Method for extracting image features and method for recognizing medical images

Abstract

The application discloses a method for extracting image features and a method for identifying medical images. The method for extracting image features comprises the following steps: obtaining image information of a computed tomography image, wherein the computed tomography image comprises a plurality of sub-images, and the image information of the computed tomography image at least comprises a layer thickness corresponding to each sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; extracting features of each sub-image to obtain image features corresponding to each sub-image; and performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image. The application solves the technical problem of low feature extraction efficiency in the prior art when extracting features of a computed tomography image.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method for extracting image features and a method for recognizing medical images. Background Art

[0002] CT (Computed Tomography) uses precisely collimated X-rays, gamma rays, ultrasound waves, etc., together with highly sensitive detectors, to perform one-by-one tomographic scans around a certain part of the human body to detect diseases.

[0003] Because the thickness of the reconstructed layers of CT images varies greatly, ranging from less than one millimeter to more than ten millimeters, if the CT image layer thickness is directly normalized in a fixed space, that is, the different layer thicknesses of the CT image are normalized to a fixed layer thickness, the image features obtained by feature extraction based on the fixed layer thickness will be significantly different from the image features obtained by feature extraction based on CT images with different layer thicknesses. In contrast, for the same tasks, such as lesion detection, classification, and segmentation, deep learning algorithms have similar semantic features at the deep level. If used properly, they can learn more generalized representations, thereby improving results.

[0004] In existing technologies, for the same task, feature extraction is performed by training corresponding algorithm modules based on CT images of different slice thicknesses. However, this approach requires more resources, such as storage. Furthermore, it requires maintaining multiple models, increasing operational costs and impacting iteration efficiency. Furthermore, the training algorithm data used in this approach may not be shared due to the slicing of data slices, thus increasing data requirements.

[0005] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0006] The embodiments of the present application provide a method for extracting image features and a method for recognizing medical images, so as to at least solve the technical problem of low feature extraction efficiency when extracting features from computed tomography images in the prior art.

[0007] According to one aspect of an embodiment of the present application, a method for extracting image features is provided, comprising: acquiring image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of the sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; performing feature extraction on each layer of the sub-image to obtain image features corresponding to each layer of the sub-image; and performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0008] According to another aspect of an embodiment of the present application, a medical image recognition method is also provided, including: reading a medical image to be recognized, wherein the medical image to be recognized is composed of multiple layers of sub-images, and the image information of the medical image to be recognized includes at least the layer thickness corresponding to each layer of sub-image; extracting image features of each layer of sub-image, and displaying target image features obtained by fusing image features of sub-images of a target number of layers, wherein the target number of layers is obtained by processing image information based on a preset depth layer in a preset neural network model; and displaying the recognition result of the medical image to be recognized obtained by analyzing the target image features.

[0009] According to another aspect of an embodiment of the present application, a device for extracting image features is also provided, including: an acquisition module for acquiring image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of sub-image; a processing module for processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; an extraction module for performing feature extraction on each layer of sub-image to obtain image features corresponding to each layer of sub-image; and a fusion module for performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0010] According to another aspect of an embodiment of the present application, a medical image recognition device is also provided, including: a reading module for reading a medical image to be recognized, wherein the medical image to be recognized is composed of multiple layers of sub-images, and the image information of the medical image to be recognized includes at least the layer thickness corresponding to each layer of sub-image; a first display module for extracting image features of each layer of sub-image, and displaying target image features obtained by fusing image features of sub-images of a target number of layers, wherein the target number of layers is obtained by processing image information based on a preset depth layer in a preset neural network model; a second display module for displaying the recognition result of the medical image to be recognized obtained by analyzing the target image features.

[0011] According to another aspect of an embodiment of the present application, a storage medium is also provided, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above-mentioned method for extracting image features and the above-mentioned method for recognizing medical images.

[0012] According to another aspect of an embodiment of the present application, a processor is further provided, which is used to run a program, wherein the program executes the above-mentioned method for extracting image features and the above-mentioned method for recognizing medical images when running.

[0013] According to another aspect of an embodiment of the present application, a system for extracting image features is also provided, including: a processor; and a memory connected to the processor, for providing the processor with instructions for processing the following processing steps: obtaining image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; performing feature extraction on each layer of sub-image to obtain image features corresponding to each layer of sub-image; and performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0014] In an embodiment of the present application, a fixed-space normalization processing method is adopted in the intermediate layer of deep learning. After acquiring the image information of the computed tomography image, the image information is processed by a preset depth layer in a preset neural network model to obtain the target number of layers corresponding to the computed tomography image, and feature extraction is performed on each layer of sub-image to obtain the image features corresponding to each layer of sub-image. Finally, feature fusion is performed on the image features based on the target number of layers to obtain the target image features corresponding to the computed tomography image.

[0015] In the above process, the same neural network model is used to extract features for CT images of different layer thicknesses, eliminating the need to occupy excessive storage resources and the need to maintain multiple neural network models. This reduces operation and maintenance costs, improves iteration efficiency, and thereby improves the efficiency of feature extraction from CT images. Furthermore, the solution provided by this application uses a preset depth layer in the neural network model to learn image features between layers, thereby preventing the problem of image information loss caused by image normalization processing of the CT images at the input of the neural network model.

[0016] It can be seen that the solution provided in the present application achieves the purpose of feature extraction of computed tomography images, thereby achieving the technical effect of improving the efficiency of image feature extraction, and further solving the technical problem of low feature extraction efficiency when extracting features from computed tomography images in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 is a hardware structure block diagram of an optional computing device for implementing the method for extracting image features according to an embodiment of the present application;

[0019] Figure 2 is a flow chart of a method for extracting image features according to an embodiment of the present application;

[0020] FIG3( a ) is a schematic diagram of an optional method for determining the target number of layers according to an embodiment of the present application;

[0021] FIG3( b ) is a schematic diagram of an optional determination of the target number of layers according to an embodiment of the present application;

[0022] Figure 4 is a flow chart of a method for extracting image features according to an embodiment of the present application;

[0023] Figure 5 is a schematic diagram of a process for extracting image features according to an embodiment of the present application;

[0024] Figure 6 is a schematic diagram of a device for extracting image features according to an embodiment of the present application;

[0025] Figure 7 is a schematic diagram of a device for extracting image features according to an embodiment of the present application;

[0026] Figure 8 This is a structural block diagram of an optional computing device according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] Example 1

[0030] According to an embodiment of the present application, an embodiment of a method for extracting image features is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0031] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computing device or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computing device (or mobile device) for implementing a method for extracting image features. Figure 1 As shown, the computing device 10 (or mobile device 10) may include one or more (illustrated as 102a, 102b, ..., 102n) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0032] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single, independent processing module, or may be incorporated in whole or in part into any of the other components of the computing device 10 (or mobile device). As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0033] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for extracting image features in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned method for extracting image features. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories may be connected to the computing device 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0034] Transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of computing device 10. In one embodiment, transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, transmission device 106 may be a radio frequency (RF) module configured to communicate with the Internet wirelessly.

[0035] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of computing device 10 (or mobile device).

[0036] It should be noted that, in some optional embodiments, the above Figure 1 The computer device (or mobile device) shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. Figure 1This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the aforementioned computer device (or mobile device).

[0037] In addition, it should be noted that a computing device, such as a computer, a tablet, etc., can serve as the executor of the method provided in this embodiment, wherein the computing device is used to perform image processing. The computing device can be connected to a CT machine directly (for example, the computing device is directly connected to the CT machine) or indirectly (for example, the user obtains an image scanned by the CT machine and inputs the image into the computing device) to obtain an image of the target object scanned by the CT machine, and then perform feature extraction on the image to detect the disease type of the target object, or detect the location of the lesion, etc.

[0038] Under the above operating environment, this application provides Figure 2 The method of extracting image features is shown. Figure 2 is a flow chart of a method for extracting image features according to the first embodiment of the present application, Figure 2 It can be seen that the method includes the following steps:

[0039] Step S202 : acquiring image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of the sub-image.

[0040] In step S202, the CT image (i.e., the computed tomography image described above) is a three-dimensional image. For the same CT image, the slice thickness corresponding to each layer of the sub-image contained therein is the same, that is, the sub-images contained in the same CT image have the same slice thickness, while the sub-images contained in different CT images may have different slice thicknesses. For example, the slice thickness of the sub-images contained in CT image A is 1 mm, while the slice thickness of the sub-images contained in CT image B is 3 mm.

[0041] In addition, the CT slice thickness represents the length covered by the computed tomography cross-sectional image in a direction perpendicular to the object or the examination bed. Optionally, the CT slice thickness is the slice thickness corresponding to each layer of sub-image contained in the CT image.

[0042] In an optional embodiment, the CT machine is directly connected to the computing device, including but not limited to a wired connection and a wireless connection, so that the computing device can directly obtain a CT image generated by the CT machine after scanning a target object (e.g., a human body). After obtaining the CT image, the computing device analyzes the CT image to obtain image information corresponding to the CT image.

[0043] In another optional embodiment, the computing device cannot directly obtain the CT image generated by the CT machine after scanning the target object. In this scenario, the user can obtain the CT image through a third-party device or the network, and input the CT image into the computing device, so that the computing device can obtain the CT image, analyze and process the CT image, and obtain image information corresponding to the CT image.

[0044] There is also an optional embodiment in which the user can set the layer thickness corresponding to the CT image when using a CT machine to scan the target object. In this scenario, the user can input the CT image and the image information corresponding to the CT image (for example, the CT layer thickness) into the computing device, so that the computing device can obtain the CT image and the image information, and perform feature extraction on the CT image.

[0045] Step S204 : Processing the image information based on the preset depth layer in the preset neural network model to obtain a target layer number corresponding to the computed tomography image.

[0046] In step S204, the above-mentioned preset depth layer can be set by the user according to actual needs. Optionally, it can be set by the user according to the task requirement information corresponding to different tasks. For example, the preset depth layer corresponding to the lesion detection task and the preset depth layer corresponding to the lesion classification task belong to different depth layers.

[0047] Additionally, in step S204, the target number of layers represents the number of sub-images from other layers that will be fused with the current sub-image when performing feature fusion on features in the CT image. For example, if the target number of layers is N, then the number of sub-images from all layers that will be fused is 2N+1, including the N sub-images before and after the current sub-image.

[0048] It should be noted that in the prior art, for CT images of varying slice thicknesses, the CT images are typically normalized at the input of the neural network model. That is, before the neural network model receives the CT images, the computing device first normalizes the slice thickness of the CT images and uses an image processing algorithm corresponding to the normalized slice thickness to perform feature extraction on the slice-thickness-normalized CT images. Because the slice thicknesses corresponding to CT images of varying slice thicknesses vary significantly, feature extraction from slice-thickness-normalized CT images is prone to loss of image information and yields poor results. Furthermore, the performance of the image processing algorithm corresponding to the normalized slice thickness often lags behind that of a fixed-slice-thickness-specific algorithm.

[0049] In this embodiment, the computing device automatically learns the information between layers in the deep learning intermediate layer (i.e., the above-mentioned preset depth layer), so that the CT image feature extraction effect can achieve the expected effect without losing CT image information.

[0050] Step S206 , performing feature extraction on each layer of sub-images to obtain image features corresponding to each layer of sub-images.

[0051] In step S206, the computing device may use an existing two-dimensional neural network model to extract features of each layer of sub-images. For example, the computing device may use a convolutional neural network model to extract features of each layer of sub-images to obtain image features corresponding to each layer of sub-images.

[0052] Step S208 , performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0053] In step S208, the computing device may fuse the image features of the sub-images of the target layer in the z-direction using an attention mechanism to obtain the target image features corresponding to the CT image. The attention mechanism is an information screening mechanism in deep learning that focuses on important points among a large amount of information, selecting key information while ignoring other unimportant information. In this embodiment, the computing device uses the attention mechanism to filter out important key image features from the multiple image features contained in each layer of sub-images, and then fuses the key image features corresponding to each layer of sub-images to obtain the target image features corresponding to the CT image.

[0054] It should be noted that after obtaining the target image features, the computing device may analyze the target image features according to actual task requirement information (eg, lesion detection, classification, segmentation, etc.).

[0055] Based on the scheme defined by the above steps S202 to S208, it can be known that in an embodiment of the present application, a fixed-space normalization processing method is adopted in the intermediate layer of deep learning. After obtaining the image information of the computed tomography image, the image information is processed by the preset depth layer in the preset neural network model to obtain the target number of layers corresponding to the computed tomography image, and feature extraction is performed on each layer of sub-image to obtain the image features corresponding to each layer of sub-image. Finally, feature fusion is performed on the image features based on the target number of layers to obtain the target image features corresponding to the computed tomography image.

[0056] It is easy to notice that in the above process, the same neural network model is used for feature extraction of CT images of different layer thicknesses, without occupying excessive storage resources or maintaining multiple neural network models. This reduces operation and maintenance costs, improves iteration efficiency, and thus improves the efficiency of feature extraction of CT images. Moreover, the solution provided by this application uses preset depth layers in the neural network model to learn image features between layers, thereby preventing the problem of image information loss caused by image normalization processing of CT images at the input end of the neural network model.

[0057] It can be seen that the solution provided in the present application achieves the purpose of feature extraction of computed tomography images, thereby achieving the technical effect of improving the efficiency of image feature extraction, and further solving the technical problem of low feature extraction efficiency when extracting features from computed tomography images in the prior art.

[0058] In an optional embodiment, after acquiring image information of the CT image, the computing device performs feature extraction on each layer of the sub-image to obtain image features corresponding to each layer of the sub-image. Specifically, the computing device performs information extraction on the computed tomography image based on a two-dimensional neural network model to obtain image features corresponding to each layer of the sub-image.

[0059] Optionally, the two-dimensional neural network model may include, but is not limited to, a convolutional layer, a batchnorm layer, and a convolutional layer, wherein the two-dimensional neural network model may perform a pooling operation on the CT image. In addition, the two-dimensional neural network model is provided at the front end of the deep learning model to extract shallow features of the CT image.

[0060] It should be noted that computing devices are used in the Batchnorm layer to accelerate the training of neural networks, speed up convergence speed and improve stability. The Batchnorm layer is an indispensable part of the deep learning model.

[0061] In an optional embodiment, after acquiring the image information of the CT image, the computing device processes the image information based on the preset depth layer in the preset neural network model to obtain the target number of layers corresponding to the computed tomography image. Specifically, the computing device first obtains the task requirement information corresponding to the preset task, and determines the target level of the preset depth layer in the preset neural network model based on the task requirement information, and then determines the preset receptive field based on the task requirement information. Finally, the target number of layers corresponding to the computed tomography image is calculated based on the target level and the preset receptive field in the preset depth layer. There is an association between the task requirement information and the target level, and the preset receptive field is the mapping area range of the pixel points on the feature map corresponding to each layer of the sub-image on each layer of the sub-image.

[0062] It should be noted that in the above process, the preset task can be, but is not limited to, lesion detection, lesion classification, image segmentation and the like, wherein different preset tasks correspond to different task requirement information. In addition, the preset depth layer corresponding to different task requirement information is also different, that is, in the present embodiment, for different task requirement information, the target level in the preset neural network model for normalizing the layer thickness of the CT image is different, for example, the preset depth layer corresponding to the lesion detection in the preset neural network model is N1 layer, and the preset depth layer corresponding to the lesion classification in the preset neural network model is N2 layer, N1 and N2 are not the same.

[0063] Optionally, the computing device inputs the task requirement information corresponding to the preset task into the first neural network model, and then obtains the target level output by the first neural network model, wherein the first neural network model is obtained by machine learning training using a plurality of groups of data, and each group of data in the plurality of groups of data includes task requirement information and a target level corresponding to the task requirement information.

[0064] It should be noted that the target level corresponding to the task requirement information can be calculated by the first neural network model described above, or can be set by the user according to actual requirements. Preferably, the user can set the target level according to the output result of the first neural network model and in combination with actual requirements, for example, the target level output by the first neural network model is N1, the user can directly set N1 as the target level, or can adjust N1, for example, adjust N1 to N2, that is, set N2 as the target level.

[0065] Optionally, the computing device inputs the task requirement information corresponding to the preset task into the second neural network model, and then obtains the preset receptive field output by the second neural network model, wherein the second neural network model is obtained by machine learning training using a plurality of groups of data, and each group of data in the plurality of groups of data includes task requirement information and a preset receptive field corresponding to the task requirement information.

[0066] It should be noted that the preset receptive field corresponding to the task requirement information can be calculated by the second neural network model described above, or can be set by the user according to actual requirements. Preferably, the user can set the preset receptive field according to the output result of the second neural network model and in combination with actual requirements, for example, the preset receptive field output by the second neural network model is M1, the user can directly set M1 as the preset receptive field, or can adjust M1, for example, adjust M1 to M2, that is, set M2 as the preset receptive field.

[0067] In an optional embodiment, after determining the preset receptive field, the computing device calculates the target number of layers based on the target layer of the CT image and the preset receptive field, wherein the target number of layers can be calculated as the ratio between the preset receptive field and the target layer. For example, in the schematic diagram of determining the target number of layers shown in FIG3(a), the CT layer thickness corresponding to the CT image is 1 mm, the preset receptive field is 9 mm, and the target number of layers is 9. For another example, in the schematic diagram of determining the target number of layers shown in FIG3(b), the CT layer thickness corresponding to the CT image is 3 mm, the preset receptive field is 9 mm, and the target number of layers is 3.

[0068] In an optional embodiment, after extracting features from each layer of sub-images and obtaining image features corresponding to each layer of sub-images, the computing device determines the feature type of the image features to be extracted based on the task requirement information, and performs feature screening on the image features corresponding to each layer of sub-images based on the feature type to obtain screened image features. Optionally, the computing device may use an attention mechanism to perform feature screening on the image features contained in each layer of sub-images. The computing device may input the task requirement information into a third neural network model and obtain the feature type output by the third neural network model, wherein the third neural network model is obtained through machine learning training using multiple sets of data, and each set of data in the multiple sets of data includes task requirement information and the feature type corresponding to the task requirement information.

[0069] It should be noted that the feature type corresponding to the above-mentioned task requirement information can be calculated by the above-mentioned third neural network model, or can be set by the user according to actual needs. Preferably, the user can set the feature type based on the feature type output by the third neural network model and in combination with actual needs. For example, if the feature type output by the third neural network model is L1, the user can directly set L1 as the feature type, or can adjust L1, for example, adjust L1 to L2, that is, set L2 as the feature type.

[0070] Furthermore, after determining the target number of layers, the computing device performs feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image. Specifically, the computing device determines a multi-layer target sub-image based on the target number of layers, obtains the image features corresponding to the multi-layer target sub-image, and then performs feature fusion on the image features corresponding to the multi-layer target sub-image to obtain target image features. The multi-layer target sub-image is a multi-layer sub-image among the multi-layer sub-images that is feature fused with the current sub-image. For example, in FIG3(a), the target number of layers is 9 layers, and the computing device performs feature fusion on the image features corresponding to the 9 layers of sub-images before the current sub-image, the image features corresponding to the 9 layers of sub-images after the current sub-image, and the image features corresponding to the current sub-image to obtain target image features. That is, in FIG3(a), the computing device performs feature fusion on 19 layers of sub-images. For another example, in Figure 3(b), the target number of layers is 3, so the computing device performs feature fusion on the image features corresponding to the 3 layers of sub-images before the current sub-image, the image features corresponding to the 3 layers of sub-images after the current sub-image, and the image features corresponding to the current sub-image to obtain the target image features. That is, in Figure 3(b), the computing device performs feature fusion on 7 layers of sub-images.

[0071] It should be noted that the preset neural network determines the target number of layers corresponding to the CT image, and the preset depth layers for feature extraction and feature fusion of the CT image may include but are not limited to z-direction average pooling layers, convolution layers, and bn layers of fixed physical size, and may also include other layers, or be replaced by other solutions, which are not specifically limited in this embodiment.

[0072] In addition, it should be noted that based on the feature representation of fixed layer thickness, image data corresponding to different layer thicknesses can also learn the same semantic features, thereby realizing end-to-end training of downstream tasks, such as lesion detection, image segmentation, and lesion classification.

[0073] In an optional embodiment, after performing feature fusion on the image features based on the target number of layers to obtain the target image features corresponding to the computed tomography image, the computing device detects whether there are lesion features corresponding to the target image features in the preset lesion database, and when there are lesion features corresponding to the target image features in the preset lesion database, the computing device determines the lesion information of the target object corresponding to the computed tomography image. For example, the computing device searches the preset lesion database for lesion features that are identical to the target image features or have a similarity greater than a preset similarity. When it is detected that there are lesion features corresponding to the target image features in the preset lesion database, the disease corresponding to the person (i.e., the target object) corresponding to the CT image can be determined based on the lesion features. That is, the computing device can use the target image features extracted by the method for extracting image features provided in this embodiment to perform lesion detection.

[0074] In another optional embodiment, after performing feature fusion on image features based on the target number of slices to obtain target image features corresponding to the computed tomography image, the computing device performs image segmentation on the computed tomography image based on the target image features to obtain multiple sub-region images, analyzes the multiple sub-region images to determine a target sub-region image, and then determines the location information of the abnormal physical sign state in the target object based on the location information of the target sub-region image in the computed tomography image. The target sub-region image at least includes the abnormal physical sign state of the target object corresponding to the computed tomography image.

[0075] For example, if there is a lesion in the lungs of a target object (e.g., a person), after the computing device obtains the target image features corresponding to the CT image using the method provided in this embodiment, the computing device divides the CT image into multiple sub-region images according to the target image features, and performs feature analysis on the image features contained in the multiple sub-region images to determine whether the features contained in the sub-region images are lesion features. If they are lesion features, the position of the sub-region image in the CT image is recorded, so that the specific position of the lesion in the human body can be determined.

[0076] From the above content, it can be seen that this application can automatically learn the common information of original images of different layer thicknesses through deep information fusion through deep learning, thereby improving the extraction effect of CT image features.

[0077] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0078] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method for extracting image features according to the above embodiment can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, disk, CD), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0079] Example 2

[0080] According to an embodiment of the present application, a medical image recognition method is also provided, such as Figure 4 As shown, the method includes:

[0081] Step S402 : reading a medical image to be identified, wherein the medical image to be identified is composed of multiple layers of sub-images, and the image information of the medical image to be identified at least includes the layer thickness corresponding to each layer of the sub-image.

[0082] In step S402, the medical image to be identified may be, but is not limited to, a computed tomography (CT) image. A CT image (i.e., the aforementioned CT image) is a three-dimensional image. For the same CT image, the slice thickness corresponding to each sub-image layer is the same, i.e., the sub-images contained in the same CT image have the same slice thickness. However, the slice thickness of the sub-images contained in different CT images may differ. For example, the slice thickness of the sub-images contained in CT image A is 1 mm, while the slice thickness of the sub-images contained in CT image B is 3 mm. Furthermore, the CT slice thickness represents the length covered by a CT cross-sectional image in a direction perpendicular to the object or examination bed. Optionally, the CT slice thickness is the slice thickness corresponding to each sub-image layer contained in the CT image.

[0083] In an optional embodiment, the CT machine is directly connected to the computing device, including but not limited to a wired connection and a wireless connection, so that the computing device can directly obtain a CT image generated by the CT machine after scanning a target object (e.g., a human body). After obtaining the CT image, the computing device analyzes the CT image to obtain image information corresponding to the CT image.

[0084] In another optional embodiment, the computing device cannot directly obtain the CT image generated by the CT machine after scanning the target object. In this scenario, the user can obtain the CT image through a third-party device or the network, and input the CT image into the computing device, so that the computing device can obtain the CT image, analyze and process the CT image, and obtain image information corresponding to the CT image.

[0085] There is also an optional embodiment in which the user can set the layer thickness corresponding to the CT image when using a CT machine to scan the target object. In this scenario, the user can input the CT image and the image information corresponding to the CT image (for example, the CT layer thickness) into the computing device, so that the computing device can obtain the CT image and the image information, and perform feature extraction on the CT image.

[0086] Step S404: extract the image features of each layer of sub-images, and display the target image features obtained by fusing the image features of the sub-images of the target layer, wherein the target layer is obtained by processing the image information based on the preset depth layer in the preset neural network model.

[0087] In step S404, the above-mentioned preset depth layer can be set by the user according to actual needs. Optionally, it can be set by the user according to the task requirement information corresponding to different tasks. For example, the preset depth layer corresponding to the lesion detection task and the preset depth layer corresponding to the lesion classification task belong to different depth layers.

[0088] In addition, if Figure 5 As shown, the computing device can extract information from the CT image to obtain the target number of layers. The target number of layers represents the number of sub-images in other layers that are to be fused with the current sub-image when performing feature fusion on features in the CT image. For example, if the target number of layers is N, the number of sub-images in all layers to be fused is 2N+1, including the N sub-images before and after the current sub-image.

[0089] It should be noted that in the prior art, for CT images of varying slice thicknesses, the CT images are typically normalized at the input of the neural network model. That is, before the neural network model receives the CT images, the computing device first normalizes the slice thickness of the CT images and uses an image processing algorithm corresponding to the normalized slice thickness to perform feature extraction on the slice-thickness-normalized CT images. Because the slice thicknesses corresponding to CT images of varying slice thicknesses vary significantly, feature extraction from slice-thickness-normalized CT images is prone to loss of image information and yields poor results. Furthermore, the performance of the image processing algorithm corresponding to the normalized slice thickness often lags behind that of a fixed-slice-thickness-specific algorithm.

[0090] In this embodiment, the computing device automatically learns the information between layers in the deep learning intermediate layer (i.e., the above-mentioned preset depth layer), so that the CT image feature extraction effect can achieve the expected effect without losing CT image information.

[0091] In addition, in this embodiment, the computing device can use an existing two-dimensional neural network model to extract features from each layer of sub-images, such as Figure 5 In the process, the computing device extracts features from the CT image using a preset neural network model to obtain image features. For example, the computing device uses a convolutional neural network model to extract features of each layer of sub-images to obtain image features corresponding to each layer of sub-images. Figure 5As shown, after obtaining the image features and the target layer number, the computing device can fuse the image features of the sub-images of the target layer number in the z direction through the attention mechanism to obtain the target image features corresponding to the CT image. The attention mechanism is an information screening mechanism in deep learning. Through this mechanism, attention can be focused on important points among a large amount of information, key information can be selected, and other unimportant information can be ignored. In this embodiment, the computing device uses the attention mechanism to filter out important key image features from the multiple image features contained in each layer of sub-images, and performs feature fusion on the key image features corresponding to each layer of sub-images to obtain the target image features corresponding to the CT image.

[0092] Step S406 , displaying the recognition result of the medical image to be recognized obtained by analyzing the target image features.

[0093] It should be noted that after obtaining the target image features, the computing device can analyze the target image features according to the actual task requirement information (for example, lesion detection, classification, segmentation, etc.) and display the recognition results, such as Figure 5 As shown, the test results of lesion detection are displayed, such as Figure 5 The recognition result is "early lung cancer lesion", and the lesion location corresponding to the "early lung cancer lesion" is marked.

[0094] Based on the scheme defined by the above steps S402 to S406, it can be known that in an embodiment of the present application, a fixed-space normalization processing method is adopted in the intermediate layer of deep learning. After reading the medical image to be identified composed of multiple layers of sub-images, the image features of each layer of sub-images are extracted, and the image information is processed based on the preset depth layer in the preset neural network model to obtain the target number of layers, and then the image features of the sub-images of the target number of layers are fused to obtain the target image features. Finally, the recognition result of the medical image to be identified is obtained by analyzing the target image features.

[0095] It is easy to notice that in the above process, the same neural network model is used for features of medical images to be identified with different layer thicknesses, without occupying too much storage resources or maintaining multiple neural network models, thus reducing operation and maintenance costs, improving iteration efficiency, and thus improving the feature extraction efficiency of the medical images to be identified. Moreover, the solution provided by the present application uses a preset depth layer in the neural network model to learn image features between layers, thereby preventing the problem of image information loss caused by image normalization processing of the medical images to be identified at the input end of the neural network model.

[0096] It can be seen that the solution provided in this application achieves the purpose of feature extraction of medical images to be identified, thereby achieving the technical effect of improving the efficiency of image feature extraction, and further solving the technical problem of low feature extraction efficiency in the existing technology when extracting features of medical images to be identified.

[0097] It should be noted that the process of obtaining the target image features through the above steps S402 to S404 is the same as the method of extracting image features in Example 1. The relevant content has been described in detail in Example 1 and will not be repeated here.

[0098] In an optional embodiment, after obtaining the medical image to be identified, the computing device analyzes the target image features to obtain an identification result for the medical image to be identified. Specifically, the computing device segments the medical image to be identified based on the target image features to obtain multiple sub-region images. Then, the computing device determines the target sub-region image containing the lesion features from the multiple sub-region images and highlights the location information of the target sub-region image in the medical image to be identified.

[0099] For example, if there is a lesion in the lungs of a target object (e.g., a person), after the computing device obtains the target image features corresponding to the medical image to be identified using the method provided in this embodiment, the computing device divides the medical image to be identified into multiple sub-region images according to the target image features, and performs feature analysis on the image features contained in the multiple sub-region images to determine whether the features contained in the sub-region images are lesion features. If they are lesion features, the position of the sub-region image in the medical image to be identified is recorded, thereby determining the specific position of the lesion in the human body. In addition, after determining the position of the lesion feature in the medical image to be identified, the computing device highlights the position where the lesion feature is located on the medical image to be identified, so that medical personnel can quickly determine the location of the lesion.

[0100] In an optional embodiment, when determining a target sub-region image containing abnormal features from multiple sub-region images, the computing device obtains the sub-image features corresponding to each sub-region image and detects whether a lesion feature corresponding to the sub-image feature exists in a preset lesion database. Upon detecting that a lesion feature corresponding to the sub-image feature exists in the preset lesion database, the computing device determines that the sub-region image corresponding to the sub-image feature is the target sub-region image. For example, the computing device searches the preset lesion database for a lesion feature that is identical to the sub-image feature or has a similarity greater than a preset similarity. Upon detecting that a lesion feature corresponding to the sub-image feature exists in the preset lesion database, the computing device determines that the human body corresponding to the medical image to be identified has an abnormality. In this case, the computing device can determine the specific location of the lesion in the human body by detecting the sub-region image in the medical image to be identified where the lesion feature exists.

[0101] Furthermore, after highlighting the position information of the target sub-region image in the medical image to be identified, the computing device obtains the lesion information corresponding to the target image feature from the preset lesion database and displays the lesion information. The preset lesion database includes at least: preset image features, preset lesion information, and the association relationship between the preset image features and the preset lesion information. Optionally, the computing device can query the lesion information corresponding to the target image feature through the preset lesion database, and can determine the disease type corresponding to the target object, and display the disease type on the display interface of the computing device, for example, Figure 5 The disease type shown in the figure is early lung cancer lesions.

[0102] From the above content, it can be seen that this application can automatically learn the common information of original images of different layer thicknesses through deep information fusion through deep learning, thereby improving the extraction effect of the medical image to be identified.

[0103] Example 3

[0104] According to an embodiment of the present application, a device for extracting image features for implementing the method in the above embodiment 1 is also provided, such as Figure 6 As shown, the device 60 includes: an acquisition module 601 , a processing module 603 , an extraction module 605 and a fusion module 607 .

[0105] Among them, the acquisition module 601 is used to obtain image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least the layer thickness corresponding to each layer of sub-image; the processing module 603 is used to process the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; the extraction module 605 is used to extract features from each layer of sub-image to obtain image features corresponding to each layer of sub-image; and the fusion module 607 is used to perform feature fusion on target image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0106] It should be noted here that the above-mentioned acquisition module 601, processing module 603, extraction module 605 and fusion module 607 correspond to steps S202 to S208 in Example 1. The four modules and the corresponding steps implement the same instances and application scenarios, but are not limited to the contents disclosed in the above-mentioned Example 1.

[0107] Example 4

[0108] According to an embodiment of the present application, a medical image recognition device for implementing the method in the above embodiment 2 is also provided. Figure 7 As shown, the device 70 includes: a reading module 701 , a first display module 703 and a second display module 705 .

[0109] The reading module 701 is configured to read a medical image to be identified, wherein the medical image to be identified is composed of multiple layers of sub-images, and the image information of the medical image to be identified includes at least the layer thickness corresponding to each layer of the sub-image;

[0110] A first display module 703 is configured to extract image features of each layer of sub-images and display target image features obtained by fusing image features of sub-images of a target number of layers, wherein the target number of layers is obtained by processing image information based on a preset depth layer in a preset neural network model;

[0111] The second display module 705 is used to display the recognition result of the medical image to be recognized obtained by analyzing the image features.

[0112] It should be noted here that the above-mentioned reading module 701, first display module 703 and second display module 705 correspond to steps S402 to S406 in Example 2. The three modules and the corresponding steps implement the same instances and application scenarios, but are not limited to the contents disclosed in the above-mentioned Example 2.

[0113] Example 5

[0114] According to an embodiment of the present application, a system for extracting image features for implementing the method in the above-mentioned embodiment 1 is also provided. The system includes: a processor and a memory.

[0115] Among them, the memory is connected to the processor and is used to provide the processor with instructions for processing the following processing steps: obtaining image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; performing feature extraction on each layer of sub-image to obtain image features corresponding to each layer of sub-image; and performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0116] As can be seen from the above, in an embodiment of the present application, a fixed-space normalization processing method is adopted in the intermediate layer of deep learning. After obtaining the image information of the computed tomography image, the image information is processed through the preset depth layer in the preset neural network model to obtain the target number of layers corresponding to the computed tomography image, and feature extraction is performed on each layer of sub-image to obtain the image features corresponding to each layer of sub-image. Finally, feature fusion is performed on the image features based on the target number of layers to obtain the target image features corresponding to the computed tomography image.

[0117] It is easy to notice that in the above process, the same neural network model is used for feature extraction of CT images of different layer thicknesses, without occupying excessive storage resources or maintaining multiple neural network models. This reduces operation and maintenance costs, improves iteration efficiency, and thus improves the efficiency of feature extraction of CT images. Moreover, the solution provided by this application uses preset depth layers in the neural network model to learn image features between layers, thereby preventing the problem of image information loss caused by image normalization processing of CT images at the input end of the neural network model.

[0118] It can be seen that the solution provided in the present application achieves the purpose of feature extraction of computed tomography images, thereby achieving the technical effect of improving the efficiency of image feature extraction, and further solving the technical problem of low feature extraction efficiency when extracting features from computed tomography images in the prior art.

[0119] It should be noted that the processor in this embodiment can execute the method for extracting image features in Example 1, wherein the content included in the method for extracting image features executed by the processor is the same as that in Example 1 and will not be repeated here.

[0120] Example 6

[0121] The embodiment of the present application may provide a computing device, which may be any computing device in a computer terminal group. Optionally, in this embodiment, the computing device may also be replaced by a terminal device such as a mobile terminal.

[0122] Optionally, in this embodiment, the computing device may be located in at least one network device among a plurality of network devices of a computer network.

[0123] In this embodiment, the computing device may execute the program code for the following steps in the method for extracting image features: obtaining image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of the sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; performing feature extraction on each layer of the sub-image to obtain image features corresponding to each layer of the sub-image; and performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0124] In this embodiment, the above-mentioned computing device can also execute the program code of the following steps in the medical image recognition method: reading the medical image to be recognized, wherein the medical image to be recognized is composed of multiple layers of sub-images, and the image information of the medical image to be recognized includes at least the layer thickness corresponding to each layer of sub-image; extracting the image features of each layer of sub-image, and displaying the target image features obtained by fusing the image features of the sub-images of the target number of layers, wherein the target number of layers is obtained by processing the image information based on the preset depth layer in the preset neural network model; and displaying the recognition result of the medical image to be recognized obtained by analyzing the target image features.

[0125] Optionally, Figure 8 This is a structural block diagram of a computing device according to an embodiment of the present application. Figure 8 As shown, the computing device 10 may include: one or more (only one is shown in the figure) processors 802 , a memory 804 and a peripheral interface 806 .

[0126] The memory can be configured to store software programs and modules, such as program instructions / modules corresponding to the method for extracting image features and the method for identifying medical images in the embodiments of the present application. The processor executes various functions and data processing by running the software programs and modules stored in the memory, that is, implements the method for extracting image features and the method for identifying medical images. The memory can include a high-speed random access memory, and can further include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory can further include a memory remotely arranged with respect to the processor, which can be connected to the computing device 10 through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0127] The processor can call information and application programs stored in the memory through the transmission device to perform the following steps: obtaining image information of a computed tomography image, wherein the computed tomography image includes a plurality of sub-images, and the image information of the computed tomography image at least includes a layer thickness corresponding to each sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; performing feature extraction on each sub-image to obtain image features corresponding to each sub-image; and performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0128] Optionally, the processor can further execute program codes of the following steps: extracting information of the computed tomography image based on a two-dimensional neural network model to obtain image features corresponding to each sub-image.

[0129] Optionally, the processor can further execute program codes of the following steps: obtaining task requirement information corresponding to a preset task; determining a target level of the preset depth layer in the preset neural network model according to the task requirement information, wherein the task requirement information and the target level have an association relationship; determining a preset receptive field according to the task requirement information, wherein the preset receptive field is a mapping area range of a pixel point located on a feature map corresponding to each sub-image on each sub-image; and calculating the target number of layers corresponding to the computed tomography image according to the target level and the preset receptive field in the preset depth layer.

[0130] Optionally, the processor can further execute program codes of the following steps: after performing feature extraction on each sub-image to obtain image features corresponding to each sub-image, determining a feature type of the image features to be extracted according to the task requirement information; and performing feature screening on the image features corresponding to each sub-image according to the feature type to obtain screened image features.

[0131] Optionally, the processor may also execute the program code of the following steps: determining a multi-layer target sub-image based on the number of target layers, wherein the multi-layer target sub-image is a multi-layer sub-image in the multi-layer sub-image that is feature-fused with the current sub-image; obtaining image features corresponding to the multi-layer target sub-image; and feature-fusing the image features corresponding to the multi-layer target sub-image to obtain target image features.

[0132] Optionally, the above-mentioned processor can also execute the program code of the following steps: after performing feature fusion on the image features based on the target number of layers to obtain the target image features corresponding to the computed tomography image, detect whether there are lesion features corresponding to the target image features in the preset lesion database; when there are lesion features corresponding to the target image features in the preset lesion database, determine the lesion information of the target object corresponding to the computed tomography image.

[0133] Optionally, the processor may also execute the program code of the following steps: after performing feature fusion on the image features based on the target number of layers to obtain the target image features corresponding to the computed tomography image, performing image segmentation on the computed tomography image based on the target image features to obtain a plurality of sub-region images; analyzing the plurality of sub-region images to determine the target sub-region image, wherein the target sub-region image at least includes the abnormal physical sign state of the target object corresponding to the computed tomography image; and determining the position information of the abnormal physical sign state in the target object based on the position information of the target sub-region image in the computed tomography image.

[0134] Optionally, the processor may also execute the program code of the following steps: reading a medical image to be identified, wherein the medical image to be identified is composed of multiple layers of sub-images, and the image information of the medical image to be identified includes at least the layer thickness corresponding to each layer of sub-image; extracting the image features of each layer of sub-image, and displaying the target image features obtained by fusing the image features of the sub-images of the target number of layers, wherein the target number of layers is obtained by processing the image information based on a preset depth layer in a preset neural network model; and displaying the recognition result of the medical image to be identified obtained by analyzing the target image features.

[0135] Optionally, the above-mentioned processor can also execute the program code of the following steps: segmenting the medical image to be identified based on the target image features to obtain multiple sub-region images; determining the target sub-region image with lesion features from the multiple sub-region images; and highlighting the position information of the target sub-region image in the medical image to be identified.

[0136] Optionally, the processor may also execute the program code of the following steps: obtaining the sub-image features corresponding to each sub-region image; detecting whether there are lesion features corresponding to the sub-image features in the preset lesion database; and upon detecting that there are lesion features corresponding to the sub-image features in the preset lesion database, determining that the sub-region image corresponding to the sub-image features is the target sub-region image.

[0137] Optionally, the above-mentioned processor can also execute the program code of the following steps: after highlighting the position information of the target sub-region image in the medical image to be identified, obtaining the lesion information corresponding to the target image feature from the preset lesion database, wherein the preset lesion database includes at least: preset image features, preset lesion information and the association relationship between the preset image features and the preset lesion information; displaying the lesion information.

[0138] Optionally, the medical image to be identified is a computed tomography image.

[0139] It can be understood by those skilled in the art that Figure 8 The structure shown is for illustration only, and the computing device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 8 It does not limit the structure of the above electronic device. For example, the computing device 10 may also include Figure 8 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 8 Different configurations shown.

[0140] A person skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0141] Example 7

[0142] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store program codes executed by the method for extracting image features and the method for recognizing medical images provided in the above embodiment.

[0143] Optionally, in this embodiment, the above-mentioned storage medium may be located in any computing device in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0144] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target number of layers corresponding to the computed tomography image; performing feature extraction on each layer of sub-image to obtain image features corresponding to each layer of sub-image; and performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image.

[0145] Optionally, in this embodiment, the storage medium is configured to store program codes for executing the following steps: extracting information from a computed tomography image based on a two-dimensional neural network model to obtain image features corresponding to each layer of sub-image.

[0146] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining task requirement information corresponding to a preset task; determining a target level of a preset depth layer in a preset neural network model based on the task requirement information, wherein the task requirement information and the target level have an associated relationship; determining a preset receptive field based on the task requirement information, wherein the preset receptive field is a mapping area range of pixel points on a feature map corresponding to each layer of sub-image on each layer of sub-image; and calculating the target number of layers corresponding to the computed tomography image in the preset depth layer based on the target level and the preset receptive field.

[0147] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: after extracting features from each layer of sub-images to obtain image features corresponding to each layer of sub-images, determining the feature types of the image features to be extracted based on task requirement information; and performing feature screening on the image features corresponding to each layer of sub-images based on the feature types to obtain screened image features.

[0148] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: determining a multi-layer target sub-image based on the target number of layers, wherein the multi-layer target sub-image is a multi-layer sub-image in the multi-layer sub-image that is feature-fused with the current sub-image; obtaining image features corresponding to the multi-layer target sub-image; and feature-fusing the image features corresponding to the multi-layer target sub-image to obtain target image features.

[0149] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: after performing feature fusion on image features based on the target number of layers to obtain target image features corresponding to the computed tomography image, detecting whether there are lesion features corresponding to the target image features in the preset lesion database; when there are lesion features corresponding to the target image features in the preset lesion database, determining the lesion information of the target object corresponding to the computed tomography image.

[0150] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: after performing feature fusion on image features based on the target number of layers to obtain target image features corresponding to the computed tomography image, performing image segmentation on the computed tomography image based on the target image features to obtain multiple sub-region images; analyzing the multiple sub-region images to determine a target sub-region image, wherein the target sub-region image at least includes an abnormal physical sign state of the target object corresponding to the computed tomography image; and determining the position information of the abnormal physical sign state in the target object based on the position information of the target sub-region image in the computed tomography image.

[0151] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: reading a medical image to be identified, wherein the medical image to be identified is composed of multiple layers of sub-images, and the image information of the medical image to be identified includes at least the layer thickness corresponding to each layer of sub-image; extracting image features of each layer of sub-image, and displaying target image features obtained by fusing image features of sub-images of a target number of layers, wherein the target number of layers is obtained by processing image information based on a preset depth layer in a preset neural network model; and displaying the recognition result of the medical image to be identified obtained by analyzing the target image features.

[0152] Optionally, in this embodiment, the storage medium is configured to store program codes for executing the following steps: segmenting the medical image to be identified based on target image features to obtain multiple sub-region images; determining a target sub-region image having lesion features from the multiple sub-region images; and highlighting the position information of the target sub-region image in the medical image to be identified.

[0153] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining sub-image features corresponding to each sub-region image; detecting whether there is a lesion feature corresponding to the sub-image feature in a preset lesion database; and upon detecting that there is a lesion feature corresponding to the sub-image feature in the preset lesion database, determining that the sub-region image corresponding to the sub-image feature is the target sub-region image.

[0154] Optionally, in the embodiment, the storage medium is configured to store program code for performing the following steps: after highlighting the position information of the target sub-region image in the medical image to be identified, obtaining the lesion information corresponding to the target image feature from a preset lesion database, wherein the preset lesion database at least includes: a preset image feature, preset lesion information, and an association relationship between the preset image feature and the preset lesion information; and displaying the lesion information.

[0155] Optionally, the medical image to be identified is a computed tomography image.

[0156] The serial numbers of the embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0157] In the above-described embodiments of the present application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0158] In the several embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. Of course, the device embodiment described above is only illustrative, and for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, unit or module, and can be electrical or other forms.

[0159] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment.

[0160] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0161] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0162] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for extracting image features, characterized in that: include: Acquiring image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of the sub-image; Processing the image information based on a preset depth layer in a preset neural network model to obtain a target layer number corresponding to the computed tomography image; Performing feature extraction on each layer of sub-images to obtain image features corresponding to each layer of sub-images; Performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image; Among them, the processing of the image information based on the preset depth layer in the preset neural network model to obtain the target number of layers corresponding to the computed tomography image includes: obtaining task requirement information corresponding to a preset task, wherein the task requirement information and the target level are correlated; determining the target level of the preset depth layer in the preset neural network model according to the task requirement information; determining a preset receptive field according to the task requirement information; and calculating the target number of layers corresponding to the computed tomography image in the preset depth layer according to the target level and the preset receptive field.

2. The method according to claim 1, characterized in that Performing feature extraction on each layer of sub-images to obtain image features corresponding to each layer of sub-images includes: Information is extracted from the computed tomography image based on a two-dimensional neural network model to obtain image features corresponding to each layer of sub-image.

3. The method according to claim 1, characterized in that The preset receptive field is a mapping area range of pixel points on the feature map corresponding to each layer of sub-images on each layer of sub-images.

4. The method according to claim 3, characterized in that After extracting features from the sub-images of each layer to obtain image features corresponding to the sub-images of each layer, the method further includes: Determining the feature type of the image feature to be extracted according to the task requirement information; Feature screening is performed on the image features corresponding to the sub-images of each layer according to the feature type to obtain screened image features.

5. The method according to claim 4, characterized in that Performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image includes: Determining a multi-layer target sub-image according to the target number of layers, wherein the multi-layer target sub-image is a multi-layer sub-image in the multi-layer sub-image that is subjected to feature fusion with the current sub-image; Obtaining image features corresponding to the multi-layer target sub-image; Feature fusion is performed on the image features corresponding to the multi-layer target sub-images to obtain the target image features.

6. The method according to claim 1, characterized in that After performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image, the method further includes: Detecting whether there is a lesion feature corresponding to the target image feature in a preset lesion database; When a lesion feature corresponding to the target image feature exists in the preset lesion database, lesion information of the target object corresponding to the computed tomography image is determined.

7. The method according to claim 1, characterized in that After performing feature fusion on the image features based on the target number of layers to obtain target image features corresponding to the computed tomography image, the method further includes: performing image segmentation on the computed tomography image based on the target image features to obtain a plurality of sub-region images; Analyzing the multiple sub-region images to determine a target sub-region image, wherein the target sub-region image at least includes an abnormal physical sign state of the target object corresponding to the computed tomography image; The position information of the abnormal vital sign state in the target object is determined according to the position information of the target sub-region image in the computed tomography image.

8. A medical image recognition method, characterized in that: include: Reading a medical image to be identified, wherein the medical image to be identified is composed of multiple layers of sub-images, and image information of the medical image to be identified includes at least a layer thickness corresponding to each layer of the sub-image; Extracting image features of each layer of sub-images, and displaying target image features obtained by fusing image features of sub-images of a target number of layers, wherein the target number of layers is calculated according to a target level and a preset receptive field in a preset depth layer in a preset neural network model, the preset receptive field is determined according to task requirement information, the target level is determined according to the task requirement information, the task requirement information and the target level are associated with each other and correspond to a preset task; Display the recognition result of the medical image to be recognized obtained by analyzing the target image features.

9. The method according to claim 8, characterized in that Displaying the recognition result of the medical image to be recognized obtained by analyzing the target image features, including: Segmenting the medical image to be identified based on the target image features to obtain a plurality of sub-region images; Determining a target sub-region image having lesion features from the multiple sub-region images; Position information of the target sub-region image in the medical image to be identified is highlighted in the medical image to be identified.

10. The method according to claim 9, characterized in that Determining a target sub-region image having abnormal features from the multiple sub-region images includes: Obtaining sub-image features corresponding to each sub-region image; Detecting whether there is a lesion feature corresponding to the sub-image feature in a preset lesion database; When it is detected that a lesion feature corresponding to the sub-image feature exists in the preset lesion database, the sub-region image corresponding to the sub-image feature is determined to be the target sub-region image.

11. The method according to claim 10, characterized in that After highlighting the position information of the target sub-region image in the medical image to be identified, the method further includes: Acquiring lesion information corresponding to the target image feature from the preset lesion database, wherein the preset lesion database includes at least: preset image features, preset lesion information, and an association relationship between the preset image features and the preset lesion information; The lesion information is displayed.

12. The method according to claim 8, characterized in that The medical image to be identified is a computed tomography image.

13. A storage medium, characterized in that The storage medium includes a stored program, wherein when the program is run, the device where the storage medium is located is controlled to execute the method for extracting image features described in any one of claims 1 to 7 and the method for recognizing medical images described in any one of claims 8 to 12.

14. A system for extracting image features, characterized in that: include: processor; as well as A memory, connected to the processor, configured to provide the processor with instructions for processing the following processing steps: Acquiring image information of a computed tomography image, wherein the computed tomography image includes multiple layers of sub-images, and the image information of the computed tomography image includes at least a layer thickness corresponding to each layer of the sub-image; processing the image information based on a preset depth layer in a preset neural network model to obtain a target layer number corresponding to the computed tomography image; performing feature extraction on each layer of the sub-image to obtain image features corresponding to each layer of the sub-image; and performing feature fusion on the image features based on the target layer number to obtain target image features corresponding to the computed tomography image; Among them, the processing of the image information based on the preset depth layer in the preset neural network model to obtain the target number of layers corresponding to the computed tomography image includes: obtaining task requirement information corresponding to a preset task, wherein the task requirement information and the target level are correlated; determining the target level of the preset depth layer in the preset neural network model according to the task requirement information; determining a preset receptive field according to the task requirement information; and calculating the target number of layers corresponding to the computed tomography image in the preset depth layer according to the target level and the preset receptive field.

Images