Pressure ulcer auxiliary detection method and system for orthopedic clinical nursing

Abstract

The present invention relates to the field of image enhancement technology, and specifically to a pressure sore auxiliary detection method and system for orthopedic clinical care, the method comprising: obtaining a target pressure sore image corresponding to a pressure sore area to be detected; based on the target pressure sore image, obtaining the pressure sore grade corresponding to the pressure sore area to be detected through a pressure sore grade recognition network; expanding the original training set of the pressure sore grade recognition network comprises the following steps: obtaining pressure sore images representing different single pressure sore grades and different sample pressure sore images; determining the texture change evaluation corresponding to the pixel points; screening out target skeletons, standard images and specific areas, and enhancing each sample pressure sore image based on the specific areas; expanding the original training set according to the target enhanced image. The present invention realizes the expansion of the training set, improves the richness of the training set, thereby improving the pressure sore grade recognition ability of the pressure sore grade recognition network to the pressure sore grade, and further improves the accuracy of the pressure sore grade recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of image enhancement, and in particular to a pressure sore auxiliary detection method and system for orthopedic clinical nursing. Background Art

[0002] Orthopedic patients often need to stay in bed for a long time after surgery. Due to long-term local pressure, their skin and underlying tissues often suffer from ischemia or necrosis, resulting in different levels of pressure sores. Different levels of pressure sores often require different care plans. Therefore, timely identification of pressure sore levels is crucial for the rehabilitation of orthopedic patients. At present, the method commonly used for object level recognition is to use a trained neural network to perform object level recognition. Among them, the training set used for neural network training is often composed of historically captured object images.

[0003] However, when the training set of the neural network for pressure ulcer grade recognition is composed of historical images of pressure ulcer areas, the following technical problems often occur:

[0004] In actual situations, the severity of pressure sores in different locations within the pressure sore area may be different, which may result in a small number of features that do not belong to the pressure sore grade in some pressure sore areas. For example, there may be a small number of blisters in a pressure sore area that is generally erythematous. At this time, when the pressure sore grade of these pressure sore areas is identified through a neural network, the pressure sore grade may be misjudged due to the poor richness of the training set. For example, the pressure sore grade of a stage I pressure sore area may be misjudged as stage II due to the small number of blisters in the pressure sore area, resulting in poor accuracy in pressure sore grade identification. Summary of the invention

[0005] In order to solve the technical problem of poor accuracy in pressure sore grade identification, the present invention proposes a pressure sore auxiliary detection method and system for orthopedic clinical nursing.

[0006] In a first aspect, the present invention provides a pressure sore auxiliary detection method for orthopedic clinical nursing, the method comprising:

[0007] Acquire a target pressure sore image corresponding to the pressure sore area to be detected;

[0008] The target pressure sore image is input into the pre-trained pressure sore grade recognition network, and the pressure sore grade corresponding to the pressure sore area to be detected is obtained through the pressure sore grade recognition network;

[0009] During the training process of the pressure sore grade recognition network, the expansion of the original training set of the pressure sore grade recognition network includes the following steps:

[0010] Obtaining pressure ulcer images representing different single pressure ulcer levels as reference images, and obtaining different sample pressure ulcer images;

[0011] Determine the texture change evaluation corresponding to each pixel point according to the gradient value corresponding to each pixel point in each reference image and each sample pressure sore image and the grayscale condition in the corresponding preset neighborhood;

[0012] Based on the texture change evaluation corresponding to the pixels, the respective target skeletons are screened out from each reference image and each sample pressure ulcer image;

[0013] According to the characteristics of the target skeletons in the reference image and the sample pressure sore image, a standard image corresponding to each sample pressure sore image is screened out from all reference images;

[0014] According to the difference between each sample pressure sore image and its corresponding standard image, a specific region is screened out from each sample pressure sore image, and based on the specific region in each sample pressure sore image, each sample pressure sore image is enhanced to obtain a target enhanced image;

[0015] The original training set is augmented based on all target augmented images.

[0016] In combination with the first aspect above, in a possible implementation, determining the texture change evaluation corresponding to each pixel point according to the gradient value corresponding to each pixel point in each reference image and each sample pressure sore image and the grayscale condition in the corresponding preset neighborhood includes:

[0017] Determine any pixel point in any reference image or any sample pressure sore image as a marked pixel point;

[0018] Determine the grayscale change factor corresponding to the marked pixel point according to the maximum grayscale value and the minimum grayscale value in the preset neighborhood corresponding to the marked pixel point;

[0019] The texture change evaluation corresponding to the marked pixel point is determined according to the gradient value and the grayscale change factor corresponding to the marked pixel point, wherein both the gradient value and the grayscale change factor are positively correlated with the texture change evaluation.

[0020] In combination with the first aspect above, in a possible implementation, the step of selecting a respective target skeleton from each reference image and each sample pressure sore image based on texture change evaluation corresponding to pixel points includes:

[0021] Determine any reference image or any sample pressure ulcer image as a marked image;

[0022] Selecting a preset number of pixel points with the largest corresponding texture change evaluation from the marked image as target points for constituting the skeleton;

[0023] Select the target point with the largest corresponding texture change evaluation from all target points as the center point of the skeleton in the marked image, and determine each target point in the marked image except the center point of the skeleton as a skeleton fulcrum;

[0024] Connect the center point of the skeleton and each skeleton fulcrum in the marked image to obtain skeleton line segments;

[0025] Construct the target skeleton in the marked image with all the skeleton line segments in the marked image.

[0026] Combined with the first aspect above, in a possible implementation manner, the method of screening out the standard image corresponding to each sample pressure ulcer image from all reference images according to the characteristics of the target skeletons in the reference image and the sample pressure ulcer image includes:

[0027] Determine any one sample pressure ulcer image as a candidate image;

[0028] Determine the target similarity between the candidate image and each reference image according to the angles and gray levels corresponding to all the skeleton line segments in the candidate image and each reference image;

[0029] Screen out the reference image with the largest target similarity with the candidate image from all reference images as the standard image corresponding to the candidate image.

[0030] Combined with the first aspect above, in a possible implementation manner, the formula corresponding to the target similarity between the candidate image and the reference image is:

[0031] ;

[0032] Wherein, is the target similarity between the candidate image and the th reference image; is the serial number of the reference image; is the natural exponential function; is the number of skeleton line segments in the target skeleton; is the serial number of the skeleton line segment in the target skeleton; is the absolute value function; is the average value of the gray levels corresponding to all the pixel points on the th skeleton line segment in the target skeleton of the candidate image; is the th reference image, and is the average value of the gray levels corresponding to all the pixel points on the th skeleton line segment in the target skeleton; is the th skeleton line segment in the target skeleton of the candidate image; It is Among the target skeletons in the reference images, The angle corresponding to the skeleton line segments.

[0033] In combination with the first aspect above, in a possible implementation, screening out a specific area from each sample pressure sore image according to a difference between each sample pressure sore image and its corresponding standard image includes:

[0034] Determine the target specific evaluation corresponding to each pixel point in each sample pressure sore image according to the target similarity between each sample pressure sore image and its corresponding standard image, the texture change evaluation corresponding to each pixel point in each sample pressure sore image, and the grayscale difference between each pixel point in each sample pressure sore image and its corresponding standard image;

[0035] Pixels whose corresponding target specific evaluation is greater than a preset specific threshold are selected from each sample pressure sore image as specific pixels;

[0036] The area formed by all specific pixel points in each sample pressure sore image is determined as the specific area.

[0037] In combination with the first aspect above, in a possible implementation, the formula corresponding to the target-specific evaluation corresponding to the pixel points in the sample pressure sore image is:

[0038] ;

[0039] in, It is Sample pressure ulcer images Target-specific evaluation corresponding to pixels; is the serial number of the sample pressure ulcer image; It is The sequence number of the pixel points in the sample pressure sore image; is the normalization function; It is The target similarity between the sample pressure ulcer images and their corresponding standard images; It is Sample pressure ulcer images Texture change evaluation corresponding to each pixel; It is The number of pixels in the standard image corresponding to the sample pressure sore image; It is The sequence number of the pixel points in the standard image corresponding to the sample pressure sore image; It is the absolute value function; It is Sample pressure ulcer images The gray value corresponding to each pixel; It is In the standard image corresponding to the sample pressure sore image, The gray value corresponding to each pixel.

[0040] In combination with the first aspect above, in a possible implementation, enhancing each sample pressure sore image based on a specific region in each sample pressure sore image to obtain a target enhanced image includes:

[0041] Perform superpixel segmentation on each sample pressure sore image to obtain a superpixel block set corresponding to each sample pressure sore image;

[0042] From the superpixel block set corresponding to each sample pressure sore image, superpixel blocks with pixels belonging to specific areas are selected as candidate blocks;

[0043] Filter out the superpixel blocks adjacent to any candidate blocks from the superpixel block set corresponding to each sample pressure sore image as the target block;

[0044] Perform connected domain extraction on each sample pressure sore image to obtain the target connected domain;

[0045] In each sample pressure sore image, the area formed by all target connected domains where all candidate blocks are located and all target connected domains where all target blocks are located is enhanced to obtain a target enhanced image.

[0046] In combination with the first aspect above, in a possible implementation manner, the step of expanding the original training set according to all target enhanced images includes:

[0047] Perform image rotation on each target enhanced image to obtain a target rotated image;

[0048] All target augmented images and all target rotated images are added to the original training set.

[0049] In a second aspect, the present invention provides a pressure sore auxiliary detection system for orthopedic clinical care, the system comprising:

[0050] An image acquisition module, used to acquire a target pressure sore image corresponding to the pressure sore area to be detected;

[0051] The pressure sore grade recognition module is used to input the target pressure sore image into a pre-trained pressure sore grade recognition network, and obtain the pressure sore grade corresponding to the pressure sore area to be detected through the pressure sore grade recognition network.

[0052] In a third aspect, a server is provided, comprising a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes the method in the first aspect or any possible implementation of the first aspect.

[0053] In a fourth aspect, a computer program product is provided, comprising: a computer program code, which, when executed on a computer, enables the computer to execute the method in the first aspect or any possible implementation of the first aspect.

[0054] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program code. When the computer program code runs on a computer, the computer executes the method in the above-mentioned first aspect or any possible implementation manner of the first aspect.

[0055] The present invention has the following beneficial effects:

[0056] The pressure sore auxiliary detection method for orthopedic clinical care of the present invention realizes the expansion of the training set, solves the technical problem of poor accuracy of pressure sore grade recognition, improves the richness of the training set, thereby improving the pressure sore grade recognition ability of the pressure sore grade recognition network, and further improves the accuracy of pressure sore grade recognition. The present invention comprehensively considers multiple factors related to non-self-pressure sore grade features, such as texture change evaluation, and the difference between the sample pressure sore image and its corresponding standard image, thereby quantifying the specific area in the sample pressure sore image that characterizes the area to which the non-self-pressure sore grade features belong, and based on the specific area in each sample pressure sore image, each sample pressure sore image is enhanced, thereby enhancing the non-self-pressure sore grade feature information in the sample pressure sore image, so that the pressure sore grade recognition network can learn the non-self-pressure sore grade feature information in the pressure sore area, which can improve the accuracy and robustness of the network, thereby reducing the misjudgment of the pressure sore grade to a certain extent, and further improve the accuracy of pressure sore grade recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 It is a flow chart of the pressure sore auxiliary detection method for orthopedic clinical nursing of the present invention;

[0059] Figure 2 A flowchart of expanding the original training set of the pressure sore grade recognition network of the present invention;

[0060] Figure 3 It is a schematic diagram of the composition structure of the pressure sore auxiliary detection system for orthopedic clinical nursing of the present invention;

[0061] Figure 4 The figure is a schematic diagram of the structure of a computer device of the present invention. DETAILED DESCRIPTION

[0062] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0063] Unless defined otherwise, 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 belongs.

[0064] refer to Figure 1 , showing the process of some embodiments of the pressure sore auxiliary detection method for orthopedic clinical care of the present invention. The pressure sore auxiliary detection method for orthopedic clinical care includes the following steps:

[0065] Step S1, obtaining a target pressure sore image corresponding to the pressure sore area to be detected.

[0066] The pressure sore area to be detected may be a pressure sore area to be tested for pressure sore level. Pressure sore levels are mainly divided into six levels, which may be: stage I, stage II, stage III, stage IV, unstageable, and deep pressure sore level. Stage I, also known as the erythema stage, is characterized by an intact skin surface but the appearance of erythema. Stage II is partial dermal loss, that is, the surface layer or part of the skin is lost, and blisters or superficial ulcers appear. Stage III is full-thickness skin loss, that is, all layers of the skin are damaged, forming a deep ulcer, but no muscle or bone is involved. Stage IV is full-thickness tissue loss, that is, the damage is deep to the muscle, bone or supporting tissue, and necrotic tissue may appear. The degree of damage of unstageable pressure sores is difficult to determine, and they are usually covered by necrotic tissue or hematoma, and the bottom tissue cannot be observed. The deep pressure sore level is characterized by an intact skin surface, but the deep tissue may be damaged, which is manifested as a local dark or purple-red area, and may develop into a more serious ulcer. The target pressure sore image may be an image of the pressure sore area to be detected, that is, the target pressure sore image can represent the pressure sore area to be detected.

[0067] As an example, the surface image of the pressure ulcer area to be detected can be collected by a camera as the target surface image, the target surface image is grayscaled, and the grayscaled target surface image is used as the target pressure ulcer image.

[0068] It should be noted that the camera used in the embodiments of the present invention can be a high-resolution camera.

[0069] Step S2: Input the target pressure ulcer image into a pre-trained pressure ulcer grade recognition network, and through the pressure ulcer grade recognition network, obtain the pressure ulcer grade corresponding to the pressure ulcer area to be detected.

[0070] Among them, the pressure ulcer grade recognition network can be a CNN (Convolutional Neural Networks) used for pressure ulcer grade recognition.

[0071] Reference Figure 2 , shows the process of expanding the original training set of the pressure ulcer grade recognition network during the training process of the present invention. Among them, the images in the original training set can be historical pressure ulcer area images collected. Each pressure ulcer area image can represent a pressure ulcer area. For example, the original training set can be a publicly available dataset of images with pressure ulcers at different stages that have been labeled. The process of expanding the original training set of the pressure ulcer grade recognition network can include the following steps:

[0072] Step 201: Obtain pressure ulcer images representing different single pressure ulcer grades as reference images, and obtain different sample pressure ulcer images.

[0073] Among them, the pressure ulcer image of a single pressure ulcer grade can represent a pressure ulcer area that only contains the characteristics of a single pressure ulcer grade. For example, if the condition of a certain pressure ulcer area only conforms to the characteristics of stage I, that is, only erythema appears in the pressure ulcer area, then the pressure ulcer area is a pressure ulcer area that only contains the characteristics of a single pressure ulcer grade, and can be recorded as a pressure ulcer area that only contains a single stage I. Similarly, if a certain pressure ulcer area only contains the characteristics of stage II, stage III, stage IV, unstageable or deep pressure ulcer grades, then the pressure ulcer area is also a pressure ulcer area that only contains the characteristics of a single pressure ulcer grade. If both erythema and blisters appear in a certain pressure ulcer area, that is, the pressure ulcer area has both the characteristics of stage I and stage II, then the pressure ulcer area contains more than one pressure ulcer grade characteristic and does not meet the requirements for obtaining reference images. The sample pressure ulcer image can be a pressure ulcer area image that contains more than one pressure ulcer grade characteristic.

[0074] As an example, this step can include the following steps:

[0075] The first step: Obtain a pressure ulcer image representing a single pressure ulcer grade as a reference image.

[0076] For example, a camera may be used to capture at least one surface image representing a single stage I pressure sore area, at least one surface image representing a single stage II pressure sore area, at least one surface image representing a single stage III pressure sore area, at least one surface image representing a single stage IV pressure sore area, at least one surface image representing a single non-stageable pressure sore area, and at least one surface image representing a single deep pressure sore area, and all of these images captured at this time are recorded as first temporary images. Next, each first temporary image is rotated, and the rotated first temporary image is recorded as a second temporary image. Finally, all first temporary images and all second temporary images are collectively referred to as reference images.

[0077] The second step is to obtain different sample pressure ulcer images.

[0078] For example, a camera can be used to capture at least one surface image of a pressure sore area with characteristics of stage I and stage II, at least one surface image of a pressure sore area with characteristics of stage II and stage III, at least one surface image of a pressure sore area with characteristics of stage III and stage IV, at least one surface image of a pressure sore area with characteristics of stage IV and unstageable, and at least one surface image of a pressure sore area with characteristics of unstageable and deep pressure sore grades, and all of these images captured at this time are recorded as sample pressure sore images.

[0079] Step 202: Determine a texture change evaluation corresponding to each pixel point according to the gradient value corresponding to each pixel point in each reference image and each sample pressure sore image and the grayscale condition in its corresponding preset neighborhood.

[0080] The gradient value is also called the gradient size. The preset neighborhood may be a preset rectangular neighborhood. For example, the preset neighborhood may be a 5×5 neighborhood. The pixel point may be located at the center of its preset neighborhood.

[0081] As an example, this step may include the following steps:

[0082] In the first step, any pixel point in any reference image or any sample pressure ulcer image is determined as a marked pixel point.

[0083] In the second step, the grayscale change factor corresponding to the marked pixel point is determined according to the maximum grayscale value and the minimum grayscale value in the preset neighborhood corresponding to the marked pixel point.

[0084] The third step is to determine the texture change evaluation corresponding to the marked pixel point according to the gradient value and grayscale change factor corresponding to the marked pixel point.

[0085] Among them, both the gradient value and the grayscale change factor can be positively correlated with the texture change evaluation.

[0086] For example, the formula for determining the texture change evaluation corresponding to the marked pixel point can be:

[0087] ;

[0088] in, It is the texture change evaluation corresponding to the marked pixel. is a normalization function. is the gradient value corresponding to the marked pixel. It is the maximum grayscale value in the preset neighborhood corresponding to the marked pixel, that is, the maximum value of the grayscale values ​​corresponding to all pixels in the preset neighborhood corresponding to the marked pixel. It is the minimum grayscale value in the preset neighborhood corresponding to the marked pixel, that is, the minimum value among the grayscale values ​​corresponding to all pixels in the preset neighborhood corresponding to the marked pixel. is the grayscale change factor corresponding to the marked pixel.

[0089] It should be noted that when The larger the value is, the greater the grayscale change at the marked pixel point is. The larger the value is, the greater the grayscale change in the preset neighborhood corresponding to the marked pixel point is. The larger the value is, the greater the grayscale change around the marked pixel is, and the more obvious the texture change around the marked pixel is.

[0090] Step 203 , based on the texture change evaluation corresponding to the pixel points, select the respective target skeletons from each reference image and each sample pressure sore image.

[0091] As an example, this step may include the following steps:

[0092] In the first step, any reference image or any sample pressure ulcer image is determined as a marked image.

[0093] In the second step, a preset number of pixel points with the largest texture change evaluation are selected from the above-mentioned marked image as target points for constituting the skeleton.

[0094] The preset number may be a preset number of pixels, for example, 50.

[0095] For example, if the preset number is 50, 50 pixel points with the largest texture change evaluation in the marker image may be used as target points.

[0096] In the third step, the target point with the largest texture change evaluation is selected from all the target points as the skeleton center point in the above-mentioned marked image, and each target point in the above-mentioned marked image except the skeleton center point is determined as a skeleton support point.

[0097] The fourth step is to connect the skeleton center point and each skeleton pivot point in the above marked image to obtain the skeleton line segment.

[0098] The skeleton line segment may be a line segment with a skeleton center point and a skeleton support point as endpoints.

[0099] The fifth step is to use all the skeleton line segments in the above labeled image to form the target skeleton in the above labeled image.

[0100] It should be noted that the pixels with large texture changes in the image can often represent the parts of the image that have undergone mutations, and the skeleton of the image is often composed of multiple points with drastic mutations. Therefore, the target skeleton in the marked image can represent the skeleton of the pressure sore area, which can facilitate the subsequent comparison of the skeleton trends between different pressure sore areas, thereby facilitating the subsequent identification of non-self-pressure sore grade characteristics in the pressure sore area.

[0101] Step 204 , based on the characteristics of the target skeletons in the reference image and the sample pressure sore images, a standard image corresponding to each sample pressure sore image is screened out from all reference images.

[0102] As an example, this step may include the following steps:

[0103] In the first step, any sample pressure ulcer image is determined as a candidate image.

[0104] In the second step, the target similarity between the candidate image and each reference image is determined based on the angles and grayscales corresponding to all skeleton line segments in the candidate image and each reference image.

[0105] The angle corresponding to the skeleton line segment can represent the direction from the skeleton center point on the skeleton line segment to the skeleton fulcrum. For example, the method for obtaining the angle corresponding to the skeleton line segment can be: determine the direction from the skeleton center point on the skeleton line segment to the skeleton fulcrum as the first direction, determine the horizontal right direction as the second direction, and determine the angle formed in the process of rotating the second direction counterclockwise to the first direction as the angle corresponding to the skeleton line segment.

[0106] For example, the formula for determining the target similarity between the candidate image and the reference image may be:

[0107] ;

[0108] in, is the candidate image and the The target similarity between reference images; is the serial number of the reference image; is a natural exponential function; is the number of skeleton segments in the target skeleton; is the sequence number of the skeleton line segment in the target skeleton. For example, the method for obtaining the arrangement order of the skeleton line segments in the target skeleton may be: according to the angles corresponding to the skeleton line segments, the skeleton line segments in the target skeleton are arranged in order from small to large angles, and the arrangement order obtained at this time is the arrangement order of the skeleton line segments in the target skeleton. It is the absolute value function; is the target skeleton in the candidate image, The mean of the gray values ​​corresponding to all pixels on the skeleton line segment; It is Among the target skeletons in the reference images, The mean of the gray values ​​corresponding to all pixels on the skeleton line segment; is the target skeleton in the candidate image, The angle corresponding to the skeleton line segment; It is Among the target skeletons in the reference images, The angle corresponding to the skeleton line segments.

[0109] It should be noted that when The smaller the value, the more likely it is that the candidate image and the first The more similar the grayscales of the skeleton line segments at the same sequence number in the first reference image are, the more similar the candidate image is to the first reference image. The more pressure ulcer areas represented by the reference images are likely to belong to the same pressure ulcer grade. The smaller the value, the more likely it is that the candidate image and the The more similar the angles between the skeleton line segments at the same sequence number in the first reference image are, the closer the candidate image is to the first reference image. The more similar the skeleton trends between the target skeletons in the first reference image are, the more similar the candidate image and the first reference image are. The more pressure ulcer areas represented by the reference images are likely to belong to the same pressure ulcer grade. When the value is larger, it often indicates that the candidate image and the The pressure ulcer areas represented by the reference images are more likely to belong to the same pressure ulcer grade.

[0110] The third step is to select the reference image with the greatest target similarity with the candidate image from all reference images as the standard image corresponding to the candidate image.

[0111] It should be noted that the pressure sore features presented by the pressure sore area represented by the candidate image are often the same as the pressure sore features presented by the pressure sore area represented by the corresponding standard image. That is, the pressure sore level corresponding to the candidate image is often the same as the pressure sore level corresponding to the corresponding standard image.

[0112] Step 205 , according to the difference between each sample pressure sore image and its corresponding standard image, a specific area is screened out from each sample pressure sore image, and based on the specific area in each sample pressure sore image, each sample pressure sore image is enhanced to obtain a target enhanced image.

[0113] As an example, this step may include the following steps:

[0114] In the first step, the target specific evaluation corresponding to each pixel in each sample pressure sore image is determined based on the target similarity between each sample pressure sore image and its corresponding standard image, the texture change evaluation corresponding to each pixel in each sample pressure sore image, and the grayscale difference between each pixel in each sample pressure sore image and its corresponding standard image.

[0115] For example, the formula for determining the target-specific evaluation corresponding to the pixel point in the sample pressure sore image can be:

[0116] ;

[0117] in, It is Sample pressure ulcer images Target-specific evaluation corresponding to pixels; is the serial number of the sample pressure ulcer image; It is The sequence number of the pixel points in the sample pressure sore image; is the normalization function; It is The target similarity between the sample pressure ulcer images and their corresponding standard images; It is Sample pressure ulcer images Texture change evaluation corresponding to each pixel; It is The number of pixels in the standard image corresponding to the sample pressure sore image; It is The sequence number of the pixel points in the standard image corresponding to the sample pressure sore image; It is the absolute value function; It is Sample pressure ulcer images The gray value corresponding to each pixel; It is In the standard image corresponding to the sample pressure sore image, The gray value corresponding to each pixel.

[0118] It should be noted that when The larger the The more likely the pressure sore areas represented by the sample pressure sore images and their corresponding standard images belong to the same pressure sore grade. The larger the Sample pressure ulcer images The greater the grayscale change around the pixel, the Sample pressure ulcer images The more obvious the texture changes around the pixel, the more obvious the texture changes around the pixel. Sample pressure ulcer images The more pixels, the more important they are. and Can be used as The weight of The larger the Sample pressure ulcer images The greater the grayscale difference between a pixel and its corresponding standard image, the greater the grayscale difference between the pixel and its corresponding standard image. Sample pressure ulcer images The more likely the pixel is to represent the The non-self-pressure ulcer grade features in the sample pressure ulcer images. The larger the Sample pressure ulcer images The more likely the pixel is to represent the The non-self-pressure ulcer grade features in the sample pressure ulcer images often indicate that the more The features represented by each pixel point are used to facilitate the pressure sore grade recognition network to learn the non-self-pressure sore grade feature information in the pressure sore area, which can reduce the misjudgment of the pressure sore grade to a certain extent and thus improve the accuracy of pressure sore grade recognition.

[0119] In the second step, pixels whose corresponding target specific evaluation is greater than a preset specific threshold are selected from each sample pressure sore image as specific pixels.

[0120] The preset specific threshold may be a preset threshold for pixel point screening, for example, the preset specific threshold may be 0.7.

[0121] In the third step, the area formed by all specific pixels in each sample pressure ulcer image is determined as the specific area.

[0122] The fourth step is to perform superpixel segmentation on each sample pressure sore image to obtain a set of superpixel blocks corresponding to each sample pressure sore image.

[0123] The superpixel block set corresponding to the sample pressure sore image may include: all superpixel blocks obtained by superpixel segmentation of the sample pressure sore image. A superpixel block, also known as a superpixel, is an area composed of a series of pixel points that are adjacent in position and have similar features such as color, brightness, and texture.

[0124] The fifth step is to select superpixel blocks with pixels belonging to specific areas from the superpixel block set corresponding to each sample pressure sore image as candidate blocks.

[0125] Among them, the candidate block is a superpixel block with unique pixels.

[0126] In the sixth step, the superpixel blocks adjacent to any candidate blocks are selected from the superpixel block set corresponding to each sample pressure sore image as the target blocks.

[0127] For example, the area formed by all candidate blocks in the sample pressure sore image may be used as the candidate area, and the super pixel blocks adjacent to the candidate area in the sample pressure sore image may be used as the target blocks.

[0128] In the seventh step, the connected domain is extracted for each sample pressure sore image to obtain the target connected domain.

[0129] The target connected domain may be a connected domain in a sample pressure sore image.

[0130] In the eighth step, in each sample pressure sore image, the area formed by all target connected domains where all candidate blocks are located and all target connected domains where all target blocks are located is enhanced to obtain a target enhanced image.

[0131] For example, histogram equalization can be performed on the area consisting of all target connected domains where all candidate blocks are located and all target connected domains where all target blocks are located in the sample pressure sore image, so as to enhance the sample pressure sore image, and the enhanced sample pressure sore image is used as the target enhanced image.

[0132] Step 206, expanding the original training set based on all target enhanced images.

[0133] As an example, this step may include the following steps:

[0134] In the first step, each target enhanced image is rotated to obtain a target rotated image.

[0135] In the second step, all target enhanced images and all target rotated images are expanded to the original training set, and the expanded training set is recorded as the target training set.

[0136] It should be noted that in order to improve the generalization ability of the network, the images in the training set can be standardized, normalized, cropped, rotated, scaled, flipped, etc. to make the data in the training set richer.

[0137] Optionally, the training process of the pressure sore grade recognition network may include the following sub-steps:

[0138] In the first sub-step, a CNN is constructed as the pre-training pressure ulcer grade recognition network.

[0139] In the second sub-step, each image in the target training set is labeled with its corresponding pressure ulcer grade through doctor observation.

[0140] In the third sub-step, the target training set is used as the training set, and the pressure sore levels corresponding to the images in the target training set are used as training labels to train the constructed pressure sore level recognition network to obtain a trained pressure sore level recognition network.

[0141] Among them, the loss function in the training process can be a cross entropy loss function, and its optimization algorithm can be a stochastic gradient descent algorithm.

[0142] refer to Figure 3 Based on the same inventive concept as the above method embodiment, the present invention provides a pressure sore auxiliary detection system for orthopedic clinical care, the system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the above computer program is executed by the processor, the steps of a pressure sore auxiliary detection method for orthopedic clinical care are implemented, which may specifically include:

[0143] An image acquisition module 301 is used to acquire a target pressure sore image corresponding to the pressure sore area to be detected;

[0144] The pressure sore level recognition module 302 is used to input the target pressure sore image into a pre-trained pressure sore level recognition network, and obtain the pressure sore level corresponding to the pressure sore area to be detected through the pressure sore level recognition network.

[0145] Figure 4 is a schematic diagram of the structure of a computer device provided by an embodiment of the present invention. Figure 4 As shown, the computer device 400 includes: a memory 401, a processor 402, and a computer program 403 stored in the memory 401 and running on the processor 402, wherein when the processor 402 executes the computer program 403, the computer device can execute any one of the pressure ulcer auxiliary detection methods for orthopedic clinical care introduced above.

[0146] Based on the same inventive concept as the above method embodiment, the present invention provides a server, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the device executes any of the above pressure sore auxiliary detection methods for orthopedic clinical care.

[0147] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer program product, which includes: computer program code, when the computer program code runs on a computer, enables the computer to execute any one of the above-mentioned pressure ulcer auxiliary detection methods for orthopedic clinical care.

[0148] Based on the same inventive concept as the above-mentioned method embodiment, the present invention provides a computer-readable storage medium, which stores a computer program code. When the computer program code runs on a computer, the computer executes any one of the above-mentioned pressure ulcer auxiliary detection methods for orthopedic clinical care.

[0149] In summary, the present invention comprehensively considers multiple factors related to non-self-pressure sore grade characteristics, such as texture change evaluation, and the difference between the sample pressure sore image and its corresponding standard image, thereby quantifying the specific area in the sample pressure sore image that characterizes the area to which the non-self-pressure sore grade characteristics belong, and based on the specific area in each sample pressure sore image, each sample pressure sore image is enhanced, thereby enhancing the non-self-pressure sore grade feature information in the sample pressure sore image, so that the pressure sore grade recognition network can learn the non-self-pressure sore grade feature information in the pressure sore area, which can improve the accuracy and robustness of the network, thereby reducing the misjudgment of pressure sore grades to a certain extent, and thereby improving the accuracy of pressure sore grade recognition.

[0150] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A pressure sore auxiliary detection method for orthopedic clinical nursing, characterized in that: The following steps are involved: Acquire a target pressure sore image corresponding to the pressure sore area to be detected; The target pressure sore image is input into the pre-trained pressure sore grade recognition network, and the pressure sore grade corresponding to the pressure sore area to be detected is obtained through the pressure sore grade recognition network; During the training process of the pressure sore grade recognition network, the expansion of the original training set of the pressure sore grade recognition network includes the following steps: Obtaining pressure ulcer images representing different single pressure ulcer levels as reference images, and obtaining different sample pressure ulcer images; Determine the texture change evaluation corresponding to each pixel point according to the gradient value corresponding to each pixel point in each reference image and each sample pressure sore image and the grayscale condition in the corresponding preset neighborhood; Based on the texture change evaluation corresponding to the pixels, the respective target skeletons are screened out from each reference image and each sample pressure ulcer image; According to the characteristics of the target skeletons in the reference image and the sample pressure sore image, a standard image corresponding to each sample pressure sore image is screened out from all reference images; According to the difference between each sample pressure sore image and its corresponding standard image, a specific region is screened out from each sample pressure sore image, and based on the specific region in each sample pressure sore image, each sample pressure sore image is enhanced to obtain a target enhanced image; The original training set is expanded based on all target enhanced images; The step of determining the texture change evaluation corresponding to each pixel point according to the gradient value corresponding to each pixel point in each reference image and each sample pressure sore image and the grayscale condition in the corresponding preset neighborhood includes: Determine any pixel point in any reference image or any sample pressure sore image as a marked pixel point; Determine the grayscale change factor corresponding to the marked pixel point according to the maximum grayscale value and the minimum grayscale value in the preset neighborhood corresponding to the marked pixel point; Determine a texture change evaluation corresponding to the marked pixel point according to the gradient value and the grayscale change factor corresponding to the marked pixel point, wherein both the gradient value and the grayscale change factor are positively correlated with the texture change evaluation; The texture change evaluation based on pixel points corresponding to the image is performed to select the respective target skeletons from each reference image and each sample pressure sore image, including: Determine any reference image or any sample pressure ulcer image as a marked image; Selecting a preset number of pixel points with the largest corresponding texture change evaluation from the marked image as target points for constituting the skeleton; Filter out the target point with the largest corresponding texture change evaluation from all the target points as the skeleton center point in the marked image, and determine each target point in the marked image except the skeleton center point as the skeleton support point; Connecting the skeleton center point and each skeleton pivot point in the marked image to obtain a skeleton line segment; All skeleton line segments in the labeled image are used to form a target skeleton in the labeled image; The method of screening out a specific area from each sample pressure sore image according to the difference between each sample pressure sore image and its corresponding standard image includes: Determine the target specific evaluation corresponding to each pixel point in each sample pressure sore image according to the target similarity between each sample pressure sore image and its corresponding standard image, the texture change evaluation corresponding to each pixel point in each sample pressure sore image, and the grayscale difference between each pixel point in each sample pressure sore image and its corresponding standard image; Pixels whose corresponding target specific evaluation is greater than a preset specific threshold are selected from each sample pressure sore image as specific pixels; The area formed by all specific pixel points in each sample pressure sore image is determined as the specific area.

2. The pressure sore auxiliary detection method for orthopedic clinical nursing according to claim 1 is characterized in that: The method of screening out a standard image corresponding to each sample pressure sore image from all reference images according to the characteristics of the target skeleton in the reference image and the sample pressure sore image comprises: Determine any sample pressure sore image as a candidate image; Determining the target similarity between the candidate image and each reference image according to the angles and grayscales corresponding to all skeleton line segments in the candidate image and each reference image; A reference image having the greatest target similarity with the candidate image is selected from all reference images as a standard image corresponding to the candidate image.

3. The pressure sore auxiliary detection method for orthopedic clinical nursing according to claim 2 is characterized in that: The formula corresponding to the target similarity between the candidate image and the reference image is: ； in, is the candidate image and the The target similarity between reference images; is the serial number of the reference image; is a natural exponential function; is the number of skeleton segments in the target skeleton; is the sequence number of the skeleton line segment in the target skeleton; It is the absolute value function; is the target skeleton in the candidate image, The mean of the gray values ​​corresponding to all pixels on the skeleton line segment; It is Among the target skeletons in the reference images, The mean of the gray values ​​corresponding to all pixels on the skeleton line segment; is the target skeleton in the candidate image, The angle corresponding to the skeleton line segment; It is Among the target skeletons in the reference images, The angle corresponding to the skeleton line segments.

4. The pressure sore auxiliary detection method for orthopedic clinical nursing according to claim 1 is characterized in that: The formula corresponding to the target-specific evaluation of the pixel points in the sample pressure sore image is: ; in, It is Sample pressure ulcer images Target-specific evaluation corresponding to pixels; is the serial number of the sample pressure ulcer image; It is The sequence number of the pixel points in the sample pressure sore image; is the normalization function; It is The target similarity between the sample pressure ulcer images and their corresponding standard images; It is Sample pressure ulcer images Texture change evaluation corresponding to each pixel; It is The number of pixels in the standard image corresponding to the sample pressure sore image; It is The sequence number of the pixel points in the standard image corresponding to the sample pressure sore image; It is the absolute value function; It is Sample pressure ulcer images The gray value corresponding to each pixel; It is In the standard image corresponding to the sample pressure sore image, The gray value corresponding to each pixel.

5. The pressure sore auxiliary detection method for orthopedic clinical nursing according to claim 1 is characterized in that: The method of enhancing each sample pressure sore image based on the specific region in each sample pressure sore image to obtain a target enhanced image includes: Perform superpixel segmentation on each sample pressure sore image to obtain a superpixel block set corresponding to each sample pressure sore image; From the superpixel block set corresponding to each sample pressure sore image, superpixel blocks with pixels belonging to specific areas are selected as candidate blocks; Filter out the superpixel blocks adjacent to any candidate blocks from the superpixel block set corresponding to each sample pressure sore image as the target block; Perform connected domain extraction on each sample pressure sore image to obtain the target connected domain; In each sample pressure sore image, the area formed by all target connected domains where all candidate blocks are located and all target connected domains where all target blocks are located is enhanced to obtain a target enhanced image.

6. The pressure sore auxiliary detection method for orthopedic clinical nursing according to claim 1 is characterized in that: The original training set is expanded according to all target enhanced images, including: Perform image rotation on each target enhanced image to obtain a target rotated image; All target augmented images and all target rotated images are added to the original training set.

7. A pressure sore auxiliary detection system for orthopedic clinical nursing, characterized in that: The method comprises a processor and a memory, wherein the processor is used to process instructions stored in the memory to implement a pressure ulcer auxiliary detection method for orthopedic clinical care according to any one of claims 1 to 6.

Images