Feature segmentation-based target identification method and device in medical image

Abstract

The invention discloses a feature segmentation-based target identification method and device in a medical image, and belongs to the technical field of computers, and the method comprises the steps: obtaining a medical image; performing region segmentation based on spatial context information extraction on the medical image to obtain a first segmentation result; directly segmenting the medical image to obtain a segmentation feature and a second segmentation result; and based on the first segmentation result, the segmentation feature and the second segmentation result, performing target identification on the related region to obtain an identification result. According to the embodiment of the invention, segmentation based on spatial context and direct segmentation are combined, the related region most possibly including the target is locked, the method is high in region positioning accuracy, the shape is most matched with the target region, the region segmentation precision is remarkably improved, and the method is suitable for large-scale popularization and application. In addition, by multiplexing related features obtained in the segmentation and direct segmentation process based on the spatial context, the recognition accuracy of the target in the segmentation and direct segmentation process can be further improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and a device for identifying a target in a medical image based on feature segmentation. Background Art

[0002] As important basic data in medical imaging, medical images carry a large amount of effective information. By fully mining the information in medical images, we can quickly determine the target area and the target therein, thereby significantly improving the speed, efficiency and accuracy of medical image analysis, thereby providing strong support for related medical methods based on medical images. Summary of the invention

[0003] The purpose of the present invention is to provide a method and device for target recognition in medical images based on feature segmentation, so as to provide strong support for related medical means based on medical images.

[0004] The technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present invention provides a method for object recognition in medical images based on feature segmentation, the method comprising:

[0006] Acquisition of medical images;

[0007] Performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result;

[0008] Directly segmenting the medical image to obtain segmentation features and a second segmentation result;

[0009] Based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result.

[0010] As a further preferred embodiment of this solution, the performing of region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result includes:

[0011] dividing the medical image into a plurality of pixel regions;

[0012] For each pixel region, determining an associated pixel region corresponding to the pixel region, wherein the associated pixel region includes the pixel region and is centered on the pixel region;

[0013] Performing a convolution operation on the associated pixel region and the difference region of the pixel region to obtain a spatial context feature corresponding to the pixel region;

[0014] Based on the spatial context features corresponding to each pixel area, the probability that the corresponding pixel area belongs to the target area is predicted, and the pixel area whose probability meets the preset requirements is determined as the first area to be identified. The first segmentation result includes each first area to be identified and the spatial context features corresponding to each of the first areas to be identified.

[0015] As a further preferred embodiment of the present invention, the second segmentation result includes at least one segmentation area, and the target recognition is performed on the relevant area based on the first segmentation result, the segmentation feature and the second segmentation result to obtain the recognition result, including:

[0016] For any first region to be identified, if there is a target segmented region in the second segmentation result whose overlap with the first region to be identified is higher than a preset threshold, the first region to be identified and the target segmented region are subjected to a region fusion operation to obtain a second region to be identified, and the second region to be identified belongs to the related region;

[0017] Based on the segmentation features and the spatial context features corresponding to the first area to be identified, target identification is performed on the second area to be identified to obtain the identification result.

[0018] As a further preferred embodiment of the present invention, the target recognition is performed on the second area to be recognized based on the segmentation feature and the spatial context feature corresponding to the first area to be recognized to obtain the recognition result, including:

[0019] Performing a feature extraction operation on the second to-be-identified region to obtain a region feature;

[0020] fusing the region feature, the spatial context feature corresponding to the first region to be identified, and the segmentation feature to obtain a fused feature;

[0021] Based on the fusion feature, target recognition is performed on the second area to be recognized to obtain a recognition result.

[0022] As a further preferred embodiment of this solution, the method is implemented by a recognition model, and the recognition model is trained by the following method:

[0023] Acquire a sample medical image, wherein the sample medical image includes a target region label and a target label in the target region;

[0024] The sample medical image is input into the first segmentation network of the recognition model, and the first segmentation network is triggered to perform the following operations: the sample medical image is divided into multiple sample pixel areas; for each sample pixel area, a sample associated pixel area corresponding to the sample pixel area is determined, and the sample associated pixel area includes the sample pixel area and is centered on the sample pixel area; a convolution operation is performed on the difference area between the sample associated pixel area and the sample pixel area to obtain a sample space context feature corresponding to the sample pixel area; based on the sample space context features corresponding to each sample pixel area, a prediction result of whether the corresponding sample pixel area belongs to a target area is obtained, and based on the difference between the prediction result and the target area label, the parameters of the first segmentation network are adjusted until the parameter adjustment stop condition is reached.

[0025] As a further preferred embodiment of this solution, the recognition model further includes a second segmentation network, and the method further includes:

[0026] Directly segmenting the sample medical image based on the second segmentation network to obtain a sample segmentation feature and a sample second segmentation result;

[0027] The parameters of the first segmentation network are fixed, and the parameters of the second segmentation network are adjusted based on the difference between the second segmentation result of the sample and the target area label until the parameter adjustment stop condition is reached.

[0028] As a further preferred embodiment of this solution, the recognition model further includes a target recognition network, and the method further includes:

[0029] Inputting the sample first segmentation result output by the first segmentation network, the sample segmentation feature output by the second segmentation network, and the sample second segmentation result into the target recognition network to obtain a sample recognition result;

[0030] The parameters of the first segmentation network and the second segmentation network are fixed, and the parameters of the target recognition network are adjusted based on the difference between the sample recognition result and the target label until a parameter adjustment stop condition is reached.

[0031] In a second aspect, the present invention further provides a device for identifying targets in medical images based on feature segmentation, the device comprising:

[0032] An image acquisition module, used for acquiring medical images;

[0033] The processing module is used to perform the following operations:

[0034] Performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result;

[0035] Directly segmenting the medical image to obtain segmentation features and a second segmentation result;

[0036] Based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result.

[0037] In a third aspect, the present invention also provides a computer device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the above-mentioned method for target recognition in medical images based on feature segmentation.

[0038] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the above-mentioned method for target recognition in medical images based on feature segmentation.

[0039] In a fourth aspect, the present invention further provides a computer program product, the computer program product comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes to implement the above-mentioned method for object recognition in medical images based on feature segmentation.

[0040] The technical solution provided by this application can bring the following beneficial effects:

[0041] The embodiment of the present application proposes a method and device for target recognition in medical images based on feature segmentation. The embodiment of the present application combines segmentation based on spatial context with direct segmentation, and locks the second area to be recognized that is most likely to include the target through the regional changes of the difference set of the area and the union set of the area. The second area to be recognized obtained by this method has high positioning accuracy, and its shape is most consistent with the target area. Therefore, the accuracy of regional segmentation can be significantly improved. In the high-precision second area to be recognized, the recognition accuracy of the target therein can be further improved by reusing the relevant features obtained in the segmentation based on spatial context and the direct segmentation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 is a flow chart of a method for target recognition in medical images based on feature segmentation provided by an embodiment of the present application;

[0044] Figure 2 This is a flow chart of a segmentation process provided by an embodiment of the present application;

[0045] Figure 3 This is a schematic diagram of regional association relationships provided by an embodiment of the present application;

[0046] Figure 4 This is a schematic diagram of the identification method flow provided by an embodiment of the present application;

[0047] Figure 5 is a schematic diagram of a training process provided by an embodiment of the present application;

[0048] Figure 6 is a block diagram of a device for identifying an object in a medical image based on feature segmentation provided by an embodiment of the present application;

[0049] Figure 7 It is a structural block diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0050] Please refer to Figure 1 , which shows a flow chart of a method for target recognition in medical images based on feature segmentation provided by an embodiment of the present application. The method can be applied to a computer device, which refers to an electronic device with data calculation and processing capabilities.

[0051] The method for object recognition in medical images based on feature segmentation proposed in the embodiment of the present application includes the following steps:

[0052] Step S101. Acquire medical images.

[0053] The embodiments of the present application do not limit the medical images, which may be medical images captured by medical means such as X-rays or magnetic resonance imaging. The embodiments of the present application do not limit the parts of the human body corresponding to the medical images.

[0054] Step S102: Perform region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result.

[0055] This step intentionally ignores or erases the information of a certain pixel area, forcing the model to focus on the spatial context of the pixel area, thereby improving the ability to mine the spatial context and mining more spatial information to improve segmentation accuracy.

[0056] Please refer to Figure 2 , which shows a schematic diagram of the segmentation process flow of an embodiment of the present application. The performing of region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result includes:

[0057] Step S201. Divide the medical image into a plurality of pixel regions;

[0058] The embodiment of the present application does not limit the method of dividing the pixel area, and the medical image can be divided into several pixel areas according to the preset length and width.

[0059] Step S202. For each pixel area, determine the associated pixel area corresponding to the pixel area, wherein the associated pixel area includes the pixel area and is centered on the pixel area; perform a convolution operation on the difference area between the associated pixel area and the pixel area to obtain the spatial context features corresponding to the pixel area; based on the spatial context features corresponding to each pixel area, predict the probability that the corresponding pixel area belongs to the target area, and determine the pixel area whose probability meets the preset requirements as the first area to be identified, and the first segmentation result includes each of the first areas to be identified and the spatial context features corresponding to each of the first areas to be identified.

[0060] Please refer to Figure 3, which shows a schematic diagram of the regional association relationship of an embodiment of the present application. A medical image can be divided into N (N is greater than 1) pixel areas. For a certain pixel area, the pixels around it and the pixel area form an associated pixel area. The embodiment of the present application does not limit the size or shape of the associated pixel area, and does not constitute an implementation obstacle. The difference area between the associated pixel area and the pixel area can be understood as the environmental area corresponding to the pixel area. The spatial context feature corresponding to the pixel area can be obtained by performing a convolution operation on the environmental area. The spatial context feature does not include the features of the pixel area itself, thereby forcing the recognition model to complete the segmentation using only the spatial context feature, forcing the recognition model to maximize the mining of the spatial context feature, thereby improving the spatial environment information extraction capability of the recognition model, and forming a first segmentation result based on the first area to be identified segmented out of the spatial context feature and the spatial context feature corresponding to each of the first areas to be identified. The embodiment of the present application does not limit the preset requirements, which belongs to the parameters that can be set by oneself according to the actual situation in the segmentation problem, and does not constitute an implementation obstacle. It can be understood that if the probability meets the preset requirements, the recognition model considers that it includes the target and considers it to be a potential target area. In the embodiment of the present application, such an area is referred to as the first area to be identified, that is, the first area to be identified is a potential target area segmented by the spatial context dimension. In the embodiment of the present application, there is no limitation on the target and the target area. The target area is the area including the target. The target may be related to the actual application. For example, if an ulcer is to be identified in an actual application, the target is the ulcer. If a nodule is to be identified in an actual application, the target is the nodule.

[0061] Step S103. Directly segment the medical image to obtain segmentation features and a second segmentation result;

[0062] The present application embodiment does not limit the direct segmentation method, and the existing segmentation network can be used. The existing segmentation network fully pays attention to the information of the medical image itself, but the information utilization rate of the spatial context is insufficient, so the first segmentation result and the second segmentation result each reflect the segmentation results obtained by segmenting using information of different dimensions.

[0063] Step S104: Based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result.

[0064] In an embodiment of the present application, the second segmentation result includes at least one segmentation area, and based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result, including: for any first area to be recognized, if there is a target segmentation area in the second segmentation result with a degree of overlap with the first area to be recognized higher than a preset threshold, the first area to be recognized and the target segmentation area are subjected to a region fusion operation to obtain a second area to be recognized, and the second area to be recognized belongs to the relevant area; based on the segmentation feature and the spatial context feature corresponding to the corresponding first area to be recognized, target recognition is performed on the second area to be recognized to obtain the recognition result.

[0065] If the second segmentation result contains a target segmentation region whose overlap with the first region to be identified is higher than a preset threshold, it means that the result obtained by using spatial context segmentation is similar to the result of direct segmentation. Therefore, a region fusion operation is performed, and the fused second region to be identified has valid information of regions segmented by different segmentation methods, and the probability that it contains the target is significantly increased. The embodiment of the present application does not limit the preset threshold and the method for calculating the overlap, and does not constitute an implementation obstacle.

[0066] Please refer to Figure 4 , which shows a schematic flow chart of the recognition method of an embodiment of the present application. The target recognition of the second area to be recognized based on the segmentation feature and the spatial context feature corresponding to the first area to be recognized to obtain the recognition result includes:

[0067] Step S401: performing a feature extraction operation on the second region to be identified to obtain region features;

[0068] The embodiment of the present application does not limit the feature extraction operation. For example, it can be at least one level of convolution operation and can be implemented using a feature pyramid.

[0069] Step S402: Fusing the region feature, the spatial context feature corresponding to the first region to be identified, and the segmentation feature to obtain a fused feature;

[0070] The embodiment of the present application proposes that fusing the three types of features, namely, the regional features, the spatial context features corresponding to the first area to be identified, and the segmentation features, can significantly improve the recognition accuracy. The embodiment of the present application does not limit the fusion method. For example, fusion can be performed through convolution, multiplication, pooling, etc., which does not constitute an implementation obstacle.

[0071] Step S403: Perform target recognition on the second area to be recognized based on the fusion features to obtain a recognition result.

[0072] The embodiment of the present application does not limit the method of performing target recognition on the second area to be recognized based on the fusion feature to obtain the recognition result, and does not constitute an implementation obstacle.

[0073] In an exemplary embodiment, the method is implemented by a recognition model, which includes a first segmentation network, a second segmentation network, and a target recognition network. Figure 5 , which shows a schematic diagram of the training process of an embodiment of the present application. The recognition model is trained by the following method:

[0074] Step S501. Acquire a sample medical image, wherein the sample medical image includes a target region label and a target label in the target region;

[0075] Step S502. Input the sample medical image into the first segmentation network of the recognition model, triggering the first segmentation network to perform the following operations: divide the sample medical image into multiple sample pixel areas; for each sample pixel area, determine the sample associated pixel area corresponding to the sample pixel area, the sample associated pixel area includes the sample pixel area and is centered on the sample pixel area; perform a convolution operation on the difference area between the sample associated pixel area and the sample pixel area to obtain the sample space context features corresponding to the sample pixel area; based on the sample space context features corresponding to each sample pixel area, obtain a prediction result of whether the corresponding sample pixel area belongs to the target area, and adjust the parameters of the first segmentation network based on the difference between the prediction result and the target area label until the parameter adjustment stop condition is reached.

[0076] The embodiments of the present application do not limit the parameter adjustment stopping conditions mentioned in various places, nor do they limit the methods of adjusting parameters. These are conventional steps in neural network training, such as setting parameter adjustment stopping conditions based on the number of parameter adjustments or adjusting parameters based on the gradient descent method. The embodiments of the present application do not elaborate on these.

[0077] Step S503. Directly segment the sample medical image based on the second segmentation network to obtain sample segmentation features and a sample second segmentation result; fix the parameters of the first segmentation network, and adjust the parameters of the second segmentation network based on the difference between the sample second segmentation result and the target area label until the parameter adjustment stop condition is reached.

[0078] Step S504. Input the first segmentation result of the sample output by the first segmentation network, the sample segmentation feature and the second segmentation result of the sample output by the second segmentation network into the target recognition network to obtain a sample recognition result; fix the parameters of the first segmentation network and the second segmentation network, and adjust the parameters of the target recognition network based on the difference between the sample recognition result and the target label until the parameter adjustment stop condition is reached.

[0079] The training process in the embodiment of the present application is based on the same inventive concept as the aforementioned application stage, and the details are not repeated. It should be pointed out that the training stage uses three-stage training, which converges quickly by focusing only on adjusting the parameters of the relevant network at each stage. And the three-stage training order cannot be changed, which is determined based on the original network structure proposed in the embodiment of the present application, and has also been proven in practice to achieve the best training effect, improve the training speed and the performance of the recognition model in the application stage.

[0080] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0081] Please refer to Figure 6 , which shows a block diagram of a device for identifying targets in medical images based on feature segmentation provided by an embodiment of the present application. The device may include:

[0082] An image acquisition module 601 is used to acquire medical images;

[0083] The processing module 602 is used to perform the following operations:

[0084] Performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result;

[0085] Directly segmenting the medical image to obtain segmentation features and a second segmentation result;

[0086] Based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result.

[0087] In an exemplary embodiment, performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result includes:

[0088] dividing the medical image into a plurality of pixel regions;

[0089] For each pixel region, determining an associated pixel region corresponding to the pixel region, wherein the associated pixel region includes the pixel region and is centered on the pixel region;

[0090] Performing a convolution operation on the associated pixel region and the difference region of the pixel region to obtain a spatial context feature corresponding to the pixel region;

[0091] Based on the spatial context features corresponding to each pixel area, the probability that the corresponding pixel area belongs to the target area is predicted, and the pixel area whose probability meets the preset requirements is determined as the first area to be identified. The first segmentation result includes each first area to be identified and the spatial context features corresponding to each of the first areas to be identified.

[0092] In an exemplary embodiment, the second segmentation result includes at least one segmented area, and the target recognition is performed on the relevant area based on the first segmentation result, the segmentation feature and the second segmentation result to obtain the recognition result, including:

[0093] For any first region to be identified, if there is a target segmented region in the second segmentation result whose overlap with the first region to be identified is higher than a preset threshold, the first region to be identified and the target segmented region are subjected to a region fusion operation to obtain a second region to be identified, and the second region to be identified belongs to the related region;

[0094] Based on the segmentation features and the spatial context features corresponding to the first area to be identified, target identification is performed on the second area to be identified to obtain the identification result.

[0095] In an exemplary embodiment, the performing target recognition on the second area to be recognized based on the segmentation feature and the spatial context feature corresponding to the first area to be recognized to obtain the recognition result includes:

[0096] Performing a feature extraction operation on the second to-be-identified region to obtain a region feature;

[0097] fusing the region feature, the spatial context feature corresponding to the first region to be identified, and the segmentation feature to obtain a fused feature;

[0098] Based on the fusion feature, target recognition is performed on the second area to be recognized to obtain a recognition result.

[0099] In an exemplary embodiment, the apparatus comprises a recognition model, wherein the recognition model is trained by:

[0100] Acquire a sample medical image, wherein the sample medical image includes a target region label and a target label in the target region;

[0101] The sample medical image is input into the first segmentation network of the recognition model, and the first segmentation network is triggered to perform the following operations: the sample medical image is divided into multiple sample pixel areas; for each sample pixel area, a sample associated pixel area corresponding to the sample pixel area is determined, and the sample associated pixel area includes the sample pixel area and is centered on the sample pixel area; a convolution operation is performed on the difference area between the sample associated pixel area and the sample pixel area to obtain a sample space context feature corresponding to the sample pixel area; based on the sample space context features corresponding to each sample pixel area, a prediction result of whether the corresponding sample pixel area belongs to a target area is obtained, and based on the difference between the prediction result and the target area label, the parameters of the first segmentation network are adjusted until the parameter adjustment stop condition is reached.

[0102] In an exemplary embodiment, the recognition model further includes a second segmentation network, and the processing module is configured to perform the following operations:

[0103] Directly segmenting the sample medical image based on the second segmentation network to obtain a sample segmentation feature and a sample second segmentation result;

[0104] The parameters of the first segmentation network are fixed, and the parameters of the second segmentation network are adjusted based on the difference between the second segmentation result of the sample and the target area label until the parameter adjustment stop condition is reached.

[0105] In an exemplary embodiment, the recognition model further includes a target recognition network, and the processing module is used to perform the following operations:

[0106] Inputting the sample first segmentation result output by the first segmentation network, the sample segmentation feature output by the second segmentation network, and the sample second segmentation result into the target recognition network to obtain a sample recognition result;

[0107] The parameters of the first segmentation network and the second segmentation network are fixed, and the parameters of the target recognition network are adjusted based on the difference between the sample recognition result and the target label until a parameter adjustment stop condition is reached.

[0108] It should be noted that the device provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0109] Please refer to Figure 7, which shows a block diagram of a computer device provided by an embodiment of the present application, for executing the above-mentioned method for identifying a target in a medical image based on feature segmentation. Specifically:

[0110] The computer device 700 includes a central processing unit (CPU) 701, a system memory 704 including a random access memory (RAM) 702 and a read only memory (ROM) 703, and a system bus 705 connecting the system memory 704 and the central processing unit 701. The computer device 700 also includes a basic input / output system (I / O (Input / Output) system) 706 for facilitating information transmission between various components in the computer, and a large-capacity storage device 707 for storing an operating system 713, application programs 714, and other program modules 715.

[0111] The basic input / output system 706 includes a display 708 for displaying information and an input device 709 such as a mouse and a keyboard for user inputting information. The display 708 and the input device 709 are connected to the central processing unit 701 through an input / output controller 710 connected to the system bus 705. The basic input / output system 706 may also include an input / output controller 710 for receiving and processing inputs from a plurality of other devices such as a keyboard, a mouse, or an electronic stylus. Similarly, the input / output controller 710 also provides output to a display screen, a printer, or other types of output devices.

[0112] The mass storage device 707 is connected to the central processing unit 701 through a mass storage controller (not shown) connected to the system bus 705. The mass storage device 707 and its associated computer readable medium provide non-volatile storage for the computer device 700. That is, the mass storage device 707 may include a computer readable medium (not shown) such as a hard disk or a CD-ROM (Compact Disc Read-Only Memory) drive.

[0113] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer readable instructions, data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other solid-state storage technologies, CD-ROM, DVD (Digital Video Disc) or other optical storage, cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above. The above-mentioned system memory 704 and mass storage device 707 can be collectively referred to as memory.

[0114] According to various embodiments of the present application, the computer device 700 can also be connected to a remote computer on the network through a network such as the Internet. That is, the computer device 700 can be connected to the network 712 through the network interface unit 711 connected to the system bus 705, or the network interface unit 711 can be used to connect to other types of networks or remote computer systems (not shown).

[0115] The above-mentioned memory also includes a computer program, which is stored in the memory and is configured to be executed by one or more processors to implement the above-mentioned target recognition method in medical images based on feature segmentation.

[0116] In an exemplary embodiment, a computer-readable storage medium is also provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. When the at least one instruction, the at least one program, the code set or the instruction set is executed by a processor, the method for target recognition in medical images based on feature segmentation is implemented.

[0117] Specifically, the method comprises:

[0118] Acquisition of medical images;

[0119] Performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result;

[0120] Directly segmenting the medical image to obtain segmentation features and a second segmentation result;

[0121] Based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result.

[0122] In an exemplary embodiment, performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result includes:

[0123] dividing the medical image into a plurality of pixel regions;

[0124] For each pixel region, determining an associated pixel region corresponding to the pixel region, wherein the associated pixel region includes the pixel region and is centered on the pixel region;

[0125] Performing a convolution operation on the associated pixel region and the difference region of the pixel region to obtain a spatial context feature corresponding to the pixel region;

[0126] Based on the spatial context features corresponding to each pixel area, the probability that the corresponding pixel area belongs to the target area is predicted, and the pixel area whose probability meets the preset requirements is determined as the first area to be identified. The first segmentation result includes each first area to be identified and the spatial context features corresponding to each of the first areas to be identified.

[0127] In an exemplary embodiment, the second segmentation result includes at least one segmented area, and the target recognition is performed on the relevant area based on the first segmentation result, the segmentation feature and the second segmentation result to obtain the recognition result, including:

[0128] For any first region to be identified, if there is a target segmented region in the second segmentation result whose overlap with the first region to be identified is higher than a preset threshold, the first region to be identified and the target segmented region are subjected to a region fusion operation to obtain a second region to be identified, and the second region to be identified belongs to the related region;

[0129] Based on the segmentation features and the spatial context features corresponding to the first area to be identified, target identification is performed on the second area to be identified to obtain the identification result.

[0130] In an exemplary embodiment, the performing target recognition on the second area to be recognized based on the segmentation feature and the spatial context feature corresponding to the first area to be recognized to obtain the recognition result includes:

[0131] Performing a feature extraction operation on the second to-be-identified region to obtain a region feature;

[0132] fusing the region feature, the spatial context feature corresponding to the first region to be identified, and the segmentation feature to obtain a fused feature;

[0133] Based on the fusion feature, target recognition is performed on the second area to be recognized to obtain a recognition result.

[0134] In an exemplary embodiment, the method is implemented by a recognition model, and the recognition model is trained by the following method:

[0135] Acquire a sample medical image, wherein the sample medical image includes a target region label and a target label in the target region;

[0136] The sample medical image is input into the first segmentation network of the recognition model, and the first segmentation network is triggered to perform the following operations: the sample medical image is divided into multiple sample pixel areas; for each sample pixel area, a sample associated pixel area corresponding to the sample pixel area is determined, and the sample associated pixel area includes the sample pixel area and is centered on the sample pixel area; a convolution operation is performed on the difference area between the sample associated pixel area and the sample pixel area to obtain a sample space context feature corresponding to the sample pixel area; based on the sample space context features corresponding to each sample pixel area, a prediction result of whether the corresponding sample pixel area belongs to a target area is obtained, and based on the difference between the prediction result and the target area label, the parameters of the first segmentation network are adjusted until the parameter adjustment stop condition is reached.

[0137] In an exemplary embodiment, the recognition model further includes a second segmentation network, and the method further includes:

[0138] Directly segmenting the sample medical image based on the second segmentation network to obtain a sample segmentation feature and a sample second segmentation result;

[0139] The parameters of the first segmentation network are fixed, and the parameters of the second segmentation network are adjusted based on the difference between the second segmentation result of the sample and the target area label until the parameter adjustment stop condition is reached.

[0140] In an exemplary embodiment, the recognition model further includes a target recognition network, and the method further includes:

[0141] Inputting the sample first segmentation result output by the first segmentation network, the sample segmentation feature output by the second segmentation network, and the sample second segmentation result into the target recognition network to obtain a sample recognition result;

[0142] The parameters of the first segmentation network and the second segmentation network are fixed, and the parameters of the target recognition network are adjusted based on the difference between the sample recognition result and the target label until a parameter adjustment stop condition is reached.

[0143] Optionally, the computer-readable storage medium may include: ROM (Read Only Memory), RAM (Random Access Memory), SSD (Solid State Drives) or an optical disk, etc. Among them, the random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0144] In an exemplary embodiment, a computer program product or a computer program is also provided, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the above-mentioned method for object recognition in medical images based on feature segmentation.

[0145] It should be understood that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In addition, the step numbers described in this article only illustrate a possible execution sequence between the steps. In some other embodiments, the above steps may not be executed in the order of the numbers, such as two steps with different numbers are executed at the same time, or two steps with different numbers are executed in the opposite order to the diagram. The embodiments of the present application are not limited to this.

[0146] In addition, in the specific implementation of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0147] The above are merely exemplary embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for target recognition in medical images based on feature segmentation, characterized in that: The method comprises: Acquisition of medical images; Performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result; Directly segmenting the medical image to obtain segmentation features and a second segmentation result; Based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result.

2. The method according to claim 1, characterized in that: The performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result includes: dividing the medical image into a plurality of pixel regions; For each pixel region, determining an associated pixel region corresponding to the pixel region, wherein the associated pixel region includes the pixel region and is centered on the pixel region; Performing a convolution operation on the associated pixel region and the difference region of the pixel region to obtain a spatial context feature corresponding to the pixel region; Based on the spatial context features corresponding to each pixel area, the probability that the corresponding pixel area belongs to the target area is predicted, and the pixel area whose probability meets the preset requirements is determined as the first area to be identified. The first segmentation result includes each first area to be identified and the spatial context features corresponding to each of the first areas to be identified.

3. The method according to claim 2, characterized in that The second segmentation result includes at least one segmentation area, and the target recognition is performed on the relevant area based on the first segmentation result, the segmentation feature and the second segmentation result to obtain the recognition result, including: For any first region to be identified, if there is a target segmented region in the second segmentation result whose overlap with the first region to be identified is higher than a preset threshold, the first region to be identified and the target segmented region are subjected to a region fusion operation to obtain a second region to be identified, and the second region to be identified belongs to the related region; Based on the segmentation features and the spatial context features corresponding to the first area to be identified, target identification is performed on the second area to be identified to obtain the identification result.

4. The method according to claim 3, characterized in that The performing target recognition on the second area to be recognized based on the segmentation feature and the spatial context feature corresponding to the first area to be recognized to obtain the recognition result includes: Performing a feature extraction operation on the second to-be-identified region to obtain a region feature; fusing the region feature, the spatial context feature corresponding to the first region to be identified, and the segmentation feature to obtain a fused feature; Based on the fusion feature, target recognition is performed on the second area to be recognized to obtain a recognition result.

5. The method according to claim 4, characterized in that The method is implemented by a recognition model, which is trained by the following method: Acquire a sample medical image, wherein the sample medical image includes a target region label and a target label in the target region; Inputting the sample medical image into the first segmentation network of the recognition model, triggering the first segmentation network to perform the following operations: dividing the sample medical image into a plurality of sample pixel regions; For each sample pixel region, determine a sample-associated pixel region corresponding to the sample pixel region, wherein the sample-associated pixel region includes the sample pixel region and is centered on the sample pixel region; perform a convolution operation on a difference set region between the sample-associated pixel region and the sample pixel region to obtain a sample space context feature corresponding to the sample pixel region; Based on the sample space context features corresponding to each sample pixel area, a prediction result of whether the corresponding sample pixel area belongs to the target area is obtained, and based on the difference between the prediction result and the target area label, the parameters of the first segmentation network are adjusted until the parameter adjustment stop condition is reached.

6. The method according to claim 5, characterized in that The recognition model also includes a second segmentation network, and the method further includes: Directly segmenting the sample medical image based on the second segmentation network to obtain a sample segmentation feature and a sample second segmentation result; The parameters of the first segmentation network are fixed, and the parameters of the second segmentation network are adjusted based on the difference between the second segmentation result of the sample and the target area label until the parameter adjustment stop condition is reached.

7. The method according to claim 6, characterized in that The recognition model also includes a target recognition network, and the method further includes: Inputting the sample first segmentation result output by the first segmentation network, the sample segmentation feature output by the second segmentation network, and the sample second segmentation result into the target recognition network to obtain a sample recognition result; The parameters of the first segmentation network and the second segmentation network are fixed, and the parameters of the target recognition network are adjusted based on the difference between the sample recognition result and the target label until a parameter adjustment stop condition is reached.

8. A device for identifying targets in medical images based on feature segmentation, characterized in that: The device comprises: An image acquisition module, used for acquiring medical images; The processing module is used to perform the following operations: Performing region segmentation on the medical image based on spatial context information extraction to obtain a first segmentation result; Directly segmenting the medical image to obtain segmentation features and a second segmentation result; Based on the first segmentation result, the segmentation feature and the second segmentation result, target recognition is performed on the relevant area to obtain a recognition result.

9. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the target recognition method in medical images based on feature segmentation as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the target recognition method in medical images based on feature segmentation as described in any one of claims 1 to 7.

Images