound target area (pixel value of 1) and the background area (pixel value of 0). In this embodiment, through network inference and calculation, the target area with a pixel value of 1 in the generated binary mask is counted, revealing that the patient's wound area contains a total of 18,500 pixels.

[0074] Step 5: Establish a spatial mapping scale based on the pixel radius and actual physical size of the effective calibration object, and calculate the actual physical size and area of ​​the wound by combining the geometric features of the binarized segmentation mask.

[0075] After accurately segmenting and locating the target, the physical length represented by each pixel is calculated based on the actual physical diameter and pixel radius of the calibration object. The total number of pixels in the target region within the binarized segmentation mask is then counted. Multiplying this total number of pixels by the square of the scale factor automatically calculates the true physical area of ​​the irregular wound. Similarly, the true length and width dimensions of the circumscribed rectangle can be obtained. In a specific embodiment, based on the actual diameter of 2.0 cm and pixel radius of 50 pixels from steps one and three, the spatial mapping scale is calculated. Combining the total number of wound pixels (18500) counted in step four, the actual physical area of ​​the irregular wound is calculated using the formula. Similarly, if the system selects a rectangle whose length and width of the wound mask are 200 pixels and 150 pixels respectively, then multiply by the scale. Afterwards, the actual maximum length and width dimensions of the wound can be obtained as follows: and .

[0076] like Figure 3 As shown in the diagram, a block diagram of a portable intelligent wound measurement system based on the fusion of Hough circular transform and deep network provided by an embodiment of the present invention includes:

[0077] A standardized acquisition module is used to acquire images and ensure the standardization of shooting pose;

[0078] The image preprocessing module is used to perform feature enhancement and normalization preprocessing on the original color image;

[0079] The scale anchoring module is used to obtain the center coordinates and pixel radius of the effective calibrator by employing the Hough circle transform algorithm with an adaptive parameter mechanism.

[0080] The wound depth segmentation module is used to extract features from the image using a pre-trained U-Net++ deep network and output a binarized segmentation mask.

[0081] The true size calculation module is used to calculate the true physical size of the wound based on the spatial mapping scale established by the pixel radius and the geometric features of the mask.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network, characterized in that, Includes the following steps: The system collects original color images containing a standard circular calibration object and the wound, and ensures the standardization of the shooting posture through visual guidance and posture perception mechanisms. The original color image is preprocessed with feature enhancement and normalization to obtain a standardized image suitable for model input; The Hough circle transform algorithm with an adaptive parameter mechanism is used to perform a global scan of the processed image to anchor the circular calibration object and obtain the center coordinates and pixel radius of the effective calibration object. The processed image is input into the pre-trained U-Net++ deep semantic segmentation network for multi-scale feature extraction and feature fusion, accurately locating the wound and outputting a pixel-level binary segmentation mask of the wound area. A spatial mapping scale is established based on the pixel radius and actual physical size of the effective calibrator. Combined with the geometric features of the binarized segmentation mask, the actual physical size and area of ​​the wound are calculated.

2. The portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network as described in claim 1, characterized in that, The acquisition includes original color images of a standard circular calibration object and the wound, and standardization of the shooting posture is ensured through visual guidance and posture perception mechanisms, including: Overlay a vertical auxiliary grid onto the mobile terminal's shooting interface; Read data from the mobile terminal's built-in gyroscope to determine the angle between the lens's main optical axis and the plane where the wound is located. If the tilt angle is less than a preset threshold, shooting is allowed; otherwise, a real-time pose correction prompt is triggered on the interface.

3. The portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network as described in claim 1, characterized in that, The feature enhancement and normalization preprocessing of the original color image includes: The image is converted into a grayscale matrix and then smoothed and denoised using an adaptive Gaussian kernel. Adaptive histogram equalization is applied to the denoised image to enhance the contrast between the skin background and the wound tissue; The image is cropped or scaled to a fixed resolution to meet the input dimension requirements of the deep learning network, and the pixel values ​​are normalized.

4. The portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network as described in claim 1, characterized in that, The Hough circle transform algorithm, which incorporates an adaptive parameter mechanism, performs a global scan of the processed image to anchor the circular calibration object, obtaining the center coordinates and pixel radius of the effective calibration object, including: The potential center of a circle is locked by accumulating gradients along the edge gradient direction in a two-dimensional plane using the Hough gradient method. Dynamically adjust the accumulator threshold based on the overall image contrast. Set edge continuity index and radius error tolerance, filter pseudo-circle interference, and output the pixel radius of the only valid calibration object.

5. The portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network as described in claim 1, characterized in that, The U-Net++ deep semantic segmentation network employs nested and dense skip connection paths to fuse semantic feature maps of different depths in the encoder with corresponding feature maps in the decoder; the network output is processed by a binarized segmentation mask of the wound target region and the background region.

6. The portable intelligent wound measurement method based on the fusion of Hough circular transform and deep network as described in claim 1, characterized in that, The step of establishing a spatial mapping scale based on the pixel radius and actual physical size of the effective calibrator, and calculating the true physical size and area of ​​the wound by combining the geometric features of the binarized segmentation mask, includes: Calculate the fixed scale factor based on the actual physical diameter and pixel radius of the calibration object; The total number of pixels in the target region of the wound in the binarized segmentation mask is counted, and the major axis pixels and minor axis pixels of the bounding rectangle of the wound contour are obtained. Multiply the length and width pixel values ​​of the bounding rectangle of the wound by the scale factor to obtain the true length and width of the wound; multiply the total number of pixels in the target area by the square of the scale factor to obtain the true area of ​​the wound.

7. A portable intelligent wound measurement system based on the fusion of Hough circular transform and deep network, characterized in that, include: The standardized acquisition module is used to acquire original color images containing a standard circular calibration object and the wound, and ensures the standardization of the shooting posture through visual guidance and posture perception mechanisms. The image preprocessing module is used to perform feature enhancement and normalization preprocessing on the original color image; The scale anchoring module is used to obtain the center coordinates and pixel radius of the effective calibrator by employing the Hough circle transform algorithm with an adaptive parameter mechanism. The wound depth segmentation module is used to extract features from the image using a pre-trained U-Net++ deep semantic segmentation network and output a pixel-level binary segmentation mask for the wound area. The true size calculation module is used to calculate the true physical size of the wound based on the spatial mapping scale established by the pixel radius and the geometric features of the mask.

Images