Automatic wound analysis method and system based on machine vision

Abstract

The invention provides an automatic wound analysis method and system based on machine vision. The method comprises the steps that a target wound image and wound record information collected by an image collection device are obtained, and the wound record information comprises a wound type identifier and a historical wound recognition result; based on the target wound image, identifying the wound through a deep learning model corresponding to the wound type identifier to obtain a target wound identification result; based on a target wound identification result and a historical wound identification result, fitting a wound change trend in regional stages to obtain a wound fitting result of each stage of each sub-region; and determining a wound analysis result based on the wound fitting result of each stage of each sub-region. According to the technical scheme, the accuracy of wound analysis is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of image processing technology, and in particular to an automated wound analysis method and system based on machine vision. Background Art

[0002] With the development of medical technology, automated wound analysis has become increasingly important in clinical diagnosis and treatment. Especially in scenarios such as chronic wound management, trauma emergency treatment, and postoperative care, doctors need to regularly evaluate the healing of wounds in order to adjust treatment plans in a timely manner. Traditional wound assessment methods rely on the doctor's subjective judgment, are easily affected by human factors, and are time-consuming and labor-intensive. Therefore, there is an urgent need for a wound analysis method that can objectively and accurately evaluate the healing of wounds and improve medical efficiency and patient recovery speed.

[0003] In recent years, image recognition technology based on deep learning has been widely used, such as using convolutional neural networks for wound recognition and classification, which has achieved good results. However, most existing deep learning methods focus on the extraction of global features, which has the problem of low accuracy in wound analysis. Summary of the invention

[0004] The embodiment of the present application provides an automated wound analysis method based on machine vision to solve the problem of low accuracy of wound analysis in the prior art.

[0005] In a first aspect, an embodiment of the present application provides an automated wound analysis method based on machine vision, comprising:

[0006] Acquire a target wound image and wound record information acquired by an image acquisition device, wherein the wound record information includes a wound type identifier and a historical wound recognition result;

[0007] Based on the target wound image, identifying the wound by using a deep learning model corresponding to the wound type identifier to obtain a target wound identification result;

[0008] Based on the target wound recognition result and the historical wound recognition result, the wound change trend is fitted in different regions and stages to obtain the wound fitting results in each stage of each sub-region;

[0009] The wound analysis result is determined based on the wound fitting results of each sub-region at each stage.

[0010] Optionally, based on the target wound recognition result and the historical wound recognition result, fitting the wound change trend in different regions and stages to obtain the wound fitting results in each stage of each sub-region includes:

[0011] Performing semantic segmentation on the target wound image to segment a wound region in the target wound image into at least two sub-regions, and different sub-regions correspond to different healing period labels;

[0012] When there are multiple sub-regions, the target wound recognition result and the historical wound recognition result are divided into the first local wound recognition result and the second local wound recognition result of each sub-region respectively;

[0013] For each sub-region, based on the first local wound recognition result and the second local wound recognition result of the sub-region, the wound change trend is fitted in stages to obtain the wound fitting results of each stage of the sub-region.

[0014] Optionally, the wound change trend is fitted in stages based on the first local wound recognition result and the second local wound recognition result of the sub-region to obtain the wound fitting results of each stage of the sub-region, including:

[0015] Acquire segmentation point information of the sub-region, and divide the first local wound recognition result and the second local wound recognition result into corresponding stages according to the segmentation point information;

[0016] Based on all local wound recognition results of each stage in the sub-region, the fitting method corresponding to each stage is used to fit the wound change trend of the corresponding stage to obtain the wound fitting results of each stage in the sub-region.

[0017] Optionally, the stages include: an untreated stage, a pre-treatment stage, a mid-treatment stage, and a post-treatment stage;

[0018] Using a fitting method corresponding to each stage to fit the wound change trend of the corresponding stage, including at least one of the following:

[0019] Linear regression was used to fit the trends of wound changes during the untreated period;

[0020] Polynomial regression was used to fit the wound change trend during the early stage of treatment;

[0021] Exponential regression was used to fit the wound change trend during the mid-treatment phase;

[0022] A saturated growth model was used to fit the wound change trend in the later stage of treatment.

[0023] Optionally, based on the target wound image, identifying the wound by using a deep learning model corresponding to the wound type identifier to obtain a target wound identification result includes:

[0024] Preprocessing the target wound image to obtain a preprocessed image;

[0025] Pre-segmenting the pre-processed image to distinguish a wound area from a non-wound area;

[0026] A deep learning model corresponding to the wound type identifier is selected from the preset model library, and the features of the wound area are extracted through the deep learning model corresponding to the wound type identifier to obtain the target wound recognition result.

[0027] Optionally, the deep learning model corresponding to the wound type identifier includes any of the following:

[0028] A burn recognition model corresponding to the burn mark, wherein the burn recognition model is obtained by jump-connecting an encoder and a decoder;

[0029] A cut injury recognition model corresponding to the cut injury identifier, wherein the cut injury recognition model adopts a masked region convolutional neural network structure;

[0030] A scratch recognition model corresponding to the scratch mark, wherein the scratch recognition model adopts a lightweight convolutional neural network structure;

[0031] An infected wound recognition model corresponding to the infection identifier, wherein the infected wound recognition model is a deep learning model based on a densely connected network;

[0032] A composite wound recognition model corresponding to the composite identifier, wherein the composite wound recognition model is a multi-task learning model.

[0033] Optionally, the target wound identification result includes: wound area, wound area color features and wound area edge features;

[0034] The determining of the wound analysis result based on the wound fitting result of each sub-region at each stage includes:

[0035] Determining the severity of the wound based on the wound area, the wound area color characteristics and the wound area edge characteristics;

[0036] The wound analysis result is determined based on the wound severity and the wound fitting results of each sub-region and each stage.

[0037] In a second aspect, an embodiment of the present application provides an automated wound analysis system based on machine vision, comprising:

[0038] An acquisition module, used to acquire a target wound image and wound record information acquired by an image acquisition device, wherein the wound record information includes a wound type identifier and a historical wound recognition result;

[0039] An identification module, configured to identify the wound based on the target wound image by using a deep learning model corresponding to the wound type identifier to obtain a target wound identification result;

[0040] A fitting module, for fitting the wound change trend in different regions and stages based on the target wound recognition result and the historical wound recognition result, to obtain the wound fitting results in each sub-region and each stage;

[0041] A determination module is used to determine the wound analysis result based on the wound fitting results of each sub-region at each stage.

[0042] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an automated wound analysis method based on machine vision as described in any one of the first aspects.

[0043] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements an automated wound analysis method based on machine vision as described in any one of the first aspects.

[0044] In an embodiment of the present application, an automated wound analysis method based on machine vision is provided. The method includes: obtaining a target wound image and wound record information collected by an image acquisition device, wherein the wound record information includes a wound type identifier and a historical wound recognition result; based on the target wound image, identifying the wound through a deep learning model corresponding to the wound type identifier to obtain a target wound recognition result; based on the target wound recognition result and the historical wound recognition result, fitting the wound change trend in different regions and stages to obtain the wound fitting results in each sub-region at each stage; determining the wound analysis result based on the wound fitting results in each sub-region at each stage. The present application identifies the wound through a deep learning model corresponding to the wound type identifier, and fits the wound change trend in different regions and stages, which can effectively overcome the defects of the existing solutions, realize the refined and automated analysis of the wound, improve the accuracy of the wound analysis, and also improve the adaptability to the analysis of wounds of different types and stages. Furthermore, the embodiments of the present application can more accurately predict and evaluate the healing speed and final state of the wound by using specific fitting methods at different stages.

[0045] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 A flowchart of an automated wound analysis method based on machine vision provided in an embodiment of the present application;

[0048] Figure 2 A schematic diagram of the structure of an automated wound analysis system based on machine vision provided in an embodiment of the present application;

[0049] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0051] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The sequence numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to different types.

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0053] Figure 1 A flowchart of an automated wound analysis method based on machine vision provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0054] S11. Acquire a target wound image and wound record information acquired by an image acquisition device, wherein the wound record information includes a wound type identifier and historical wound recognition results.

[0055] It should be understood that the image acquisition device can be a device for capturing the target wound image, such as a digital camera, a smartphone camera, etc., which can produce high-quality digital images for subsequent analysis. Wound type identification includes burn identification, cut identification, abrasion identification, infection identification, composite identification, etc. The historical recognition results are specific information obtained by using a deep learning model to identify historical wound images, such as wound location, wound area, wound area color characteristics, and wound area edge characteristics, etc., where the wound area refers to the wound area in the image, and the wound area can be converted into the actual wound area through a coordinate system conversion method. The actual wound area is a quantitative representation of the wound size. The wound size not only includes the actual area of ​​the wound, but may also involve dimensions such as the depth, length, and width of the wound. The acquisition time of the historical wound image is earlier than the acquisition time of the target wound image.

[0056] S12. Based on the target wound image, the wound is identified by a deep learning model corresponding to the wound type identifier to obtain a target wound identification result.

[0057] In this embodiment, burns, cuts, abrasions, infections, and complex wounds have different characteristics, so designing different deep learning models can better capture and identify these characteristics.

[0058] S13. Based on the target wound recognition result and the historical wound recognition result, the wound change trend is fitted in different regions and stages to obtain the wound fitting results of each sub-region and each stage.

[0059] Among them, regional stage fitting refers to dividing the wound area in the target wound image into multiple sub-areas, and different sub-areas correspond to different healing periods, and then fitting the wound change trend in stages for different areas. Among them, the wound change trend can refer to the change of wound area, wound color, wound edge, redness and swelling degree in the area; the healing period includes: hemostasis period, inflammation period, proliferation period, remodeling period, etc., and different healing periods can be represented by wound characteristics, for example: light red area, dark red area, yellow area, black area, etc. are divided separately. Alternatively, the basis for regionalization can be anatomical structure. Specifically: because the skin structure and healing characteristics of different anatomical positions may be different, regionalization based on anatomical structure helps to more accurately evaluate the healing status of each part.

[0060] Exemplarily, in this embodiment, a wound area A may be divided into two sub-areas, where sub-area 1 is in the inflammatory stage and sub-area 2 is in the proliferative stage.

[0061] S14. Determine wound analysis results based on the wound fitting results of each sub-region at each stage.

[0062] It should be understood that the wound fitting results of each sub-region at each stage are detailed descriptions of the wound, and the wound analysis results can be a wound analysis report. For the same wound, there may be areas in different healing stages at the same time. Therefore, in order to carry out targeted treatment for each part, this embodiment comprehensively considers factors such as the healing trend of all sub-regions at each stage and the current wound status to form a detailed wound analysis report. The wound analysis results may include the current severity of the wound, the current severe area, the estimated healing time of each sub-region, and recommendations for treatment measures to be taken for each sub-region.

[0063] The embodiment of the present application identifies wounds through a deep learning model corresponding to the wound type identifier, and fits the wound change trend by region and stage, which can effectively overcome the defects of existing solutions, realize refined and automated analysis of wounds, improve the accuracy of wound analysis, and also improve the adaptability of wound analysis of different types and stages.

[0064] It should be noted that for different types of wounds, the same deep learning model has different recognition accuracy. In order to improve the recognition accuracy of each type of wound, the embodiments of the present application provide different types of deep learning models. In some optional embodiments, S12, based on the target wound image, the wound is identified by the deep learning model corresponding to the wound type identifier to obtain the target wound recognition result, including:

[0065] Step 121: preprocess the target wound image to obtain a preprocessed image.

[0066] Step 122: Pre-segment the pre-processed image to distinguish the wound area from the non-wound area.

[0067] Step 123: Select a deep learning model corresponding to the wound type identifier from a preset model library, and extract features of the wound area through the deep learning model corresponding to the wound type identifier to obtain a target wound recognition result.

[0068] It should be understood that preprocessing includes binarization, denoising, grayscale, edge processing, etc. After pre-segmentation, only the wound area is input into the deep learning model corresponding to the wound type identifier, which can reduce the impact of the non-wound area on the recognition of the deep learning model, thereby improving the recognition accuracy of the deep learning model.

[0069] It is also important to understand that different types of wounds have different characteristics that determine differences in their treatment and management. Capturing these characteristics can help select appropriate treatments and monitoring tools. Here are the characteristics of several common wound types and why they are designed:

[0070] Burn-type wounds have the following characteristics: Burns usually cause discoloration, blisters, eschars, etc. of the skin. The degree of pain caused by burns varies depending on the depth of the burn. Superficial burns are more painful, while deep burns may feel numb. Superficial burns are often accompanied by blisters, while deep burns are less likely to have blisters. Severe burns may cause tissue necrosis and form eschars. In view of the above characteristics, this embodiment may adopt an encoder-decoder architecture to effectively capture and reconstruct the complex structure of the wound, especially for features such as blisters and eschars of burns. Through jump connections, more detailed information can be retained, which helps to more accurately identify and segment the burn area.

[0071] Cut-type wounds have the following characteristics: Cuts usually appear as straight or curved wounds. Cuts are often accompanied by obvious bleeding, especially when arteries are injured. The edges of cuts are usually neat and easy to identify. The depth of cuts varies, and may be limited to the epidermis, or may penetrate deep into the muscle or even the bone. In view of the above characteristics, this embodiment may use a masked region convolutional neural network, which can perform object detection and segmentation at the same time and is suitable for identifying linear and edge features of cuts. In addition, the masked region convolutional neural network can accurately locate the position and edges of the wound, which helps to evaluate the depth and length of the wound.

[0072] Abrasion type wounds have the following characteristics: Abrasions mainly affect the epidermis and usually appear as wear on the skin surface. A small amount of blood and tissue fluid often seep out of the abrasion site. The shape of the abrasion is usually irregular and may be flaky or strip-shaped. The abrasion site is often accompanied by redness, swelling and slight pain. In view of the above characteristics, this embodiment may use a lightweight convolutional neural network. Since the characteristics of abrasions are relatively simple, a lightweight convolutional neural network is sufficient to complete the recognition task while maintaining a high computational efficiency. The lightweight model can run quickly in a resource-limited environment and is suitable for real-time monitoring and evaluation.

[0073] Infected wounds have the following characteristics: Infected wounds are often accompanied by symptoms such as redness, swelling, fever, and pain. Infected wounds may have pus exuding, which is usually yellow or green in color. Infected wounds may emit a foul odor. The skin temperature of the infected area may be higher than the surrounding normal skin. In view of the above characteristics, this embodiment can use a densely connected network to effectively extract complex features, which is suitable for identifying multiple symptoms of infected wounds.

[0074] Complex wounds have the following characteristics: Complex wounds may contain multiple types of injuries such as burns, cuts, abrasions, etc. Complex wounds have a more complex structure and may involve multiple levels of tissue damage. Multiple symptoms such as redness, swelling, exudation, and pain may appear at the same time. In view of the above characteristics, this embodiment can use a multi-task learning model to process multiple tasks at the same time. The model is suitable for identifying and evaluating multiple characteristics of complex wounds. This embodiment can also improve the overall performance of the model by jointly optimizing multiple tasks to ensure accurate identification of different types of injuries.

[0075] Therefore, the deep learning model corresponding to the wound type identifier includes any of the following: (1) a burn recognition model corresponding to the burn identifier, which is obtained by jump-connecting the encoder and the decoder; (2) a cut wound recognition model corresponding to the cut wound identifier, which adopts a masked area convolutional neural network structure; (3) an abrasion recognition model corresponding to the abrasion identifier, which adopts a lightweight convolutional neural network structure; (4) an infected wound recognition model corresponding to the infection identifier, which is a deep learning model based on a densely connected network; (5) a composite wound recognition model corresponding to the composite identifier, which is a multi-task learning model.

[0076] In summary, each type of wound has its own unique characteristics, and different deep learning models are designed to better capture and identify these characteristics, thereby improving the accuracy and efficiency of wound identification and assessment. The selection and design of these models are all to meet the needs of specific types of wounds in clinical diagnosis and treatment.

[0077] It should be noted that the present embodiment can segment the wound area according to different healing stages, and can also divide the same wound area into stages in time according to factors such as whether the same wound area has been treated, how many times it has been treated, and what type of treatment method it has undergone, so as to achieve regional stage fitting to improve the accuracy of wound analysis. In some optional embodiments, S13, based on the target wound recognition result and the historical wound recognition result, the wound change trend is fitted by regional stage to obtain the wound fitting results of each stage in each sub-area, including:

[0078] Step 131: semantically segment the target wound image to segment the wound area in the target wound image into at least two sub-areas, and different sub-areas correspond to different healing period labels. Exemplary: the healing period label includes a label corresponding to the hemostatic period, a label corresponding to the inflammatory period, a label corresponding to the proliferative period, a label corresponding to the remodeling period, etc. Using a deep learning model for semantic segmentation can more accurately identify various areas of the wound and reduce misjudgment.

[0079] Step 132: When there are multiple sub-regions, the target wound recognition result and the historical wound recognition result are divided into the first local wound recognition result and the second local wound recognition result of each sub-region. By decomposing the overall recognition result into each sub-region, the change trend of each region can be analyzed more accurately.

[0080] Exemplarily, the number of sub-regions is 3, then this embodiment can divide the target wound recognition result into the first local wound recognition result of sub-region 1, the first local wound recognition result of sub-region 2, and the first local wound recognition result of sub-region 3. Similarly, this embodiment can divide the historical wound recognition result into the second local wound recognition result of sub-region 1, the second local wound recognition result of sub-region 2, and the second local wound recognition result of sub-region 3.

[0081] Step 133 : for each sub-region, based on the first local wound recognition result and the second local wound recognition result of the sub-region, fitting the wound change trend in stages to obtain the wound fitting results of each stage of the sub-region.

[0082] It should be understood that the second local wound recognition result may be the local wound recognition results corresponding to multiple historical moments. Therefore, for each sub-region, its corresponding first local wound recognition result and second local wound recognition result may involve one stage or multiple stages. When multiple stages are involved, this embodiment can fit the wound change trend in stages.

[0083] Among them, staging refers to dividing the healing process of local wounds into different stages in order to more systematically evaluate and manage the healing process of local wounds. The staging criteria are usually based on the following aspects: time, treatment intervention, etc. Exemplarily, sub-region 1 is currently in the growth stage, so sub-region 1 has previously experienced the hemostasis stage and the inflammation stage. The embodiment of the present application can analyze the wound fitting results in the hemostasis stage, the inflammation stage and the growth stage through the first local wound recognition result and the second local wound recognition result of the region, while sub-region 2 is currently in the inflammation stage, so sub-region 2 has previously experienced the hemostasis stage. The embodiment of the present application can analyze the wound fitting results in the hemostasis stage and the inflammation stage through the first local wound recognition result and the second local wound recognition result of the region.

[0084] By dividing the wound area into multiple sub-areas, the embodiments of the present application can more finely manage the healing of each area and provide more targeted treatment suggestions. By dividing the wounds in the same area into different stages in time, the healing process of the wound can be more systematically evaluated and managed, and the treatment plan can be adjusted in time.

[0085] The stage division in this embodiment is of great significance for generating accurate wound analysis results. Optionally, the stage division in this embodiment can be performed based on segmentation point information. As an optional implementation method, step 133, based on the first local wound recognition result and the second local wound recognition result of the sub-region, the wound change trend is fitted in stages to obtain the wound fitting results of each stage of the sub-region, including:

[0086] Step a1, obtain the segmentation point information of the sub-area, and divide the first local wound recognition result and the second local wound recognition result into corresponding stages according to the segmentation point information; step a2, based on all local wound recognition results of each stage in the sub-area, use the fitting method corresponding to each stage to fit the wound change trend of the corresponding stage, and obtain the wound fitting results of each stage in the sub-area.

[0087] Among them, the segmentation point information includes the scratch time point, the wound treatment time point, etc. The wound treatment time point can be the treatment time point of using a specified wound treatment method, can be the treatment time point of using any wound treatment method, or can be a periodic segmentation time point. According to the segmentation point information, for example, the first stage is: the period from the scratch time point to the time point of the first wound treatment, the second stage is from the 1st day to the 10th day after the first wound treatment, and the third stage is from the 1st day to the 10th day after the second wound treatment.

[0088] It should be noted that by using different fitting methods to fit the wound change trends at different stages, the healing speed and final state of the wound can be predicted more accurately, providing a scientific basis for medical personnel. Among them, each stage includes: untreated stage, pre-treatment stage, mid-treatment stage and late-treatment stage. Accordingly, the wound change trends at the corresponding stages are fitted using the fitting methods corresponding to each stage, including at least one of the following: using linear regression to fit the wound change trends at the untreated stage; using polynomial regression to fit the wound change trends at the pre-treatment stage; using exponential regression to fit the wound change trends at the mid-treatment stage; using a saturated growth model to fit the wound change trends at the late-treatment stage.

[0089] The wound change trends at each stage may include the change trends of wound indicators such as area, depth, degree of redness and swelling, and healing speed. The above fitting strategy is met for the change trend of any wound indicator.

[0090] Among them, the linear regression used in the untreated stage adopts the following formula:

[0091]

[0092] Among them, y1 represents the wound index value in the untreated stage, t represents time, and a 01The wound index value at the initial moment of the untreated stage can be determined by the first local wound identification result of the sub-area in the untreated stage. 11 It represents the linear change rate of wound indicators over time, which can be estimated by the change trend of the local wound identification results of the sub-region in the untreated stage. λ1 represents the regularization parameter in linear regression, which is used to control the complexity of the model and prevent overfitting. Its value can be adjusted according to the amount of data in the sub-region in the untreated stage. The larger the amount of data, the smaller the regularization parameter can be.

[0093] For example, a 11 When it is equal to -0.5, it means that the wound area decreases by 0.5 square centimeters every day.

[0094] The polynomial regression used in the pre-treatment phase was as follows:

[0095]

[0096] Among them, y2 represents the wound index value in the early stage of treatment, t represents time, and a 02 Indicates the wound index value at the initial moment of the early stage of treatment, a 13 、a 22 、a 32 , …, a n2 The polynomial coefficients in the early stage of treatment describe the nonlinear changes of wound indicators over time and can be determined by least squares fitting of all local wound identification results in the sub-region in the early stage of treatment. λ2 is the regularization parameter in the early stage of treatment, which is used to control the complexity of the model.

[0097] For example, n=2, a 12 =-0.5, a 22 =0.1, it means that the wound area decreases by 0.5 square centimeters every day, but the rate of decrease increases gradually with time, because a 22 The value of is positive.

[0098] The exponential regression used in the mid-treatment phase was as follows:

[0099]

[0100] Among them, y3 represents the wound index value in the middle stage of treatment, t represents time, a 03 It represents the initial value of the mid-treatment stage, which can be determined by the first local wound identification result of the sub-area in the mid-treatment stage. 13 It represents the wound healing growth rate in the middle stage of treatment, which can be estimated by the changing trend of the local wound identification results in the sub-region in the middle stage of treatment. 23The quadratic term coefficient representing the mid-treatment stage is used to adjust the curvature of the growth curve and can be determined by least squares fitting of the local wound identification results of the sub-region in the mid-treatment stage.

[0101] The saturation growth model used in the late treatment phase uses the following formula:

[0102]

[0103] Among them, y4 represents the wound index value in the later stage of treatment, t represents time, a 04 is the maximum value in the late stage of treatment, indicating the final state of wound healing, and its value can be estimated by the last local wound identification result of the sub-region in the late stage of treatment. 14 is the wound healing growth rate in the later stage of treatment. The larger the value, the faster the healing speed. Its value can be estimated by the changing trend of the local wound identification results of the sub-region in the later stage of treatment. 24 is the time constant in the late stage of treatment, indicating the time point when half of the maximum value is reached, which can be used to evaluate the intermediate stage of the healing process. Its value can be determined by least square fitting of the local wound identification results of the sub-region in the late stage of treatment. 34 It is the offset in the later stage of treatment and is used to adjust the baseline position of the curve. It can reflect the basic state of the wound healing process. Its value can be determined by least square fitting of the local wound identification results of the sub-region in the later stage of treatment.

[0104] In other words, in the untreated stage, the change of the wound is usually linear, because there is no external intervention, the state of the wound is relatively stable, and the change trend is relatively simple. Therefore, this embodiment can use linear regression to fit the change trend of the wound in the untreated stage. In the early stage of treatment, the change of the wound may be more complicated, because after the treatment starts, the state of the wound will change rapidly, and polynomial regression can better capture this nonlinear change trend. Therefore, this embodiment can use polynomial regression to fit the change trend of the wound in the early stage of treatment. In the middle stage of treatment, the healing speed of the wound may be accelerated, showing an exponential growth trend. Therefore, exponential regression can better fit this accelerated healing process. In the late stage of treatment, the wound is usually basically closed, scars are formed, exudate basically disappears, and pain basically disappears. The main feature of this stage is that the healing speed of the wound gradually slows down and enters a stable or slowly remodeled stage. Therefore, the saturated growth model is suitable for describing the trend of gradually approaching a maximum value. In the late stage of treatment, the healing process of the wound gradually approaches a stable state.

[0105] Through these steps, the wound healing process can be more accurately evaluated and managed, providing medical staff with scientific treatment recommendations and improving treatment outcomes.

[0106] In some optional embodiments, the target wound recognition result includes: wound area, wound area color features and wound area edge features; wherein the wound area may refer to the area occupied by the wound in the image, usually in square centimeters. The wound area color features may refer to the color distribution of the wound, which may include areas of different colors such as red, yellow, and black. The wound area edge features may refer to the shape of the wound edge, such as whether it is neat, whether there are irregular edges, etc.

[0107] Accordingly, S14, determining the wound analysis result based on the wound fitting result of each sub-region at each stage, including:

[0108] Step 141: Determine the severity of the wound based on the wound area, the wound area color characteristics, and the wound area edge characteristics.

[0109] Among them, the wound severity of the corresponding sub-region is comprehensively evaluated according to the wound area, color characteristics and edge characteristics of each sub-region, which is usually used to judge the urgency and treatment needs of the partial area of ​​the wound. Reasons for comprehensive evaluation of multiple features: The wound area of ​​the sub-region corresponding to the hemostatic period or the inflammatory period is an important indicator for evaluating the severity of the wound. The larger the wound area of ​​these two sub-regions, the higher the severity. The color characteristics of the wound area in each region can reflect the state of the local wound, such as red may indicate inflammation, and black may indicate necrosis. The edge characteristics of the wound area can reflect the healing of the local wound, such as neat edges may indicate good healing, and irregular edges may indicate poor healing. The severity of the wound can be the severity of the wound as a whole, which can be comprehensively determined according to the severity of the wound corresponding to each sub-region.

[0110] As a possible implementation manner, this embodiment can assess the severity of the wound by setting a threshold or using a scoring system.

[0111] Step 142: Determine the wound analysis result based on the wound severity and the wound fitting results of each sub-region and each stage.

[0112] It should be understood that a comprehensive assessment report on the wound status is formed by comprehensively considering the severity of the wound and the wound fitting results of each sub-area and each stage. The wound analysis results include the current state of the wound, healing progress, potential risks, treatment recommendations, etc.

[0113] In this embodiment, by comprehensively evaluating the severity of the wound and the wound change trend of each sub-region at each stage, the healing condition of each sub-region of the wound can be comprehensively evaluated, and more comprehensive treatment suggestions can be provided for each sub-region.

[0114] Figure 2 A schematic diagram of the structure of an automated wound analysis system based on machine vision provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes:

[0115] The acquisition module 21 is used to acquire the target wound image and wound record information acquired by the image acquisition device, wherein the wound record information includes a wound type identifier and a historical wound recognition result.

[0116] The identification module 22 is used to identify the wound based on the target wound image through a deep learning model corresponding to the wound type identifier to obtain a target wound identification result.

[0117] The fitting module 23 is used to fit the wound change trend in different regions and stages based on the target wound recognition result and the historical wound recognition result, so as to obtain the wound fitting results in each sub-region and each stage.

[0118] The determination module 24 is used to determine the wound analysis result based on the wound fitting result of each sub-region at each stage.

[0119] Figure 2 The automated wound analysis system based on machine vision can perform Figure 1 The implementation principle and technical effect of the automated wound analysis method based on machine vision described in the illustrated embodiment will not be described in detail. The specific manner in which each module and unit performs operations in the automated wound analysis system based on machine vision in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0120] In one possible design, Figure 2 The machine vision-based automated wound analysis system of the illustrated embodiment may be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0121] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0122] The processing component 32 is used to: obtain the target wound image and wound record information collected by the image acquisition device, wherein the wound record information includes a wound type identifier and a historical wound recognition result; based on the target wound image, identify the wound through a deep learning model corresponding to the wound type identifier to obtain a target wound recognition result; based on the target wound recognition result and the historical wound recognition result, fit the wound change trend in different regions and stages to obtain the wound fitting results of each sub-region and each stage; determine the wound analysis result based on the wound fitting results of each sub-region and each stage.

[0123] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0124] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0125] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0126] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0127] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0128] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0129] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is an automated wound analysis method based on machine vision.

[0130] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0131] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0132] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An automated wound analysis method based on machine vision, characterized in that: include: Acquire a target wound image and wound record information acquired by an image acquisition device, wherein the wound record information includes a wound type identifier and a historical wound recognition result; Based on the target wound image, identifying the wound by using a deep learning model corresponding to the wound type identifier to obtain a target wound identification result; Based on the target wound recognition result and the historical wound recognition result, the wound change trend is fitted in different regions and stages to obtain the wound fitting results in each stage of each sub-region; The wound analysis result is determined based on the wound fitting results of each sub-region at each stage.

2. The method according to claim 1, characterized in that The step of fitting the wound change trend by region and stage based on the target wound recognition result and the historical wound recognition result to obtain the wound fitting results of each stage in each sub-region includes: Performing semantic segmentation on the target wound image to segment a wound region in the target wound image into at least two sub-regions, and different sub-regions correspond to different healing period labels; When there are multiple sub-regions, the target wound recognition result and the historical wound recognition result are divided into the first local wound recognition result and the second local wound recognition result of each sub-region respectively; For each sub-region, based on the first local wound recognition result and the second local wound recognition result of the sub-region, the wound change trend is fitted in stages to obtain the wound fitting results of each stage of the sub-region.

3. The method according to claim 2, characterized in that The wound change trend is fitted in stages based on the first local wound recognition result and the second local wound recognition result of the sub-region to obtain the wound fitting results of each stage of the sub-region, including: Acquire segmentation point information of the sub-region, and divide the first local wound recognition result and the second local wound recognition result into corresponding stages according to the segmentation point information; Based on all local wound recognition results of each stage in the sub-region, the fitting method corresponding to each stage is used to fit the wound change trend of the corresponding stage to obtain the wound fitting results of each stage in the sub-region.

4. The method according to claim 3, characterized in that The stages include: untreated stage, pre-treatment stage, mid-treatment stage, and post-treatment stage; Using a fitting method corresponding to each stage to fit the wound change trend of the corresponding stage, including at least one of the following: Linear regression was used to fit the trends of wound changes during the untreated period; Polynomial regression was used to fit the wound change trend during the early stage of treatment; Exponential regression was used to fit the wound change trend during the mid-treatment phase; A saturated growth model was used to fit the wound change trend in the later stage of treatment.

5. The method according to claim 1, characterized in that The step of identifying the wound based on the target wound image by using a deep learning model corresponding to the wound type identifier to obtain a target wound identification result includes: Preprocessing the target wound image to obtain a preprocessed image; Pre-segmenting the pre-processed image to distinguish a wound area from a non-wound area; A deep learning model corresponding to the wound type identifier is selected from the preset model library, and the features of the wound area are extracted through the deep learning model corresponding to the wound type identifier to obtain the target wound recognition result.

6. The method according to claim 5, characterized in that A deep learning model corresponding to a wound type identifier, including any of the following: A burn recognition model corresponding to the burn mark, wherein the burn recognition model is obtained by jump-connecting an encoder and a decoder; A cut injury recognition model corresponding to the cut injury identifier, wherein the cut injury recognition model adopts a masked region convolutional neural network structure; A scratch recognition model corresponding to the scratch mark, wherein the scratch recognition model adopts a lightweight convolutional neural network structure; An infected wound recognition model corresponding to the infection identifier, wherein the infected wound recognition model is a deep learning model based on a densely connected network; A composite wound recognition model corresponding to the composite identifier, wherein the composite wound recognition model is a multi-task learning model.

7. The method according to claim 1, characterized in that The target wound recognition result includes: wound area, wound area color characteristics and wound area edge characteristics; The determining of the wound analysis result based on the wound fitting result of each sub-region at each stage includes: Determining the severity of the wound based on the wound area, the wound area color characteristics and the wound area edge characteristics; The wound analysis result is determined based on the wound severity and the wound fitting results of each sub-region and each stage.

8. An automated wound analysis system based on machine vision, characterized in that: include: An acquisition module, used to acquire a target wound image and wound record information acquired by an image acquisition device, wherein the wound record information includes a wound type identifier and a historical wound recognition result; An identification module, configured to identify the wound based on the target wound image by using a deep learning model corresponding to the wound type identifier to obtain a target wound identification result; A fitting module, used to fit the wound change trend in different regions and stages based on the target wound recognition result and the historical wound recognition result, so as to obtain the wound fitting results in each sub-region and each stage; A determination module is used to determine the wound analysis result based on the wound fitting results of each sub-region at each stage.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an automated wound analysis method based on machine vision as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, an automated wound analysis method based on machine vision as described in any one of claims 1 to 7 is implemented.

Images