Image processing method, model training method and related device

Abstract

The present application discloses an image processing method, a model training method and a related device, which relate to the field of image processing technology. The method comprises: obtaining a pressure sore image to be processed; determining the actual area of ​​a preset stage in the pressure sore image to be processed; inputting the preset stage and the actual area of ​​the preset stage into a pressure sore classification model to obtain the category corresponding to the pressure sore image; the pressure sore classification model is trained based on sample clustering results, the sample clustering results are obtained by clustering analysis based on the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and the sample clustering results include the actual area of ​​the preset stage in the first sample pressure sore image and the severity category of the pressure sore corresponding to the actual area of ​​the preset stage in the first sample pressure sore image. Therefore, the method can timely give accurate classification results for the image of the pressure sore part, and improve the classification efficiency.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to an image processing method, a model training method, a device, equipment and a medium. Background Art

[0002] With the development of modern medicine, the Intensive Care Unit (ICU) has become an important department in clinical medicine. The patients monitored in the ICU are all critically ill patients, who are in a coma most of the time, such as lying in bed for a long time, and patients who lie in bed for a long time often develop pressure sores on their bodies.

[0003] Currently, doctors need to analyze the severity of the patient's pressure sores based on the images of the patient's pressure sores, and then obtain the corresponding category of the pressure sore images. This analysis process mainly depends on the doctor's clinical experience. For doctors with less clinical experience, they often cannot give accurate classification results for the pressure sore images in a timely manner, and the classification efficiency is poor. Summary of the invention

[0004] The purpose of this application is to provide an image processing method, a model training method and related devices, which can timely give accurate classification results for images of pressure sores and improve classification efficiency.

[0005] In a first aspect, the present application provides an image processing method, comprising:

[0006] Obtain images of the pressure ulcer to be treated;

[0007] Determining the actual area of ​​the preset stage in the pressure sore image to be processed;

[0008] The preset stage and the actual area of ​​the preset stage are input into the pressure sore classification model to obtain the category corresponding to the pressure sore image; the pressure sore classification model is trained based on the sample clustering results, and the sample clustering results are obtained by clustering analysis based on the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and the sample clustering results include the actual area of ​​the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image corresponding to the severity category of the pressure sore.

[0009] Optionally, determining the actual area of ​​the preset stage in the pressure sore image to be processed includes:

[0010] Determine the area of ​​each pressure sore stage in the pressure sore image to be processed, wherein the pressure sore stage includes the marker stage and the preset stage;

[0011] The actual area of ​​the preset stage is determined according to the ratio between the regional pixel area of ​​the marker stage and the actual area of ​​the marker, and the regional pixel area of ​​the preset partition.

[0012] Optionally, determining the regions of each pressure sore stage in the pressure sore image to be processed includes:

[0013] The pressure sore image to be processed is subjected to instance segmentation to obtain regions of various pressure sore stages in the pressure sore image to be processed.

[0014] Optionally, performing instance segmentation on the pressure sore image to be processed to obtain regions of each pressure sore stage in the pressure sore image to be processed includes:

[0015] The pressure sore image to be processed is input into a segmentation model to obtain regions of each pressure sore stage in the pressure sore image to be processed; the segmentation model is trained based on a second sample pressure sore image and regions of each pressure sore stage in the second sample pressure sore image.

[0016] Optionally, the method further includes:

[0017] Determining prompt information corresponding to the category of the pressure sore image according to the category of the pressure sore image and a mapping relationship between a preset category and preset prompt information;

[0018] The prompt information is displayed.

[0019] In a second aspect, the present application provides a model training method, characterized by comprising:

[0020] acquiring a first sample pressure ulcer image;

[0021] Determining an actual area of ​​a preset stage in the first sample pressure sore image;

[0022] After fusing the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image, cluster analysis is performed to obtain the severity category of the pressure sore corresponding to the actual areas of the preset stages in the first sample pressure sore image;

[0023] The actual area of ​​the preset stages with severity categories and the preset stages are used for model training to obtain a pressure sore classification model.

[0024] Optionally, determining the actual area of ​​the first sample pressure sore image includes:

[0025] acquiring a second sample pressure ulcer image;

[0026] Model training is performed using the second sample pressure sore image and the regions of each pressure sore stage in the second sample pressure sore image to obtain a segmentation model; the pressure sore stage includes a marker stage and the preset stage;

[0027] Inputting the first sample pressure sore image into the segmentation model to obtain regions of each pressure sore stage in the first sample pressure sore image;

[0028] The actual area of ​​the preset stage in the first sample pressure sore image is determined based on the ratio between the regional pixel area of ​​the marker stage in the first sample pressure sore image and the actual area of ​​the marker, and the regional pixel area of ​​the preset partition in the first sample pressure sore image.

[0029] In a third aspect, the present application provides an image processing device, comprising:

[0030] A first acquisition module is used to acquire a pressure sore image to be processed;

[0031] A first determination module is used to determine the actual area of ​​the preset stage in the pressure sore image to be processed;

[0032] A classification module is used to input the preset stage and the actual area of ​​the preset stage into a pressure sore classification model to obtain the category corresponding to the pressure sore image; the pressure sore classification model is trained based on sample clustering results, and the sample clustering results are obtained by clustering analysis based on the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and the sample clustering results include the actual area of ​​the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image corresponding to the severity category of the pressure sore.

[0033] Optionally, the first determination module is specifically used to determine the area of ​​each pressure sore stage in the pressure sore image to be processed, and the pressure sore stage includes the marker stage and the preset stage; according to the ratio between the regional pixel area of ​​the marker stage and the actual area of ​​the marker, and the regional pixel area of ​​the preset partition, the actual area of ​​the preset stage is determined.

[0034] Optionally, the first determination module is specifically configured to perform instance segmentation on the pressure sore image to be processed to obtain regions of each pressure sore stage in the pressure sore image to be processed.

[0035] Optionally, the first determination module inputs the pressure sore image to be processed into a segmentation model to obtain regions of each pressure sore stage in the pressure sore image to be processed; the segmentation model is trained based on the second sample pressure sore image and the regions of each pressure sore stage in the second sample pressure sore image.

[0036] Optionally, the device also includes a display module, which is used to determine the prompt information corresponding to the category of the pressure sore image according to the category of the pressure sore image and the mapping relationship between the preset category and the preset prompt information; and display the prompt information.

[0037] In a fourth aspect, a model training device includes:

[0038] A second acquisition module, used to acquire a first sample pressure sore image;

[0039] A second determination module is used to determine the actual area of ​​the preset stage in the first sample pressure sore image;

[0040] a clustering module, configured to perform cluster analysis after fusing the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image, so as to obtain a severity category of the pressure sore corresponding to the actual areas of the preset stages in the sample pressure sore image;

[0041] The training module is used to perform model training using the actual area of ​​the preset stages with severity categories and the preset stages to obtain a pressure sore classification model.

[0042] Optionally, the second determination module is used to obtain a second sample pressure sore image; use the second sample pressure sore image and the areas of each pressure sore stage in the second sample pressure sore image to perform model training to obtain a segmentation model; the pressure sore stage includes the marker stage and the preset stage; input the first sample pressure sore image into the segmentation model to obtain the areas of each pressure sore stage in the first sample pressure sore image; determine the actual area of ​​the preset stage in the first sample pressure sore image based on the ratio between the regional pixel area of ​​the marker stage in the first sample pressure sore image and the actual area of ​​the marker, and the regional pixel area of ​​the preset partition in the first sample pressure sore image.

[0043] In a fifth aspect, the present application provides a device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the device executes any one of the optional methods in the first aspect or the second aspect.

[0044] In a sixth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes any optional method in the first aspect or the second aspect.

[0045] In a seventh aspect, the present application provides a computer program product, which, when executed on a computer, enables the computer to execute any optional method in the first aspect or the second aspect.

[0046] It can be seen that this application has the following beneficial effects:

[0047] The present application provides an image processing method, which first fuses the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and then performs cluster analysis to obtain the pressure sore severity category corresponding to the actual area of ​​the preset stage of the first sample pressure sore image, that is, labeled data, and establishes the relationship between the pressure sore area of ​​the preset stage and the corresponding pressure sore severity category, providing a data basis for model training. Then, the actual area of ​​the preset stage of the first sample pressure sore image and the corresponding pressure sore severity category are used for model training to obtain a pressure sore classification model. Then, after obtaining the pressure sore image to be processed, the actual area of ​​the preset stage in the pressure sore image to be processed is first determined, and then the preset stage and the actual area of ​​the preset stage are input into the above-mentioned pressure sore classification model, thereby obtaining the category corresponding to the pressure sore image. Through this method, the pressure sore image can be accurately and efficiently classified without the participation of a doctor, thereby improving the classification accuracy and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of an image processing method provided in an embodiment of the present application;

[0049] Figure 2 A schematic diagram of model training and application provided in an embodiment of the present application;

[0050] Figure 3 A schematic diagram of a pressure sore image provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of a cluster analysis provided in an embodiment of the present application;

[0052] Figure 5 A flowchart of a model training method provided in an embodiment of the present application;

[0053] Figure 6 A schematic diagram of an image processing device provided in an embodiment of the present application;

[0054] Figure 7 A schematic diagram of a model training device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] The terms "first" and "second" in the embodiments of the present application are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0056] The following is an introduction to the technical terms involved in this application.

[0057] Pressure sores refer to the phenomenon of local tissue ulceration and ischemic necrosis caused by persistent ischemia, hypoxia, and malnutrition due to long-term pressure on local tissues of the human body. Pressure sores usually occur in the lumbar sacral region, bone prominences, and areas in contact with medical devices or other equipment. For patients monitored in the ICU, since patients in the ICU are in a coma for a long time, pressure sores often occur on their bodies.

[0058] At present, when doctors analyze patients' pressure sores, they often rely on their clinical experience to determine the type of pressure sores that the patient has. Doctors with less clinical experience often cannot give accurate classification results for the images of the pressure sores in a timely manner, and the classification efficiency is poor.

[0059] In view of this, an embodiment of the present application provides an image processing method, which can be applied to a processing device. The processing device can be a terminal or a server. Among them, the terminal includes but is not limited to a desktop, a laptop, a tablet computer and a smart phone. The server can be a cloud server in a cloud environment, such as a central server in a central cloud computing cluster, or an edge server in an edge cloud computing cluster. Of course, the server can also be a server in a local data center, and the so-called local data center refers to the data center to which the user belongs. It should also be noted that the server can be a single server or a server cluster formed by multiple servers. The embodiment of the present application does not specifically limit the processing device, and those skilled in the art can make a choice according to actual needs.

[0060] Specifically, the method includes: a processing device obtains a pressure sore image to be processed, determines a preset stage in the pressure sore image to be processed and an actual area of ​​the preset stage, and then inputs the actual area of ​​the preset stage into a pressure sore classification model to obtain a category corresponding to the pressure sore image. The pressure sore classification model is trained based on a sample clustering result, and the sample clustering result is obtained by clustering analysis based on the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and the sample clustering result includes the actual area of ​​the preset stage in the first sample pressure sore image and its corresponding pressure sore severity category.

[0061] In the method, the processing device first fuses the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and then performs cluster analysis to obtain the pressure sore severity category corresponding to the actual area of ​​the preset stage of the first sample pressure sore image, and establishes the relationship between the pressure sore area of ​​the preset stage and the corresponding pressure sore severity category, which provides a data basis for model training. Then, the actual area of ​​the preset stage of the first sample pressure sore image and the corresponding pressure sore severity category are used for model training to obtain a pressure sore classification model. Then, after obtaining the pressure sore image to be processed, the actual area of ​​the preset stage in the pressure sore image to be processed is first determined, and then the preset stage and the actual area of ​​the preset stage are input into the above-mentioned pressure sore classification model, and then the category corresponding to the pressure sore image is obtained. Through this method, the pressure sore image can be accurately and efficiently classified without the participation of a doctor, thereby improving the classification accuracy and efficiency.

[0062] In order to make the technical solution of the present application clearer and easier to understand, the image processing method provided by the embodiment of the present application is introduced below in conjunction with the accompanying drawings. Figure 1 As shown in FIG. 1 , the figure is a flow chart of an image processing method provided in an embodiment of the present application, the method comprising:

[0063] S101. A processing device obtains a pressure sore image to be processed.

[0064] The pressure sore image refers to an image of a pressure sore site on the body of a user (eg, a patient). The pressure sore image generally includes a marker of a preset area for subsequent calculation of the actual area of ​​each pressure sore stage in the pressure sore image, which will be described later.

[0065] Based on this, before photographing the area where the pressure sore appears on the user's body, it is necessary to place the above-mentioned marker at a position on the same plane as the pressure sore area. The marker can be a marker of a fixed shape or a marker of a non-fixed shape. The embodiment of the present application does not limit the shape of the marker, and those skilled in the art can select the shape of the marker based on actual needs.

[0066] The image acquisition device then acquires the pressure sore site and the markers on the user's body to obtain a pressure sore image. In some examples, the image acquisition device may be a device independent of the processing device. In this case, the image acquisition device may be a mobile camera device, such as a mobile phone, a camera, etc. After the image acquisition device acquires the pressure sore image, it may transmit the pressure sore image to the processing device via a network connection (such as the Internet, Bluetooth, etc.), and the processing device may acquire the pressure sore image to be processed.

[0067] In other examples, the image acquisition device may be a sub-device integrated on the processing device, the image acquisition device may be a movable camera, etc. The processing device uses the image acquisition device to capture the locations of pressure sores and markers on the user's body, and thereby obtain the pressure sore image to be processed.

[0068] It should be noted that the above is merely an exemplary introduction to the processing device obtaining the pressure ulcer images to be processed. Those skilled in the art may choose other methods to obtain the pressure ulcer images to be processed based on actual needs. For example, the processing device may receive the pressure ulcer images to be processed sent from other hospitals.

[0069] S102: The processing device determines the actual area of ​​the preset stage in the pressure sore image to be processed.

[0070] The pressure sore staging in the pressure sore image can be divided into preset staging and marker staging. Among them, the preset staging can generally be divided into stage one, stage two, stage three, stage four and unstaging in medicine. Stage one is a milder pressure sore, stage two is more serious than stage one, stage four is the most serious, and unstaging refers to the undetermined staging. Marker staging refers to the marker in the pressure sore image, which is used to determine the pixel area of ​​the marker in the image.

[0071] After the processing device determines the preset stage in the pressure sore image to be processed, it can calculate the pixel area of ​​the region of the preset stage, and then calculate the actual area of ​​the preset stage.

[0072] In some embodiments, the processing device may first determine the regions of each pressure sore stage in the pressure sore image to be processed, such as the region of the above-mentioned marker stage and the region of the preset stage, and then calculate the regional pixel area of ​​the above-mentioned marker stage, the regional pixel area of ​​the first stage, the regional pixel area of ​​the second stage, the regional pixel area of ​​the third stage, the regional pixel area of ​​the fourth stage, and the pixel area of ​​the non-staged region in the pressure sore image, and then calculate the ratio between the regional pixel area of ​​the marker stage and the actual area of ​​the marker, and use the ratio and the regional pixel area of ​​each stage in the above-mentioned preset stage to calculate the actual area of ​​each stage in the above-mentioned preset stage. For example, the processing device may determine the actual area of ​​each stage in the preset stage by the following formula:

[0073]

[0074] Among them, S1 is the regional pixel area of ​​the first stage, S′1 is the actual area of ​​the first stage, S2 is the regional pixel area of ​​the second stage, S′2 is the actual area of ​​the second stage, S3 is the regional pixel area of ​​the third stage, S′3 is the actual area of ​​the third stage, S4 is the regional pixel area of ​​the fourth stage, S′4 is the actual area of ​​the fourth stage, S5 is the regional pixel area of ​​the non-stageable stage, S′5 is the actual area of ​​the fifth stage, S6 is the regional pixel area of ​​the marker stage, S′6 is the actual area of ​​the marker stage, and S′6 is a known quantity.

[0075] Based on the above formula (1), the processing equipment can calculate as well as The actual area of ​​each stage in the preset stage is then obtained. It can be seen from the above formula (1) that the processing device does not rely on the shape of the marker in the process of calculating the actual area of ​​each stage in the preset stage, thereby providing more choices for the shape of the marker, thereby avoiding the selection of a marker of a specified shape (such as a circle), which results in the marker being similar to the shape of the pressure sore and causing the problem of misidentification.

[0076] The following describes how the processing device determines the regions of each pressure sore stage in the pressure sore image to be processed.

[0077] In some embodiments, the processing device may perform instance segmentation on the pressure sore image to be processed to obtain regions of various pressure sore stages in the pressure sore image to be processed. Instance segmentation refers to segmenting different instances from the pressure sore image using a target detection method, that is, segmenting regions of different pressure sore stages in the pressure sore image.

[0078] Based on this, the processing device can use the second sample pressure sore image and the areas of each pressure sore stage in the second sample pressure sore image to perform model training, and then obtain a segmentation model. Among them, algorithms such as mask r-cnn can be used in the model training process. Then the processing device can use the trained segmentation model to segment the pressure sore image to be processed, and then obtain the areas of each pressure sore stage in the pressure sore image to be processed.

[0079] like Figure 2As shown, this figure is a schematic diagram of a model training and application provided by an embodiment of the present application. Among them, image data processing refers to pixel-level annotation of the second sample pressure sore image, and staging the second sample pressure sore image. As mentioned above, it can generally be divided into stage one, stage two, stage three, stage four and unstaging. The present application further adds marker staging on this basis; the establishment and division of the data set refers to constructing a data set of the second sample pressure sore image, and dividing the data set of the second sample pressure sore image into a training set, a validation set and a test set according to a preset ratio. For example, the data set of the second sample pressure sore image is divided into a training set, a validation set and a test set according to a ratio of 6:1:3.

[0080] Model training refers to using the training set and the validation set to train the model and obtain a segmentation model; model testing refers to using the test set to test the accuracy and robustness of the trained segmentation model; model application refers to using the tested segmentation model to segment new images (such as pressure ulcer images to be processed).

[0081] Among them, Dropout refers to the use of the Dropout strategy to prevent model overfitting during model training, BN refers to the use of the BN strategy to prevent the gradient disappearance problem during the back propagation of the model gradient, cross-validation refers to the verification strategy used in the training process; learning self-attenuation refers to the learning strategy used in the training process; mean average precision (MAP) and intersection over union (IOU) values ​​are model evaluation indicators.

[0082] like Figure 3 As shown in the figure, this figure is a schematic diagram of a pressure sore image provided by an embodiment of the present application. It can be seen from the figure that both region 310 and region 320 represent pressure sores, wherein the severity of the pressure sores in region 310 is higher than the severity of the pressure sores in region 320. In traditional solutions, generally only the actual area of ​​region 310, where the severity of the pressure sores is more serious, is calculated, while the pressure sores in region 320 are ignored, which in turn affects the subsequent classification results. In the present application, after the above-mentioned pressure sore image is input into the segmentation model, the above-mentioned region 310 and region 320 can be automatically segmented, and then the actual area of ​​region 310 and the actual area of ​​region 320 are calculated.

[0083] S103: The processing device inputs the preset stages and the actual areas of the preset stages into the pressure sore classification model to obtain the category corresponding to the pressure sore image.

[0084] In some examples, the preset stages of the pressure sore image can be represented by Y = [y1, y2, y3, y4, y5], where y1, y2, y3, y4, y5 respectively represent whether the pressure sores of stage 1, stage 2, stage 3, stage 4, and unstageable exist. Taking y1 as an example, y1 = 0 indicates that stage 1 does not exist, and y1 = 1 indicates that stage 1 exists. For example, taking pressure sore image A as an example, Y A =[0, 1, 0, 1, 0] indicates that the pressure sore image A includes stage II and stage IV pressure sores but does not include stage I, stage III and pressure sores that cannot be staged.

[0085] Continuing with the above example, the actual area can be represented by S′=[s′1, s′2, s′3, s4′, s′5], where s′1, s′2, s′3, s′4, s′5 represent the actual areas of stage 1, stage 2, stage 3, stage 4, and stage-indistinguishable pressure ulcers, respectively. Among them, s′1, s′2, s′3, s′4, s′5 can be normalized values.

[0086] Based on the above Y and S′, we can know that if y i = 0, then s′ i =0; if y i =1, then s′ i >0, that is, s′ i =s′ i , where i can be 1, 2, 3, 4, or 5. Based on this, the above Y and S′ can be fused to obtain X=[x1, x2, x3, x4, x5], and X is used to represent the preset stages and the actual areas of the preset stages. Taking x1 as an example, x1=0 represents that there is no stage 1 pressure sore, and x1≠0 represents that there is a stage 1 pressure sore, and the actual area of ​​the stage 1 pressure sore is the value of x1.

[0087] The pressure sore classification model is used to determine the category corresponding to the pressure sore image based on the actual area of ​​the preset stage of the pressure sore image. The training process of the pressure sore classification model is introduced below. The processing device can first obtain a first sample pressure sore image, and then determine the actual area of ​​the preset stage in the sample pressure sore image. For example, the processing device can obtain the area of ​​each pressure sore stage in the first sample pressure sore image through the above-mentioned segmentation model, and then determine the actual area of ​​the preset stage in the first sample pressure sore image based on the ratio between the regional pixel area of ​​the marker stage in the first sample pressure sore image and the actual area of ​​the marker, as well as the regional pixel area of ​​the preset partition in the first sample pressure sore image.

[0088] Then the processing device can characterize the actual area of ​​the preset stage in the first sample pressure sore image by the above X to digitize it. Then the processing device performs cluster analysis on the actual area of ​​the preset stage in the first sample pressure sore image to obtain the severity category of the pressure sore corresponding to the actual area of ​​the preset stage in the first sample pressure sore image.

[0089] like Figure 4 As shown, this figure is a schematic diagram of a cluster analysis provided by an embodiment of the present application. The selection of the K value refers to presetting a number of pressure sore severity categories, for example, K can be 4, K can also be =5, and the embodiment of the present application does not specifically limit the value of K. Those skilled in the art can make a selection based on actual needs. In some examples, K can be an empirical value.

[0090] Unsupervised learning means classifying data with similar features into one category through unsupervised learning, so as to obtain data of each severity level. That is to say, a severity category label is generated for each piece of data through unsupervised learning. The unsupervised learning algorithm can be a clustering algorithm or other related algorithms.

[0091] K value evaluation refers to whether the K-type results obtained after unsupervised learning can meet the requirements. The user can evaluate the K-type results obtained based on prior knowledge. If they cannot meet the user's requirements, the user is required to reset the K value based on the results of this learning and prior knowledge until the results of unsupervised learning meet the requirements. At this time, data with severity categories are output, such as the actual area of ​​preset stages with severity categories, which provides a data basis for subsequent model training.

[0092] Next, the processing device can perform model training based on the actual area of ​​the preset stages with severity categories to obtain a pressure sore classification model. In the model training process, a support vector machine (SVM) or other related algorithms can be used to learn the data distribution.

[0093] After obtaining the above pressure sore classification model, the processing device can input the actual area of ​​the preset stage in the pressure sore image to be processed into the pressure sore classification model to obtain the category corresponding to the pressure sore image. Through this method, the processing device can automatically identify the category of the pressure sore corresponding to the pressure sore image to be processed, and can accurately and efficiently classify the pressure sore image without the participation of a doctor, thereby improving the classification accuracy and efficiency.

[0094] The image processing method provided in the embodiment of the present application can perform intelligent analysis on the pressure sore image, automatically obtain the area of ​​each pressure sore stage of the pressure sore image, and accurately calculate the actual area of ​​each pressure sore stage, that is, realize the intelligent and automated preprocessing of the pressure sore image, and provide accurate data support for subsequent care, treatment and other work. The present application also digitizes the actual area of ​​the preset area in the above-mentioned pressure sore image, and then performs a cluster analysis to obtain the pressure sore severity category corresponding to the preset stage of the pressure sore image and the actual area of ​​the preset stage, and establishes the relationship between the pressure sore area of ​​the preset stage and the corresponding pressure sore severity category, which provides a data basis for subsequent model training. Then, the pressure sore classification model obtained after training is used to classify the pressure sore image to be processed, and the category of the pressure sore corresponding to the pressure sore image is obtained, thereby improving the classification accuracy and efficiency.

[0095] The embodiment of the present application also provides a model training method, which can be applied to a training device, which can be the above-mentioned processing device or not. When the training device is not the above-mentioned processing device, after completing the model training, the training device can deploy the trained model to the above-mentioned processing device.

[0096] like Figure 5 As shown, the figure is a flow chart of a model training method provided in an embodiment of the present application, the method comprising:

[0097] S501. The training device obtains a first sample pressure ulcer image.

[0098] S502: The training device determines the actual area of ​​the preset stage in the first sample pressure sore image.

[0099] In some examples, the training device may first train a segmentation model, then input the first sample pressure sore image into the segmentation model to obtain the areas of each pressure sore stage in the first sample pressure sore image, and then calculate the actual area of ​​the preset stage in the first sample pressure sore image.

[0100] Specifically, the training device obtains the second sample pressure sore image, uses the second sample pressure sore image and the regions of each pressure sore stage in the second sample pressure sore image to perform model training, and obtains a segmentation model, and the pressure sore stage includes the marker stage and the preset stage. Then the first sample pressure sore image is input into the segmentation model to obtain the regions of each pressure sore stage in the first sample pressure sore image. Next, based on the ratio between the regional pixel area of ​​the marker stage in the first sample pressure sore image and the actual area of ​​the marker, and the regional pixel area of ​​the preset partition in the first sample pressure sore image, the actual area of ​​the preset stage in the first sample pressure sore image is determined. The specific process can be found in the introduction of the above embodiment, and will not be repeated here.

[0101] S503: After fusing the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image, the training device performs cluster analysis to obtain the severity category of the pressure sore corresponding to the actual areas of the preset stages in the first sample pressure sore image.

[0102] After the training device fuses the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image (i.e., obtaining the above X), it performs cluster analysis to obtain the severity category of the pressure sore corresponding to the actual area of ​​the preset stage in the first sample pressure sore image. In this way, the relationship between the actual area of ​​the preset stage in the pressure sore image and the severity category of the pressure sore is established, providing a data basis for subsequent model training. The specific process can be referred to the introduction in the above embodiment, which will not be repeated here.

[0103] S504: The training device performs model training using the actual area of ​​the preset stages with severity categories and the preset stages to obtain a pressure sore classification model.

[0104] The specific process of model training can be found in the introduction in the above embodiment and will not be repeated here.

[0105] Based on the above description, the embodiment of the present application provides an image processing method and a model training method. First, after the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image are fused, a cluster analysis is performed to obtain the pressure sore severity category corresponding to the actual area of ​​the preset stage of the first sample pressure sore image, and the relationship between the pressure sore area of ​​the preset stage and the corresponding pressure sore severity category is established, which provides a data basis for model training. Then, the model training is performed using the actual area of ​​the preset stage of the first sample pressure sore image corresponding to the pressure sore severity category to obtain a pressure sore classification model. Then, after obtaining the pressure sore image to be processed, the actual area of ​​the preset stage in the pressure sore image to be processed is first determined, and then the preset stage and the actual area of ​​the preset stage are input into the above-mentioned pressure sore classification model, thereby obtaining the category corresponding to the pressure sore image. The pressure sore classification model is obtained by the model training method, and the pressure sore image is processed by the image processing method, so that the pressure sore image can be automatically classified, and the pressure sore image can be accurately and efficiently classified without the participation of a doctor, thereby improving the classification accuracy and efficiency.

[0106] The present application also provides an image processing device, such as Figure 6 As shown in FIG. 1 , the figure is a schematic diagram of an image processing device provided in an embodiment of the present application, the device comprising:

[0107] A first acquisition module 601 is used to acquire a pressure sore image to be processed;

[0108] A first determination module 602 is used to determine the actual area of ​​the preset stage in the pressure sore image to be processed;

[0109] The classification module 603 is used to input the preset stage and the actual area of ​​the preset stage into the pressure sore classification model to obtain the category corresponding to the pressure sore image; the pressure sore classification model is trained based on the sample clustering results, and the sample clustering results are obtained by clustering analysis based on the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and the sample clustering results include the actual area of ​​the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image corresponding to the severity category of the pressure sore.

[0110] Optionally, the first determination module 602 is specifically used to determine the area of ​​each pressure sore stage in the pressure sore image to be processed, and the pressure sore stage includes the marker stage and the preset stage; according to the ratio between the regional pixel area of ​​the marker stage and the actual area of ​​the marker, and the regional pixel area of ​​the preset partition, the actual area of ​​the preset stage is determined.

[0111] Optionally, the first determination module 602 is specifically configured to perform instance segmentation on the pressure sore image to be processed to obtain regions of each pressure sore stage in the pressure sore image to be processed.

[0112] Optionally, the first determination module 602 inputs the pressure sore image to be processed into a segmentation model to obtain regions of each pressure sore stage in the pressure sore image to be processed; the segmentation model is trained based on the second sample pressure sore image and the regions of each pressure sore stage in the second sample pressure sore image.

[0113] Optionally, the device also includes a display module, which is used to determine the prompt information corresponding to the category of the pressure sore image according to the category of the pressure sore image and the mapping relationship between the preset category and the preset prompt information; and display the prompt information.

[0114] Based on the above description, the device can perform intelligent analysis on the pressure sore image, automatically obtain the area of ​​each pressure sore stage of the pressure sore image, and accurately calculate the actual area of ​​each pressure sore stage, that is, it realizes the intelligent and automated preprocessing of the pressure sore image, and provides accurate data support for subsequent care, treatment and other work. The device also digitizes the actual area of ​​the preset area in the above pressure sore image, and then performs cluster analysis to obtain the pressure sore severity category corresponding to the preset stage of the pressure sore image and the actual area of ​​the preset stage, and establishes the relationship between the pressure sore area of ​​the preset stage and the corresponding pressure sore severity category, which provides a data basis for subsequent model training. Then, the pressure sore classification model obtained after training is used to classify the pressure sore image to be processed, and the category of the pressure sore corresponding to the pressure sore image is obtained, thereby improving the classification accuracy and efficiency.

[0115] The present application also provides a model training device, such as Figure 7 As shown, a schematic diagram of the model training device, the device includes:

[0116] The second acquisition module 701 is used to acquire a first sample pressure sore image;

[0117] A second determination module 702 is used to determine the actual area of ​​the preset stage in the first sample pressure sore image;

[0118] The clustering module 703 is used to perform cluster analysis after fusing the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image, so as to obtain the severity category of the pressure sore corresponding to the actual areas of the preset stages in the sample pressure sore image;

[0119] The training module 704 is used to perform model training using the actual area of ​​the preset stages with severity categories and the preset stages to obtain a pressure sore classification model.

[0120] Optionally, the second determination module 702 is used to obtain a second sample pressure sore image; use the second sample pressure sore image and the areas of each pressure sore stage in the second sample pressure sore image to perform model training to obtain a segmentation model; the pressure sore stage includes the marker stage and the preset stage; input the first sample pressure sore image into the segmentation model to obtain the areas of each pressure sore stage in the first sample pressure sore image; determine the actual area of ​​the preset stage in the first sample pressure sore image based on the ratio between the regional pixel area of ​​the marker stage in the first sample pressure sore image and the actual area of ​​the marker, and the regional pixel area of ​​the preset partition in the first sample pressure sore image.

[0121] Based on the above description, the device first fuses the preset stage in the first sample pressure sore image and the actual area of ​​the preset stage in the first sample pressure sore image, and then performs cluster analysis to obtain the pressure sore severity category corresponding to the actual area of ​​the preset stage of the first sample pressure sore image, and establishes the relationship between the pressure sore area of ​​the preset stage and the corresponding pressure sore severity category, which provides a data basis for model training. Then, the model training is performed using the actual area of ​​the preset stage of the first sample pressure sore image corresponding to the severity category of the pressure sore to obtain a pressure sore classification model. Then, after obtaining the pressure sore image to be processed, the actual area of ​​the preset stage in the pressure sore image to be processed is first determined, and then the preset stage and the actual area of ​​the preset stage are input into the above pressure sore classification model, thereby obtaining the category corresponding to the pressure sore image. The pressure sore classification model is obtained by the model training method, and the pressure sore image is processed by the image processing method, so that the pressure sore image can be automatically classified, and the pressure sore image can be accurately and efficiently classified without the participation of a doctor, thereby improving the classification accuracy and efficiency.

[0122] In addition, an embodiment of the present application further provides a computing device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described in any one of the above embodiments is implemented.

[0123] In addition, an embodiment of the present application further provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes the method described in any of the above embodiments.

[0124] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the system or device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0125] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0126] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0127] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0128] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An image processing method, characterized in that: include: Obtain images of the pressure ulcer to be treated; Determining the regions of each pressure sore stage in the pressure sore image to be processed, wherein the pressure sore stages include marker stages and preset stages; Determine the actual area of ​​the preset stage according to the ratio between the pixel area of ​​the region of the marker stage and the actual area of ​​the marker, and the pixel area of ​​the region of the preset stage; Inputting the preset stages and the actual areas of the preset stages into a pressure sore classification model to obtain a category corresponding to the pressure sore image; The pressure sore classification model is trained based on sample clustering results, and the sample clustering results are obtained by clustering analysis based on preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image. The sample clustering results include the severity categories of pressure sores corresponding to the actual areas of the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image.

2. The method according to claim 1, characterized in that The step of determining the regions of each pressure sore stage in the pressure sore image to be processed includes: The pressure sore image to be processed is subjected to instance segmentation to obtain regions of various pressure sore stages in the pressure sore image to be processed.

3. The method according to claim 2, characterized in that The performing instance segmentation on the pressure sore image to be processed to obtain regions of each pressure sore stage in the pressure sore image to be processed includes: The pressure sore image to be processed is input into a segmentation model to obtain regions of each pressure sore stage in the pressure sore image to be processed; the segmentation model is trained based on a second sample pressure sore image and regions of each pressure sore stage in the second sample pressure sore image.

4. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: Determining prompt information corresponding to the category of the pressure sore image according to the category of the pressure sore image and a mapping relationship between a preset category and preset prompt information; The prompt information is displayed.

5. A model training method, characterized in that: include: acquiring a first sample pressure ulcer image; acquiring a second sample pressure ulcer image; Model training is performed using the second sample pressure sore image and the regions of each pressure sore stage in the second sample pressure sore image to obtain a segmentation model; the pressure sore stage includes a marker stage and a preset stage; Inputting the first sample pressure sore image into the segmentation model to obtain regions of each pressure sore stage in the first sample pressure sore image; Determine the actual area of ​​the preset stage in the first sample pressure sore image according to the ratio between the pixel area of ​​the region of the marker stage in the first sample pressure sore image and the actual area of ​​the marker, and the pixel area of ​​the region of the preset stage in the first sample pressure sore image; After fusing the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image, cluster analysis is performed to obtain the severity category of the pressure sore corresponding to the actual areas of the preset stages in the first sample pressure sore image; The actual area of ​​the preset stages with severity categories and the preset stages are used for model training to obtain a pressure sore classification model.

6. An image processing device, characterized in that: include: A first acquisition module is used to acquire a pressure sore image to be processed; A first determination module is used to determine the area of ​​each pressure sore stage in the pressure sore image to be processed, wherein the pressure sore stage includes a marker stage and a preset stage; according to the ratio between the pixel area of ​​the area of ​​the marker stage and the actual area of ​​the marker, and the pixel area of ​​the area of ​​the preset stage, determine the actual area of ​​the preset stage; A classification module, used for inputting the preset stages and the actual areas of the preset stages into a pressure sore classification model to obtain a category corresponding to the pressure sore image; The pressure sore classification model is trained based on sample clustering results, and the sample clustering results are obtained by clustering analysis based on preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image. The sample clustering results include the severity categories of pressure sores corresponding to the actual areas of the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image.

7. A model training device, characterized in that: include: A second acquisition module, used to acquire a first sample pressure sore image; A second determination module is used to obtain a second sample pressure sore image; The second sample pressure sore image and the regions of each pressure sore stage in the second sample pressure sore image are used for model training to obtain a segmentation model; the pressure sore stage includes a marker stage and a preset stage; the first sample pressure sore image is input into the segmentation model to obtain the regions of each pressure sore stage in the first sample pressure sore image; according to the ratio between the pixel area of ​​the region of the marker stage in the first sample pressure sore image and the actual area of ​​the marker, and the pixel area of ​​the region of the preset stage in the first sample pressure sore image, the actual area of ​​the preset stage in the first sample pressure sore image is determined; a clustering module, configured to perform cluster analysis after fusing the preset stages in the first sample pressure sore image and the actual areas of the preset stages in the first sample pressure sore image, so as to obtain a severity category of the pressure sore corresponding to the actual areas of the preset stages in the sample pressure sore image; The training module is used to perform model training using the actual area of ​​the preset stages with severity categories and the preset stages to obtain a pressure sore classification model.

8. A device, characterized in that include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes the method according to any one of claims 1 to 5.

Images