Method and system for identifying target tissue in image, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a target tissue identification method and system in an image, electronic equipment and a storage medium. The method comprises the following steps: acquiring a to-be-recognized image, wherein the to-be-recognized image is a section image of a target part of a shot object; the to-be-recognized image is input into the first model, a first result is output, and the first result comprises position information of various tissues contained in the to-be-recognized image; and according to the type of the section corresponding to the to-be-identified image, the position information of the target tissue in the to-be-identified image is screened out from the first result, and if the types of the sections are different, the types of the target tissue included in the sections are different. According to the scheme, the recognition accuracy is high, the consumed calculation amount is remarkably reduced, and the recognition efficiency is also remarkably improved. In addition, the expansibility of the scheme is high, and the recognition effect on various types of target organizations is good. Therefore, the user experience is better.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular, to a method for identifying target tissue in an image, a system for identifying target tissue in an image, an electronic device and a storage medium. Background Art

[0002] In recent years, image recognition technology has been widely used in various fields. For example, in the medical field, biomedical imaging technologies such as X-rays, computer-assisted tomography (CT), ultrasound, etc. are usually used to obtain images of the target part of the patient. Then, recognition algorithms or deep learning models are used to obtain the location of the target tissue to be identified in the image. Finally, auxiliary diagnosis and treatment of the disease are performed based on the location of the identified target tissue.

[0003] The method of using deep learning models to identify fixed types of target tissues in images has many advantages, such as high recognition efficiency and good generalization ability. However, when the types of target tissues to be identified are not fixed, the accuracy of using a deep learning model to identify multiple different images is greatly reduced.

[0004] Take the identification of musculoskeletal tissues in the main joints of the human body, such as the shoulder joint, ankle joint, wrist joint, elbow joint, knee joint and hip joint, as an example. In order to accurately obtain the true state of various musculoskeletal tissues in the patient's target joint, it is necessary to scan multiple section images of the patient's target joint, and then assist in diagnosis and treatment based on the morphology of different musculoskeletal tissues in each section image. Since the tissue morphology, characteristics and relative positions of different sections of the same joint are similar, there are some musculoskeletal tissues that need to be identified on some sections, but not on other sections. In other words, for different section images, the target tissues expected to be identified are different. Since it is impossible to distinguish different section images using a single deep learning model, most of the existing technologies use multiple recognition models for recognition, which consumes a huge amount of calculation and is difficult to implement. Summary of the invention

[0005] In order to at least partially solve the problems existing in the prior art, according to a first aspect of the present application, a method for identifying a target tissue in an image is provided, comprising:

[0006] Acquire an image to be identified, wherein the image to be identified is a cross-sectional image of a target part of the photographed object;

[0007] Inputting the image to be identified into the first model and outputting a first result, wherein the first result includes position information of various tissues contained in the image to be identified; and

[0008] According to the type of the section corresponding to the image to be identified, the position information of the target tissue in the image to be identified is filtered out from the first result, wherein different types of sections have different categories of target tissue included in the sections.

[0009] Exemplarily, the first model is a segmentation model,

[0010] Inputting the image to be recognized into the first model and outputting the first result includes:

[0011] Inputting the image to be identified into the segmentation model, and outputting a first number of first segmentation maps, wherein the first number is equal to the total number of tissue categories that can be identified by the segmentation model, wherein each first segmentation map is a segmentation map for a category of tissue, and different first segmentation maps correspond to different categories of tissue;

[0012] According to the type of the section corresponding to the image to be identified, the location information of the target tissue in the image to be identified is screened out from the first result, including:

[0013] Determining, according to the type of the section corresponding to the image to be identified, category information of the target tissue included in the section; and

[0014] According to the first number of first segmentation graphs and the category information, position information of the target tissue in the image to be identified is determined.

[0015] Exemplarily, the first number of first segmentation maps is represented by a first tensor,

[0016] According to the type of the section corresponding to the image to be identified, the category information of the target tissue included in the section is determined, including:

[0017] Determine, according to the type of the section corresponding to the image to be identified, a first array consisting of a first number of elements, wherein each element in the first array corresponds one-to-one to each type of tissue, and the value of the element corresponding to the first tissue in the first array is a non-zero number, and the values ​​of the remaining elements are 0, and the first tissue is the target tissue in the section corresponding to the image to be identified;

[0018] Determining the location information of the target tissue in the image to be identified according to the first number of first segmentation maps and the category information includes:

[0019] multiply each layer in the first tensor by the corresponding element in the first array to obtain a second tensor; and

[0020] Determine the position of the target tissue in the image to be identified according to the non-zero elements in the second tensor;

[0021] The i-th layer in the first tensor corresponds to the i-th element in the first array, and both correspond to the same category of tissues, and i is greater than or equal to 1 and less than or equal to the first number.

[0022] Exemplarily, before filtering out the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified, the method further includes:

[0023] The image to be identified is input into the second model, and the type information of the section corresponding to the image to be identified is output.

[0024] Exemplarily, the method further comprises:

[0025] Acquire annotation data of the sample image, wherein the annotation data includes real position information of various target tissues in the sample image;

[0026] Inputting the sample image into the first initial model and outputting a first initial result, wherein the first initial result includes position information of various tissues contained in the sample image;

[0027] According to the type of the section corresponding to the sample image, the predicted position information of various tissues in the sample image is screened out from the first initial result;

[0028] According to the difference between the actual position information and the predicted position information, the first initial model is trained to obtain a first model.

[0029] Exemplarily, the first initial result is a first number of second segmentation maps, wherein each second segmentation map is a segmentation map for a category of tissue, and different second segmentation maps correspond to different categories of tissue;

[0030] According to the type of the section corresponding to the sample image, the predicted position information of various tissues in the sample image is screened out from the first initial result, including:

[0031] Mapping the type information of the section corresponding to each sample image to the category information of the target tissue included in the section; and

[0032] According to the first number of second segmentation maps and the category information, predicted position information of the target tissue in the sample image is determined.

[0033] Exemplarily, obtaining the annotation data of the sample image includes:

[0034] For each sample image, obtaining annotation information of the sample image; and

[0035] Converting the sample image into a first number of labeled images according to the labeling information;

[0036] Among them, each annotated image corresponds one-to-one to each category of tissue, and in the annotated image corresponding to the first tissue, the pixel value of the pixel corresponding to the position of the first tissue is different from the pixel values ​​of other pixels; in the annotated image where the corresponding tissue does not belong to the first tissue, the pixel values ​​of each pixel are the same, and the first tissue is the target tissue in the section corresponding to the sample image.

[0037] Exemplarily, the first model is a semantic segmentation model or an instance segmentation model.

[0038] Exemplarily, the image to be identified is an ultrasound image, and the tissue is musculoskeletal tissue.

[0039] According to another aspect of the present application, a system for identifying target tissue in an image is also provided, comprising:

[0040] An acquisition module, used for acquiring an image to be identified, wherein the image to be identified is a cross-sectional image of a target part of a photographed object;

[0041] a recognition module, configured to input the image to be recognized into a first model and output a first result, wherein the first result includes position information of various tissues contained in the image to be recognized; and

[0042] The screening module is used to screen the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified, wherein different types of sections have different categories of target tissue included in the sections.

[0043] According to another aspect of the present application, an electronic device is also provided, including a processor and a memory, wherein the memory stores computer program instructions, and the computer program instructions are used to execute the above-mentioned method for identifying target tissue in an image when the processor is executed.

[0044] According to another aspect of the present application, a storage medium is further provided, on which program instructions are stored, and the program instructions are used to execute the above-mentioned method for identifying target tissue in an image when running.

[0045] As mentioned above, the method of using multiple recognition models to identify target tissues in different section images in the prior art consumes a large amount of computation and time cost, and is also difficult to implement. However, the above method in the embodiment of the present application uses a recognition model to identify the tissue contained in the image to be identified, and then accurately screens out the location information of the target tissue to be identified from the recognition result according to the section type corresponding to the image. This solution not only has a high recognition accuracy, but also significantly reduces the amount of computation consumed and significantly improves the recognition efficiency. In addition, the scalability of the solution is also strong, and the recognition effect for various types of target tissues is good. Therefore, the user experience is better.

[0046] A series of simplified concepts are introduced in the invention content, which will be further described in detail in the detailed description. The content of this application does not mean to attempt to define the key features and essential technical features of the technical solution claimed for protection, nor does it mean to attempt to determine the scope of protection of the technical solution claimed for protection.

[0047] The advantages and features of the present application are described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The following drawings of the present application are used as part of the present application for understanding the present application. The drawings show the implementation methods of the present application and their descriptions, and are used to explain the principles of the present application. In the drawings,

[0049] Figure 1 A schematic flow chart showing a method for identifying target tissue in an image according to an embodiment of the present application;

[0050] Figure 2 and Figure 3 Different parts of a flowchart of a method for identifying a target tissue in an image according to another embodiment of the present application are respectively shown;

[0051] Figure 4 A schematic block diagram showing a target tissue recognition system in an image according to an embodiment of the present application; and

[0052] Figure 5 A schematic block diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0053] In the following description, a large amount of details are provided so that the present application can be thoroughly understood. However, it will be appreciated by those skilled in the art that the following description only exemplarily illustrates the preferred embodiments of the present application, and the present application can be implemented without one or more such details. In addition, in order to avoid confusion with the present application, some technical features well known in the art are not described in detail.

[0054] In order to at least partially solve the above technical problems, according to one aspect of the present application, a method for identifying target tissue in an image is provided. This method can effectively solve the problem of low recognition accuracy caused by different recognition requirements of similar tissues in images of different sections, and achieves the effect of "differential recognition" of target tissues in images of different types of sections through a model, with high recognition accuracy and efficiency.

[0055] Figure 1 A schematic flow chart of a method 100 for identifying target tissue in an image according to an embodiment of the present application is shown. As shown in the figure, the method 100 includes step S120, step S140 and step S160.

[0056] Step S120, obtaining an image to be identified, wherein the image to be identified is a cross-sectional image of a target part of the photographed object.

[0057] The subject can be any suitable object, including but not limited to human body, animal or still life, etc., which is not limited by the present application. The target part can also be any part that is desired to be identified, which is not limited by the present application. Exemplarily, the subject is a patient, and the target part can be any injured part or lesion part of the patient. Specifically, the target part can be a joint part, such as a shoulder joint. Alternatively, the target part can also be multiple joint parts, such as a shoulder joint and an elbow joint. The section of the target part can be arbitrarily set according to actual needs, including but not limited to a sagittal plane, a coronal plane, a horizontal plane, etc.

[0058] According to the embodiments of the present application, any existing or future developed image acquisition method can be used to acquire the image to be identified. The image to be identified can be the original image directly acquired by the image acquisition device, or it can be the image after the original image is preprocessed. The preprocessing operation may include all operations to improve the visual effect of the image, improve the clarity of the image, or highlight certain features in the image to facilitate the beautification of the target. For example, the preprocessing operation may include denoising operations such as filtering, and may also include adjustments to image parameters such as image grayscale, contrast, brightness, etc., and may also include operations such as image cropping or image scaling. The image to be identified can be any suitable image. It can be a two-dimensional image or a three-dimensional image; it can be a black and white image or a color image; it can be an image of any size, or it can be an image that meets the preset resolution requirements.

[0059] The following is an example in which the subject is a patient and the target part is the shoulder joint of the patient, and the implementation of the method 100 is explained. In this example, medical imaging equipment such as medical X-ray machines, digital imaging equipment, X-ray computer tomography equipment, magnetic resonance imaging equipment, ultrasonic imaging equipment, nuclear medicine imaging equipment, etc. can be used to obtain images of various sections of the shoulder joint to obtain images to be identified. For example, the image to be identified is an ultrasonic image, and an ultrasonic probe can be used to scan different positions of the human shoulder joint from different angles to obtain images of various sections of the shoulder joint as images to be identified. Exemplarily and not restrictively, the images of these sections may include images of the short-axis section of the long head of the biceps brachii tendon, images of the long-axis section of the long head of the biceps brachii tendon, images of the short-axis section of the subscapularis tendon, images of the long-axis section of the subscapularis tendon, images of the short-axis section of the supraspinatus tendon, and images of the long-axis section of the supraspinatus tendon, etc., 8 different sections.

[0060] Step S140: input the image to be identified into the first model and output a first result, wherein the first result includes the position information of various tissues contained in the image to be identified.

[0061] It can be understood that in each section image of the human shoulder joint obtained above, the imaging area of ​​each muscle tissue and bone tissue at the position that can be detected by the ultrasonic signal can be included. In this step, the first model can be used to identify the position of the imaging area of ​​each tissue in each section image. The first model can be any existing or future developed deep learning model with target recognition capability, including but not limited to various target detection models and various image segmentation models. A preset number of sample images can be collected in advance. Exemplarily, these sample images can be ultrasound sample images of the above 8 sections of the shoulder joint of different patients. In addition, any suitable annotation method can be used to annotate the tissues contained in the ultrasound sample image. For example, the category and position of each target tissue expected to be identified in each ultrasound sample image can be annotated. Take the identification of 10 categories of tissues in the ultrasound image of 8 sections of the shoulder joint as an example. If the ultrasound image of the A-type section includes imaging areas of 5 types of tissues in these 10 types of tissues. For ultrasound images of type A sections, only three of the five types of tissues need to be identified, so only the categories and locations of the three types of tissues can be marked in ultrasound sample images belonging to type A sections. Then, these marked ultrasound sample images can be used to iteratively train the model to be trained using a suitable loss function until a good training effect is achieved.

[0062] It can be understood that in this example, the trained model has the ability to identify the above 10 types of tissues. Therefore, in this step, the image to be identified is input into the trained first model respectively, and the recognition result of the first model for the various types of tissues contained in the image can be obtained. The recognition result includes the position information of the various types of tissues identified by the model in the image. The position information can be represented in various suitable forms. For different models, the form of the position information of the tissue output is different. In the case where the first model is a target detection model, the target detection model can output a list of various types of tissues and the position coordinates of the position box representing the location of the tissue in this step. In the case where the first model is a segmentation model, a segmentation map representing the segmentation results of various types of tissues can be output in this step. For example, if the current image to be identified belongs to an A-type section image, after the image to be identified is input into the trained semantic segmentation model, a segmentation map of 10 channels output by the semantic segmentation model can be obtained. Each channel corresponds to a category of tissue. In each segmentation map, the size of the pixel value of each pixel can be proportional to the probability that the pixel at that position belongs to the corresponding category of tissue.

[0063] Step S160, based on the type of the section corresponding to the image to be identified, the location information of the target tissue in the image to be identified is filtered out from the first result. Different types of sections have different categories of target tissue included in the sections.

[0064] The type of the section can be set arbitrarily according to actual needs. In the example of the ultrasonic image of the 8 sections of the shoulder joint, the 8 sections can be divided into a plurality of suitable types. For example, if the category of the target tissue expected to be identified by each section is different from that of the other sections in the 8 sections, the 8 sections can be divided into 8 different types. If the categories of the target tissue expected to be identified by 2 sections are the same, and the categories of the target tissue expected to be identified by the other 6 sections are different, the 8 sections can be divided into 7 types. In this way, for different sections of the same type, the categories of the target tissue expected to be identified are exactly the same, while for different sections of different types, the categories of the target tissue expected to be identified are different. The type of section corresponding to each image to be identified can be determined in any suitable manner. In one example, the type of each image to be identified can be manually labeled. For example, the image to be identified can be manually named or numbered, and the images to be identified of different types of sections can be named with different names or numbers. In another example, the type of each image to be identified can also be determined using a trained classification model. In other words, the image to be identified can be input into a trained classification model, so that the classification model outputs type information of the section corresponding to the image to be identified. The type information can be a type number or the like.

[0065] As described above, the output result of the first model includes the position information of each type of tissue identified by the model in the image. Only some of these tissues may belong to the target tissue of the image. In this step, any suitable method can be used to filter out the position information of the target tissue in the image to be identified from the first result. Exemplarily but not restrictively, the various categories of the target tissue corresponding to the section can be determined according to the type of section corresponding to the image to be identified. Then, these categories can be matched with the various categories output by the first model, and the position information of the tissue that is successfully matched can be extracted. For example, the section corresponding to the image to be identified belongs to the A-type section, and the target tissue expected to be identified in the A-type section includes, for example, three types of tissues, such as the greater nodule, the lesser nodule and the deltoid muscle. Similarly, taking the first model as a semantic segmentation model, outputting the image to be identified to the semantic segmentation model can output a segmentation map of 10 channels as an example. It can be understood that there must be 3 channels in the segmentation map of these 10 channels corresponding to the segmentation map of the semantic segmentation model for the greater nodule, the lesser nodule and the deltoid muscle. Therefore, the three segmentation maps can be screened out, and the positions of the three tissues in the image can be further determined based on the three segmentation maps. For example, the position of the large nodule in the image to be identified can be determined based on the position of the pixel with a pixel value of 1 in the segmentation map of the first channel corresponding to the large nodule. Specifically, in this step, any suitable method can be used to extract the position information of the target tissue in the image to be identified, including but not limited to various suitable logical operation methods, mathematical operation methods, etc.

[0066] As mentioned above, the method of using multiple recognition models to identify target tissues in different section images in the prior art consumes a large amount of computation and time cost, and is also difficult to implement. However, the above method in the embodiment of the present application uses a recognition model to identify the tissue contained in the image to be identified, and then accurately screens out the location information of the target tissue to be identified from the recognition result according to the section type corresponding to the image. This solution not only has a high recognition accuracy, but also significantly reduces the amount of computation consumed and significantly improves the recognition efficiency. In addition, the scalability of the solution is also strong, and the recognition effect for various types of target tissues is good. Therefore, the user experience is better.

[0067] Exemplarily, the image to be identified is an ultrasonic image, and the tissue is musculoskeletal tissue. On the one hand, since ultrasonic imaging is fast and this imaging method is non-radioactive to the human body, images of different sections of the target part of the human body can be quickly and safely acquired. In addition, for doctors, in order to accurately acquire the condition of the musculoskeletal tissue of the patient's affected part, it is necessary to repeatedly check and compare multiple tissues in multiple section images. Not only is the workload huge, but it is also very dependent on the doctor's experience and is very prone to errors. The above method 100 of the embodiment of the present application can automatically, in real time and accurately obtain the position information of the target tissue in different section images, which can provide great convenience for doctors' diagnosis and significantly improve the efficiency of diagnosis.

[0068] Exemplarily, the method 100 further includes steps S111 to S114.

[0069] Step S111, obtaining the annotation data of the sample image. The annotation data includes the real position information of various target tissues in the sample image. For example, the sample database stores the ultrasound sample images of the above 8 sections of the shoulder joint of 200 patients. Exemplarily, various suitable image annotation tools can be used to annotate the real position of the target tissue in each ultrasound sample image. Specifically, in this step, the ultrasound sample images can be divided according to the corresponding section types, and the ultrasound sample images corresponding to the same section type are divided into a group. Then, each of each group of ultrasound sample images is annotated separately. For example, for the image of the A-type section, the target tissues expected to be identified are the three types of tissues, namely, the greater tubercle, the lesser tubercle and the deltoid muscle. Then, the positions of the three types of tissues in each ultrasound sample image corresponding to the A-type section can be annotated separately. For example, in the case where the first model is a semantic segmentation model, the image areas where the greater tubercle, the lesser tubercle and the deltoid muscle are located can be annotated with three different color blocks in each ultrasound sample image corresponding to the A-type section, thereby obtaining the annotation data of the ultrasound sample image.

[0070] Step S112, input the sample image into the first initial model, and output the first initial result. The first initial result includes the location information of various tissues contained in the sample image. It can be understood that the first initial model is an untrained first model. Exemplarily, all ultrasound sample images in the above-mentioned sample database can be input into the first initial model in sequence to obtain the prediction result of the model for each ultrasound sample image. Specifically, the prediction result includes the category information of each tissue contained in each ultrasound sample image and the location information of each tissue in the image. The implementation scheme of this step is similar to the reasoning step S140 of the aforementioned first model. Ordinary technicians in this field can understand the various implementation methods of this step, and for the sake of brevity, they will not be repeated here.

[0071] Step S113, based on the type of section corresponding to the sample image, filter out the predicted position information of various tissues in the sample image from the first initial result. As described above, the output result of the first initial model includes the predicted position information of various tissues in the image predicted by the model. Only some of these tissues may belong to the target tissue of the image. In this step, any suitable method can be used to filter out the predicted position information of the target tissue in each ultrasound sample image from the first initial result. The implementation method of this step is similar to the above-mentioned step S160, and for the sake of brevity, it will not be repeated here.

[0072] Step S114, training the first initial model according to the difference between the real position information and the predicted position information to obtain the first model. Exemplarily, for each sample image, the difference between the predicted position information of each target tissue in the image predicted by the model and the real position information of the target tissue in the image that has been marked can be calculated, and the value of the preset loss function, i.e., the loss value, can be calculated according to the difference. Moreover, the model can back-propagate and update each model parameter according to the difference until the loss value loss converges to the ideal value, and then the training can be terminated to obtain the trained first model.

[0073] It can be understood that, through this training scheme, the model can pay more attention to the recognition of the target tissue expected to be recognized in each image, so that different emphasis can be placed on the recognition of images of different section types. For example, although both the A-type section image and the B-type section image include tissue 1, tissue 1 is the target tissue of the B-type section image, not the target tissue of the A-type section image. It can be understood that since the features of tissue 1 in the A-type section image and the B-type section image may be different, interference will be caused to the recognition of the model. In this way, through the above training scheme, the model can have a higher recognition accuracy rate for tissue 1 when recognizing the B-type section image, while it does not pay attention to the recognition of tissue 1 in the A-type section image when recognizing the A-type section image. Thus, the interference of such similar tissues on the model recognition can be effectively avoided, so that the model can achieve the effect of distinguishing the tissues in different images, so as to ensure that the model can accurately recognize the target tissues expected to be recognized in images of different section types.

[0074] According to the above scheme, during the model training process, the output results of the model are screened, and the value of the loss function is calculated based on the screened prediction results for the target tissue and the actual results of the pre-labeled target tissue, so that the model can have a higher recognition accuracy for the target tissue. In addition, this scheme can avoid the interference of similar tissues on images with different effective sections on the model recognition, so that the model can have a higher recognition accuracy and recognition efficiency for the target tissue to be recognized in images of different section types.

[0075] Exemplarily, step S111 obtains annotation data of a sample image, including step S111a and step S111b. Step S111a, for each sample image, obtains annotation information of the sample image. Step S111b, converts the sample image into a first number of annotated images according to the annotation information. The first number is the total number of categories of tissues that the model can recognize. Each annotated image corresponds one-to-one to each category of tissue. And in the annotated image corresponding to the first tissue, the pixel value of the pixel corresponding to the position of the first tissue is different from the pixel values ​​of other pixels. In the annotated image where the corresponding tissue does not belong to the first tissue, the pixel values ​​of each pixel are the same. The first tissue is the target tissue in the section corresponding to the sample image.

[0076] Taking the identification of 10 categories of tissues in the ultrasound image of 8 sections of the shoulder joint as an example, the first number is 10. Exemplarily, in step S111a, various suitable image annotation tools can be used to annotate the true position of the target tissue in each ultrasound sample image. For example, for the sample image of the A-type section, the areas where the three types of tissues, namely the greater tubercle, the lesser tubercle and the deltoid muscle, are located can be annotated with different color blocks, and then the annotated sample image can be converted into a grayscale image. It can be understood that in the grayscale image, the pixel values ​​of the pixels of the three types of tissues are different. In step S111b, the grayscale image can be converted into a mask image of 10 channels according to the pixel values ​​of the pixels as an annotated image. Each channel corresponds to a tissue, and the label of each channel is the same as the category label of the corresponding tissue. Exemplarily, the pixel value of each pixel in each mask image is 0 or 1, and the position of the pixel with a pixel value of 1 is the same as the position of the pixel of the target tissue corresponding to the channel in the sample image. It can be understood that among the mask images of the 10 channels, the mask images of the 3 channels corresponding to the 3 types of tissues, namely the greater tubercle, the lesser tubercle and the deltoid muscle, have pixels whose pixel values ​​are not 0. However, the pixel values ​​of each pixel in the mask images of the 7 channels corresponding to the other 7 types of tissues are all 0.

[0077] In the above scheme, when there are multiple types of target tissues to be identified in the sample image, the real position information of target tissues of different categories can be distinguished and annotated, which is not only convenient for comparison with the predicted position information of target tissues of different categories output by the model, but also for quickly and accurately obtaining the value of the calculated loss function, thereby improving the efficiency of model training and ensuring a good training effect. Moreover, this scheme can also avoid confusion caused by different target tissue categories in different sample images.

[0078] Exemplarily, the first initial result is a first number of second segmentation maps. Each second segmentation map is a segmentation map for a category of tissue, and different second segmentation maps correspond to different categories of tissue. Taking the identification of 10 categories of tissue in the ultrasound image of 8 sections of the shoulder joint as an example, the first number is 10. Then in step S112, after each sample image is input into the first initial model, 10 channels of second segmentation maps can be obtained, and each channel can correspond to a category of tissue. For example, channel 1 to channel 10 correspond to tissue 1 to tissue 10, respectively. It can be understood that the second segmentation map of channel 3 is the segmentation result of the first initial model for tissue 3 in the sample image. Exemplarily, the pixel value of each pixel in each second segmentation map can be a value between 0 and 1, and the size of each pixel value represents the probability that the pixel at that position predicted by the model belongs to the corresponding category of tissue.

[0079] Step S113 selects predicted position information of various tissues in the sample image from the first initial result according to the type of the section corresponding to the sample image, including step S113a and step S113b.

[0080] Step S113a, mapping the type information of the section corresponding to each sample image to the category information of the target tissue in the section. The type information of the section is represented by a number, for example. In this step, the type number of the section corresponding to each sample image can be mapped to the category information of each target tissue in the section. The category information is represented by binary coding, vector, etc., for example, and the present application does not limit it. Exemplarily and non-restrictively, the category information can be represented by an array or sequence whose length is equal to the total number of categories of tissues that the model can recognize. For example, the type number of the section can be mapped to an array with a length of 10, and each element of the array can correspond to a category of tissue. For example, the 10 elements in the array correspond one-to-one to the above 10 tissues. The mapping relationship between the type number and the array can be set according to actual needs. The type information of the type A section is, for example, the type number 01, and the target tissues in the type A section include, for example, tissue 1, tissue 3, and tissue 5. In this step, the number 01 can be mapped to the array a, in which the values ​​of the 1st, 3rd, and 5th elements can be 1, and the values ​​of other elements can be 0.

[0081] Step S113b, according to the first number of second segmentation maps and category information, determine the predicted position information of the target tissue in the sample image. In this step, any suitable logical operation or mathematical operation can be performed according to the first number of second segmentation maps and category information to obtain the predicted position information of the target tissue in the sample image. For example, the above 10 channels correspond to the 10 elements of the array one by one. In this step, the tensor of the second segmentation map of the 10 channels output by the first initial model or each layer in the three-dimensional matrix (i.e., the second segmentation map of each channel) can be multiplied with the corresponding element in the array obtained by mapping to obtain another three-dimensional matrix. The three-dimensional matrix can be regarded as 10 segmentation maps, in which the values ​​of each element on the segmentation map corresponding to each non-target tissue are 0. In this way, in step S114, it can be compared with the 10 annotated images (which can also be regarded as three-dimensional matrices or tensors) obtained in step S111b. For example, the difference between the two three-dimensional matrices can be directly calculated, and the value of the loss function can be determined according to the difference.

[0082] This training scheme can effectively avoid the recognition confusion caused by the same or similar tissues in different sections being defined as foreground and background, and can shield the false positive interference caused by the misdetection of tissues that do not belong to the section. The calculation of the loss value can be focused on the target tissue that the image expects to identify, thereby achieving a better training effect and significantly improving the recognition accuracy of the trained model. In addition, the above scheme of using tensor and vector operations to screen the predicted position information of the target tissue has a simple execution logic and a small amount of calculation, which can significantly improve the training efficiency of the model.

[0083] Exemplarily, the first model is a segmentation model. The segmentation model may be any existing or future developed model that can achieve target segmentation, including but not limited to a semantic segmentation model or an instance segmentation model.

[0084] Step S140 inputs the image to be identified into the first model and outputs the first result, including step S141. Step S141, inputs the image to be identified into the segmentation model, and outputs a first number of first segmentation maps. The first number is equal to the total number of categories of tissues that can be identified by the segmentation model. Each first segmentation map is a segmentation map for a category of tissue. Different first segmentation maps correspond to different categories of tissues.

[0085] Taking the identification of 10 categories of tissues in the ultrasound image of 8 sections of the shoulder joint as an example, the first number is 10. Taking the semantic segmentation model as an example, it can be understood that the trained semantic segmentation model has the ability to identify the above 10 categories of tissues. Therefore, by inputting the image to be identified into the trained semantic segmentation model, a segmentation map of 10 channels can be output, and each segmentation map corresponds to a type of tissue. For example, the segmentation map in channel 1 is the segmentation result of the model for tissue 1. Each pixel in the segmentation map corresponds one-to-one to each pixel in the image to be identified. In the segmentation map, the pixel value of each pixel can be equal to the probability calculated by the model that the pixel belongs to tissue 1.

[0086] Step S160 filters out the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified, and includes step S161 and step S162.

[0087] Step S161, according to the type of the section corresponding to the image to be identified, determine the category information of the target tissue included in the section. Exemplarily, as described above, the type information of the section corresponding to the image to be identified can be first obtained, and then the type information can be mapped to the category information of the target tissue in the section. For example, the image to be identified can be input into a trained section classification model to obtain a section type number. Then, according to the mapping relationship between the preset type number and the vector, the section type number can be mapped to the category information of the target tissue contained in the section. In another example, the images to be identified can also be grouped and named in advance according to the corresponding model type, for example, the images to be identified corresponding to the same section type are named with the same digital number. In this step, the name of the image to be identified can be retrieved, and then its name can be mapped to the category information of the target tissue. As described above, the category information can be an array or sequence composed of a first number of elements, each element can correspond to a category of tissue.

[0088] Step S162, determining the location information of the target tissue in the image to be identified according to the first number of first segmentation graphs and the category information. The implementation principle of this step is similar to that of step S113b, and will not be described here for brevity.

[0089] In the above scheme, the trained segmentation model can accurately output a segmentation map representing the precise position of each tissue in the image to be identified. In addition, the category information of the target tissue in the section is determined according to the type of section corresponding to the image to be identified, and then the location information of each target tissue is filtered according to the category information. The execution logic of this scheme is simple and reasonable, the amount of calculation is small, and the overall recognition efficiency can be significantly improved, and the real-time and accurate recognition of the target tissue in the image can be realized.

[0090] Exemplarily and not restrictively, the first model is a semantic segmentation model or an instance segmentation model. It is understandable that the semantic segmentation model or the instance segmentation model can be used to output the exact position of each tissue in the image. That is, the position information of the contour of each tissue can be included. Therefore, different adjacent target tissues can be accurately located, so that more accurate position information of each target tissue can be obtained. For example, in the case of identifying musculoskeletal tissue in a patient's joint, the contour area of ​​the musculoskeletal tissue desired to be identified in each joint section image can be obtained using a semantic segmentation model or an instance segmentation model, so that based on this information, a diagnosis can be made quickly and accurately, which can further improve the efficiency and accuracy of the diagnosis.

[0091] Exemplarily, before filtering out the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified in step S160, the method 100 further includes step S150. In step S150, the image to be identified is input into the second model, and the type information of the section corresponding to the image to be identified is output.

[0092] Exemplarily, the second model can be various existing or future developed classification models. Take the identification of 10 categories of tissues in ultrasound images of 8 sections of the shoulder joint as an example. If in the 8 sections, the category of the target tissue that each section expects to identify is different from that of other sections, then the 8 sections can be divided into 8 different types. A multi-classification model can be pre-trained using different sample images containing the above 8 sections, and the number of categories of the classification model is 8. The type information of the section is, for example, a number. For example, the image to be identified can be input into a trained classification model, and the classification model calculates the probability that the image to be identified belongs to each section type. The model output can be set to correspond to the number of the section type with the highest probability.

[0093] In the above solution, the second model can be used to quickly and accurately obtain the type information of the section corresponding to the image to be identified, thereby significantly improving the overall recognition efficiency.

[0094] Exemplarily, the first number of first segmentation maps is represented by a first tensor. Step S161 determines the category information of the target tissue included in the section according to the type of the section corresponding to the image to be identified, including step S161a. Step S161a, determines a first array consisting of a first number of elements according to the type of the section corresponding to the image to be identified. Each element in the first array corresponds one-to-one to each category of tissue, and the value of the element corresponding to the first tissue in the first array is a non-zero number, and the values ​​of the remaining elements are 0, and the first tissue is the target tissue in the section corresponding to the image to be identified.

[0095] Take the identification of 10 categories of tissues in the ultrasound image of 8 sections of the shoulder joint as an example. The first model is, for example, a semantic segmentation model. For example, the category labels of the 10 categories of tissues identified by the semantic segmentation model are 1 to 10, respectively, and the segmentation maps of the 10 channels output by the semantic segmentation model are channels 1 to 10, respectively. It can be understood that each image can be regarded as a two-dimensional tensor, and the 10 segmentation maps arranged in channel order can be regarded as a three-dimensional tensor, that is, the first tensor mentioned above. It can be understood that each layer of the first tensor is a segmentation map of a channel. In step S161a, the type number of the section corresponding to each image to be identified can be obtained by using the section classification model. And the type number can be mapped to an array according to the mapping relationship between the preset type number and the vector. The length of the array can be 10, where each element corresponds to a category of tissue. For example, elements 1 to 10 correspond to tissues 1 to 10, respectively. For example, the image to be identified corresponds to section type A. The type information of the A-type section is, for example, type number 01, and the target tissues in the A-type section include, for example, tissue 1, tissue 3, and tissue 5. In this step, number 01 may be mapped to array a. Exemplarily, the values ​​of the 1st, 3rd, and 5th elements in the array a may be 1, and the values ​​of the other elements may be 0.

[0096] Step S162 determines the position information of the target tissue in the image to be identified according to the first number of first segmentation maps and the category information, and includes step S162a and step S162b.

[0097] Step S162a, multiply each layer in the first tensor by the corresponding element in the first array to obtain a second tensor. For example, for the first segmentation map of the 10 channels represented by the first tensor, channels 1 to 10 correspond to tissues 1 to 10 in sequence, and elements 1 to 10 in the first array obtained in step S161a also correspond to tissues 1 to 10 in sequence. Then in step S162a, the segmentation map of each channel represented by each layer in the first tensor is multiplied by the corresponding element in the first array, and the obtained second tensor can also represent the segmentation map of the above 10 channels. This segmentation map can be called a screening segmentation map. Taking the value of each element in the first array as 0 or 1 as an example, it can be understood that in the screening segmentation map of the 10 channels, the value of each element on the screening segmentation map corresponding to the non-target tissue is 0, and the screening segmentation map corresponding to the target tissue is the same as the first segmentation map of the same channel. In the example where the values ​​of the 1st, 3rd, and 5th elements in the above array a are 1 and the values ​​of the other elements are 0, the filtered segmentation maps corresponding to tissue 1, tissue 3, and tissue 5 in the second tensor are the same as the first segmentation map, while the pixel values ​​of each pixel in the filtered segmentation maps corresponding to other tissues are 0.

[0098] Step S162b, determining the position of the target tissue in the image to be identified based on the non-zero elements in the second tensor. The i-th layer in the first tensor corresponds to the i-th element in the first array, and both correspond to the same category of tissue, and i is greater than or equal to 1 and less than or equal to the first number. For example, each pixel with a pixel value greater than 0.5 in the screening segmentation map corresponding to tissue 1 in the second tensor can be searched, and each pixel in the image to be identified that has the same position as these pixels can be called the first pixel, and the image area composed of the first pixel can be determined as the imaging area of ​​tissue 1. Similarly, the imaging areas of tissues 3 and 5 in the image to be identified can be determined based on each pixel with a pixel value greater than 0.5 in the screening segmentation map corresponding to tissues 3 and 5 in the second tensor.

[0099] In this scheme, through mathematical operations, the location information of the target tissue output by the model that does not belong to the image to be identified can be quickly and accurately filtered out, and then the location of the target tissue in the image to be identified can be determined based on the non-zero elements in the tensor obtained after screening. This can effectively avoid the false positive interference of these misdetected tissues and improve the accuracy of the model output results. In addition, this method has a small amount of calculation and a relatively simple execution logic, which can greatly improve the overall recognition efficiency and realize real-time recognition of target tissues in the image. In addition, this solution is also highly scalable and suitable for the effective recognition of various target tissues in various types of images.

[0100] Figure 2 and Figure 3 Different parts of the flowchart of the method for identifying target tissue in an image according to another embodiment of the present application are respectively shown. Take the example of using the first model to identify 10 types of tissues in an ultrasound image of 8 sections of a shoulder joint. The 10 types of tissues are, for example, tissue 1 to tissue 10. Figure 2As shown, before the image to be identified is identified, the first initial model can be trained using each sample image in the sample database. The first initial model can be a semantic segmentation model. For each sample image, each target tissue that is expected to be identified in the sample image can be annotated to obtain a training label. Exemplarily, the training label can be a grayscale image of the same size as the sample image, in which the grayscale values ​​of pixels representing different target tissues are different. And the training label can be converted into a mask map of 10 channels, that is, an annotated image. The mask map of each channel corresponds to a class of tissue. The sample image can be input into the first initial model, and the prediction result of the model can be output. The prediction result is the second segmentation map of the 10 channels output by the semantic segmentation model, and channels 1 to 10 can correspond to tissues 1 to 10 in turn. The pixel value of each pixel in each second segmentation map represents the probability that the pixel at the position belongs to the corresponding category of tissue predicted by the semantic segmentation model. For example, the second segmentation map of channel 3 is the probability that the pixel at the position belongs to tissue 2 predicted by the semantic segmentation model. And the type number of the section corresponding to the sample image can be obtained, and then the type number can be mapped to an array of length 10. The value of each element can be 1 or 0, and elements 1 to 10 can correspond to tissues 1 to 10 in sequence. Exemplarily, in the array, the value of the element corresponding to the target tissue of the sample image is 1, and the value of the other elements is 0. Each layer in the third tensor representing the second segmentation map of the above-mentioned 10 second channels can be multiplied by the corresponding element in the mapped array to obtain a fourth tensor. It can be understood that the fourth tensor can also represent 10 segmentation maps, and the segmentation map of the channel corresponding to the target tissue in these 10 segmentation maps can be the same as the second segmentation map of the same channel, while the pixel value of each pixel in the segmentation map of the channel corresponding to the non-target tissue is 0. In this way, the 10 segmentation maps represented by the fourth tensor correspond to the annotated images of the 10 channels of the sample image. The fourth tensor is the predicted value of the position of each target tissue in the sample image by the semantic segmentation model, and the annotated image of the 10 channels is the true value of the position of each target tissue in the sample image, so the loss value loss can be calculated based on the difference between the two. The model can back-propagate and update each model parameter according to the difference until the loss value converges to the ideal value, then the training can be terminated to obtain a trained semantic segmentation model. For example, a section classification model can also be trained. The section type number corresponding to each sample image can be marked, and then the section classification model can be trained using the marked sample images to obtain a trained section classification model. Figure 3, after obtaining the current image to be identified, the image to be identified can be input into the trained semantic segmentation model and the section classification model respectively, so as to obtain the segmentation results of each tissue in the image by the semantic segmentation model and the type number of the section corresponding to the image by the section classification model. Similar to the training process, the semantic segmentation model can output the first segmentation map of the 10 channels corresponding to the above 10 tissues. These 10 first segmentation maps can be represented by a first tensor. The section type number output by the section classification model can be mapped to an array. The length of the array is 10, each element corresponds to a category of tissue, and in the array, the value of the element corresponding to the section target tissue can be 1, while the value of the element corresponding to other tissues can be 0. Then, each layer in the first tensor can be multiplied by the corresponding element in the array to obtain a second tensor, which can be regarded as a screening segmentation map of 10 channels. The screening segmentation map of the channel corresponding to the target tissue is the same as the first segmentation map of the same channel, and the pixel value of each pixel in the screening segmentation map of the channel corresponding to the non-target tissue can be 0. Finally, the position of each target tissue in the image to be identified can be determined according to the non-zero elements in the screening segmentation map corresponding to the target tissue.

[0101] According to another aspect of the present application, a system for identifying target tissue in an image is also provided. Figure 4 A schematic block diagram of a target tissue recognition system 400 in an image according to an embodiment of the present application is shown. As shown in the figure, the system 400 includes an acquisition module 410, a recognition module 420 and a screening module 430.

[0102] The acquisition module 410 is used to acquire the image to be identified, wherein the image to be identified is a cross-sectional image of a target part of the photographed object.

[0103] The recognition module 420 is used to input the image to be recognized into the first model and output a first result, wherein the first result includes the position information of various tissues contained in the image to be recognized.

[0104] The screening module 430 is used to screen the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified. Different types of sections have different categories of target tissue included in the sections.

[0105] Exemplarily, the first model is a segmentation model.

[0106] The recognition module 420 includes a first recognition unit. The first recognition unit is used to input the image to be recognized into the segmentation model and output a first number of first segmentation maps. The first number is equal to the total number of tissue categories that can be recognized by the segmentation model. Each first segmentation map is a segmentation map for a category of tissue, and different first segmentation maps correspond to different categories of tissue.

[0107] The screening module 430 includes a first screening submodule and a second screening submodule.

[0108] The first screening submodule is used to determine the category information of the target tissue included in the section according to the type of the section corresponding to the image to be identified.

[0109] The second screening submodule is used to determine the position information of the target tissue in the image to be identified according to the first number of first segmentation images and the category information.

[0110] Exemplarily, the first number of first segmentation maps are represented by a first tensor.

[0111] The first screening submodule includes a first determination unit. The first determination unit is used to determine a first array consisting of a first number of elements according to the type of the section corresponding to the image to be identified. Each element in the first array corresponds to each category of tissue one by one, and the value of the element corresponding to the first tissue in the first array is a non-zero number, and the values ​​of the remaining elements are 0, and the first tissue is the target tissue in the section corresponding to the image to be identified.

[0112] The second screening submodule includes a calculating unit and a second determining unit.

[0113] The computation unit is used to multiply each layer in the first tensor by the corresponding element in the first array to obtain a second tensor.

[0114] The second determination unit is used to determine the position of the target tissue in the image to be identified according to the non-zero elements in the second tensor, wherein the i-th layer in the first tensor corresponds to the i-th element in the first array, and both correspond to the same type of tissue, and i is greater than or equal to 1 and less than or equal to the first number.

[0115] Exemplarily, the system 400 further includes a type determination module.

[0116] The type determination module is used to input the image to be identified into the second model and output the type information of the section corresponding to the image to be identified.

[0117] Exemplarily, the system 400 further includes a labeling module, a prediction module, a prediction screening module and a training module.

[0118] The annotation module is used to obtain annotation data of the sample image, wherein the annotation data includes the real position information of various target tissues in the sample image.

[0119] The prediction module is used to input the sample image into the first initial model and output a first initial result, wherein the first initial result includes the position information of various tissues contained in the sample image.

[0120] The prediction and screening module is used to screen out the predicted position information of various tissues in the sample image from the first initial result according to the type of the section corresponding to the sample image.

[0121] The training module is used to train the first initial model according to the difference between the real position information and the predicted position information to obtain the first model.

[0122] Exemplarily, the first initial result is a first number of second segmentation graphs, wherein each second segmentation graph is a segmentation graph for a category of tissue, and different second segmentation graphs correspond to different categories of tissue.

[0123] The prediction and screening module includes a mapping unit and a third determination unit.

[0124] The mapping unit is used to map the type information of the section corresponding to each sample image into the category information of the target tissue included in the section.

[0125] The third determination unit is used to determine the predicted position information of the target tissue in the sample image according to the first number of second segmentation maps and the category information.

[0126] Exemplarily, the labeling module includes a first labeling submodule and a second labeling submodule.

[0127] The first labeling submodule is used to obtain labeling information of each sample image.

[0128] The second annotation submodule is used to convert the sample image into a first number of annotated images according to the annotation information. Each annotated image corresponds to each category of tissue one by one, and in the annotated image corresponding to the first tissue, the pixel value of the pixel corresponding to the position of the first tissue is different from the pixel values ​​of other pixels; in the annotated image corresponding to the tissue that does not belong to the first tissue, the pixel values ​​of each pixel are the same, and the first tissue is the target tissue in the section corresponding to the sample image.

[0129] According to another aspect of the present application, an electronic device is also provided. Figure 5 A schematic block diagram of an electronic device 500 according to an embodiment of the present application is shown. As shown in the figure, the electronic device 500 includes a processor 510 and a memory 520. The memory 520 stores computer program instructions, which are used by the processor 510 to execute the above-mentioned method 100 for identifying target tissue in an image when the computer program instructions are executed.

[0130] According to another aspect of the present application, a storage medium is also provided. Program instructions are stored on the storage medium, and the program instructions are used to execute the above-mentioned method 100 for identifying target tissues in images when running. The storage medium may include, for example, an erasable programmable read-only memory (EPROM), a portable read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The storage medium may be any combination of one or more computer-readable storage media.

[0131] A person skilled in the art can understand the specific implementation scheme and beneficial effects of the target tissue identification system, electronic device and storage medium in the above-mentioned image by reading the above-mentioned description of the target tissue identification method in the above-mentioned image, and for the sake of brevity, they will not be repeated here.

[0132] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present application to this. Those of ordinary skill in the art may make various changes and modifications therein without departing from the scope and spirit of the present application. All these changes and modifications are intended to be included within the scope of the present application as required by the appended claims.

[0133] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0134] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0135] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0136] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various inventive aspects, in the description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the method of the present application should not be interpreted as reflecting the following intention: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the corresponding claims, the inventive point is that the corresponding technical problem can be solved with features less than all the features of a single disclosed embodiment. Therefore, the claims following the specific embodiment are hereby explicitly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.

[0137] It will be understood by those skilled in the art that, except for mutually exclusive features, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this specification may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose.

[0138] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0139] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some modules in the target tissue recognition system in the image according to the embodiment of the present application. The present application can also be implemented as a device program (e.g., computer program and computer program product) for executing part or all of the methods described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0140] It should be noted that the above embodiments illustrate the present application rather than limit the present application, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of multiple such elements. The present application may be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names.

[0141] The above is only a specific implementation or description of a specific implementation of the present application, and the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for identifying target tissue in an image, characterized in that: include: Acquire an image to be identified, wherein the image to be identified is a cross-sectional image of a target part of a photographed object; Inputting the image to be identified into a first model and outputting a first result, wherein the first result includes position information of various tissues contained in the image to be identified; as well as According to the type of the section corresponding to the image to be identified, the position information of the target tissue in the image to be identified is filtered out from the first result, wherein different types of sections have different categories of target tissue included in the sections.

2. The method for identifying target tissue in an image according to claim 1, characterized in that: The first model is a segmentation model, The step of inputting the image to be recognized into a first model and outputting a first result includes: Input the image to be identified into a segmentation model, and output a first number of first segmentation maps, wherein the first number is equal to the total number of tissue categories that can be identified by the segmentation model, wherein each first segmentation map is a segmentation map for a category of tissue, and different first segmentation maps correspond to different categories of tissue; The step of filtering out the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified includes: Determining, according to the type of the section corresponding to the image to be identified, category information of the target tissue included in the section; and The position information of the target tissue in the image to be identified is determined according to the first number of first segmentation maps and the category information.

3. The method for identifying target tissue in an image according to claim 2, characterized in that: The first number of first segmentation maps is represented by a first tensor, The step of determining the category information of the target tissue included in the section according to the type of the section corresponding to the image to be identified includes: According to the type of the section corresponding to the image to be identified, determining a first array consisting of the first number of elements, each element in the first array corresponds to each category of tissue one by one, and the value of the element corresponding to the first tissue in the first array is a non-zero number, and the values ​​of the remaining elements are 0, and the first tissue is the target tissue in the section corresponding to the image to be identified; determining the position information of the target tissue in the image to be identified according to the first number of first segmentation maps and the category information, includes: multiplying each layer in the first tensor by the corresponding element in the first array to obtain a second tensor; and Determining the position of the target tissue in the image to be identified according to the non-zero elements in the second tensor; The i-th layer in the first tensor corresponds to the i-th element in the first array, and both correspond to the same category of tissues, and i is greater than or equal to 1 and less than or equal to the first number.

4. The method for identifying target tissue in an image according to claim 1, characterized in that: Before filtering out the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified, the method further includes: The image to be identified is input into the second model, and type information of the section corresponding to the image to be identified is output.

5. The method for identifying target tissue in an image according to any one of claims 1 to 4, characterized in that: The method further comprises: Acquire annotation data of a sample image, wherein the annotation data includes real position information of various target tissues in the sample image; Inputting the sample image into a first initial model and outputting a first initial result, wherein the first initial result includes position information of various tissues contained in the sample image; According to the type of the section corresponding to the sample image, the predicted position information of various tissues in the sample image is screened out from the first initial result; The first initial model is trained according to the difference between the actual position information and the predicted position information to obtain the first model.

6. The method for identifying target tissue in an image as claimed in claim 5 as cited in claim 2, characterized in that: The first initial result is the first number of second segmentation graphs, wherein each second segmentation graph is a segmentation graph for a category of tissue, and different second segmentation graphs correspond to different categories of tissue; The step of screening out predicted position information of various tissues in the sample image from the first initial result according to the type of the section corresponding to the sample image includes: Mapping the type information of the section corresponding to each sample image to the category information of the target tissue included in the section; and The predicted position information of the target tissue in the sample image is determined according to the first number of second segmentation maps and the category information.

7. The method for identifying target tissue in an image as claimed in claim 5 as cited in claim 2, characterized in that: The step of obtaining the annotation data of the sample image includes: For each sample image, obtaining annotation information of the sample image; and Converting the sample image into a first number of annotated images according to the annotation information; Among them, each annotated image corresponds one-to-one to each category of tissue, and in the annotated image corresponding to the first tissue, the pixel value of the pixel corresponding to the position of the first tissue is different from the pixel values ​​of other pixels; in the annotated image corresponding to the tissue that does not belong to the first tissue, the pixel values ​​of each pixel are the same, and the first tissue is the target tissue in the section corresponding to the sample image.

8. The method for identifying target tissue in an image according to any one of claims 1 to 4, characterized in that: The first model is a semantic segmentation model or an instance segmentation model.

9. The method for identifying target tissue in an image according to any one of claims 1 to 4, characterized in that: The image to be identified is an ultrasound image, and the tissue is musculoskeletal tissue.

10. A target tissue recognition system in an image, characterized in that: include: An acquisition module, used for acquiring an image to be identified, wherein the image to be identified is a cross-sectional image of a target part of a photographed object; A recognition module, used for inputting the image to be recognized into a first model and outputting a first result, wherein the first result includes position information of various tissues contained in the image to be recognized; as well as The screening module is used to screen out the location information of the target tissue in the image to be identified from the first result according to the type of the section corresponding to the image to be identified, wherein different types of sections have different categories of target tissue included in the sections.

11. An electronic device comprising a processor and a memory, characterized in that: The memory stores computer program instructions, which are used by the processor to execute the method for identifying target tissue in an image as claimed in any one of claims 1 to 9 when the processor is running the computer program instructions.

12. A storage medium having program instructions stored thereon, wherein the program instructions are used to execute the method for identifying target tissue in an image as claimed in any one of claims 1 to 9 when run.

Images