A method, system, device and medium for segmenting ovarian lesions

Abstract

The present application discloses a method, system, device, and medium for segmenting ovarian lesions. The method comprises inputting an ultrasound image containing an ovarian region into an encoding module, extracting a multi-scale encoding feature map corresponding to the ultrasound image through the encoding module; inputting the encoding feature map with the smallest scale into a Radon projection module, outputting a fine-grained feature map through the Radon projection module; inputting each encoding feature map except the encoding feature map with the smallest scale and the fine-grained feature map into a decoding module, and outputting an ovarian lesion segmentation result through the decoding module. After obtaining the multi-scale encoding features, the present application performs spatial feature mining on the encoding feature map with the smallest scale through the Radon projection module to obtain fine-grained features. This operation not only converts the spatial features, but also retains the inherent properties of the spatial features before conversion, i.e., the encoding feature domain with the smallest scale, thus reducing the loss of key information and facilitating the improvement of the accuracy of the ovarian lesion segmentation results.

Description

Technical Field

[0001] The present application relates to the field of ultrasound technology, and in particular to a method, system, device and medium for segmenting ovarian lesions. Background Art

[0002] Ultrasound imaging is one of the preferred imaging techniques for screening and diagnosing ovarian lesions due to its noninvasive, radiation-free, real-time, and low-cost advantages. Ultrasound physicians' identification of ovarian lesions on ultrasound images is highly dependent on clinical experience and is subject to subjective variation. The quality of ultrasound images also affects lesion identification to a certain extent, making accurate assessment of ovarian lesions with blurred boundaries particularly challenging.

[0003] With the development of artificial intelligence, deep learning-based technologies have been widely studied in the field of medical image analysis and have achieved remarkable results, promising to reduce the clinical pressure on doctors. Although existing methods use deep learning to segment ovarian lesions in ultrasound images, these methods generally stack deep learning modules to construct neural networks for image processing and learning, neglecting the extraction of unique information through feature space transformation, which affects the accuracy of ovarian lesion segmentation.

[0004] Therefore existing technology still needs to be improved and improved. Summary of the Invention

[0005] The technical problem to be solved by the present application is to provide a method, system, device and medium for segmenting ovarian lesions in response to the deficiencies of the existing technology.

[0006] In order to solve the above technical problems, the first aspect of the present application provides a method for segmenting ovarian lesions, wherein a trained ovarian lesion segmentation model is applied, and the ovarian lesion segmentation model includes an encoding module, a Radon projection module, and a decoding module; the ovarian lesion segmentation method specifically includes:

[0007] Inputting an ultrasound image containing an ovarian region into a coding module, and extracting a multi-scale coding feature map corresponding to the ultrasound image through the coding module;

[0008] Inputting the smallest-scale coded feature map among the multi-scale coded feature maps into the Radon projection module, and outputting a fine-grained feature map through the Radon projection module;

[0009] Each encoding feature map except the encoding feature map with the smallest scale and the fine-grained feature map are input into the decoding module, and the ovarian lesion segmentation result is output through the decoding module.

[0010] The ovarian lesion segmentation method, wherein the Radon projection module includes a latent feature projection unit, a feature extraction unit, and a back projection unit; the output of the fine-grained feature map by the Radon projection module specifically includes:

[0011] The ovarian lesion segmentation method, wherein the Radon projection module includes a latent feature projection unit, a feature extraction unit, and a back projection unit; the output of the fine-grained feature map by the Radon projection module specifically includes:

[0012] Performing a Radon transform on the smallest-scale encoding feature map at a preset angle by the latent feature projection unit to obtain a mapping feature map, wherein the mapping feature map includes a channel dimension, an angle dimension, and a width dimension;

[0013] Performing feature extraction on the mapping feature map from a channel dimension and an angle dimension by the feature extraction unit to output a channel feature map and an angle feature map;

[0014] The back projection unit performs a back Radon transform on the channel feature map and the angle feature map to output a fine-grained feature map.

[0015] The ovarian lesion segmentation method, wherein the preset angle includes multiple preset angles, and the latent feature projection unit performs Radon transform on the smallest-scale encoding feature map at the preset angle to obtain a mapping feature map specifically includes:

[0016] The latent feature projection unit performs a Radon transform on the smallest-scale encoding feature map at each preset angle to obtain a mapping feature map.

[0017] In the ovarian lesion segmentation method, the angular dimension of the mapping feature graph is equal to the number of preset angles.

[0018] The ovarian lesion segmentation method, wherein the feature extraction unit extracts features from the mapping feature map from the channel dimension and the angle dimension to output the channel feature map and the angle feature map, specifically includes:

[0019] Splitting the mapping feature map into a plurality of channel dimension features according to the channel dimension and into a plurality of angle dimension features according to the angle dimension by the feature extraction unit;

[0020] Input several angle dimension features into the transformer, and determine the angle feature map through the transformer;

[0021] Several channel dimension features are input into the transformer, and the dimension feature map is determined by the transformer.

[0022] The ovarian lesion segmentation method, wherein the performing of a reverse Radon transform on the channel feature map and the angle feature map by the reverse projection unit to output a fine-grained feature map specifically includes:

[0023] Performing a reverse Radon transform on the channel feature map and the angle feature map respectively by the reverse projection unit to obtain a reverse channel feature map and a reverse angle feature map;

[0024] A fine-grained feature map is determined by the reverse projection unit based on the reverse channel feature map, the reverse angle feature map, and the encoding feature map with the smallest scale.

[0025] The ovarian lesion segmentation method, wherein the determining of the fine-grained feature map by the back projection unit based on the back channel feature map, the back angle feature map, and the smallest-scale encoding feature map specifically includes:

[0026] Cascading along the channel dimension based on the reverse channel feature map, the reverse angle feature map, and the encoding feature map with the smallest scale by the reverse projection unit to form a cascade feature map;

[0027] The convolution calculation is performed on the cascade feature map through the back projection unit to obtain a fine-grained feature map, and the image scale of the fine-grained feature map is the same as the image scale of the smallest-scale encoding feature map.

[0028] A second aspect of the present application provides an ovarian lesion segmentation system, which uses a trained ovarian lesion segmentation model, wherein the ovarian lesion segmentation model includes an encoding module, a Radon projection module, and a decoding module; the ovarian lesion segmentation system specifically includes:

[0029] an acquisition module, configured to acquire an ultrasound image including an ovarian region;

[0030] A control module is configured to input an ultrasound image containing an ovarian region into an encoding module, extract a multi-scale encoding feature map corresponding to the ultrasound image through the encoding module; input the encoding feature map with the smallest scale among the multi-scale encoding feature maps into the Radon projection module, output a fine-grained feature map through the Radon projection module; and input each encoding feature map except the encoding feature map with the smallest scale and the fine-grained feature map into the decoding module, and output an ovarian lesion segmentation result through the decoding module.

[0031] A third aspect of the present application provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in any of the above-mentioned methods for segmenting ovarian lesions.

[0032] A fourth aspect of the present application provides a terminal device, comprising: a processor and a memory;

[0033] The memory stores a computer-readable program executable by the processor;

[0034] When the processor executes the computer-readable program, the processor implements the steps in any of the above methods for segmenting ovarian lesions.

[0035] Beneficial effects: Compared with the prior art, the present application provides a method, system, device and medium for segmenting ovarian lesions. The method includes inputting an ultrasound image containing an ovarian region into an encoding module, extracting a multi-scale encoding feature map corresponding to the ultrasound image through the encoding module; inputting the encoding feature map with the smallest scale in the multi-scale encoding feature map into the Radon projection module, outputting a fine-grained feature map through the Radon projection module; inputting each encoding feature map except the encoding feature map with the smallest scale and the fine-grained feature map into the decoding module, and outputting the ovarian lesion segmentation result through the decoding module. After obtaining the multi-scale encoding features through the encoding module, the present application performs spatial feature mining on the encoding feature map with the smallest scale through the Radon projection module to obtain fine-grained features. Such an operation not only converts the spatial features, but also retains the inherent properties of the spatial features before the conversion, i.e., the encoding feature domain with the smallest scale, after the conversion, thereby reducing the loss of key information and helping to improve the accuracy of the ovarian lesion segmentation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0037] Figure 1 Block diagram of the working principle of the ovarian lesion segmentation model.

[0038] Figure 2 This is a flow chart of the ovarian lesion segmentation method provided in an embodiment of the present application.

[0039] Figure 3 This is the structural block diagram of the Radon projection module.

[0040] Figure 4 This is a principle block diagram of the ovarian lesion segmentation system provided in an embodiment of the present application.

[0041] Figure 5 This is a block diagram of the principles of the terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The present application provides a method, system, device, and medium for segmenting ovarian lesions. To clarify the purpose, technical solutions, and effects of this application, the present application is further described below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate this application and are not intended to limit this application.

[0043] It will be understood by those skilled in the art that, unless expressly stated otherwise, the singular forms "a", "an", "said" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, "connected" or "coupled" as used herein may include wireless connections or wireless couplings. The term "and / or" used herein includes all or any units and all combinations of one or more associated listed items.

[0044] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0045] It should be understood that the sequence numbers and sizes of the steps in this embodiment do not imply the order of execution. The order of execution of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of this application.

[0046] The application content will be further explained below through description of embodiments in conjunction with the accompanying drawings.

[0047] This embodiment provides a method for segmenting ovarian lesions, using a trained ovarian lesion segmentation model, such as Figure 1As shown, the ovarian lesion segmentation model includes an encoding module, a Radon projection module and a decoding module. The Radon projection module is located between the encoding module and the decoding module and is connected to the encoding module and the decoding module respectively. The encoding module is used to extract features from the ultrasound image to obtain a multi-scale encoding feature map. The Radon projection module is used to perform spatial learning on the encoding feature map with the smallest scale to obtain a fine-grained feature map. The decoding module is used to decode based on the fine-grained feature map and the encoding feature maps other than the encoding feature map with the smallest scale to obtain a decoding feature map, and determine the ovarian lesion segmentation result based on the decoding feature map. The embodiment of the present application embeds the Radon projection module in the U-structured network, and performs spatial feature learning through the Radon projection module while retaining the integrity of the source domain information to enrich the feature information included in the encoding feature, thereby improving the accuracy of the ovarian lesion segmentation result.

[0048] like Figure 2 As shown, the ovarian lesion segmentation method provided in the embodiment of the present application specifically includes:

[0049] S10. Input an ultrasound image containing an ovarian region into a coding module, and extract a multi-scale coding feature map corresponding to the ultrasound image through the coding module.

[0050] Specifically, the ultrasound image may be a real-time ultrasound image acquired in real time by an ultrasound device, an offline ultrasound image stored offline, or a remote ultrasound image sent by another external device. The ultrasound image includes the ovarian region, that is, the ultrasound image is obtained by performing an ovarian ultrasound examination on the patient. That is, before the ultrasound image including the ovarian region is input into the encoding module, an ultrasound image including the ovarian region is first obtained, and the obtained ultrasound image including the ovarian region may also be pre-processed, for example, by performing clarity detection, ovarian region detection, and noise removal.

[0051] After obtaining the ultrasound image containing the ovarian region, the ultrasound image containing the ovarian region is input into the encoding module, and the encoding module extracts features of the ultrasound image containing the ovarian region. The encoding module includes several encoding units, which are cascaded in sequence, and each encoding unit extracts a scaled encoding feature map. For example, assuming the image size of the ultrasound image is , Indicates the image height, Indicates the image width, such as Figure 1 As shown, the encoding module includes five encoding units. The image scales of the encoding feature maps extracted by each encoding unit in the cascade order are (48, 1), (96, 1 / 2), (192, 1 / 4), (384, 1 / 8), and (768, 1 / 16), where: Indicates that the image width of the coded feature map is the image width of the ultrasound image , the image height of the coded feature map is the image height of the ultrasound image (For example, 1 / 2 means the image height and image width of the encoded feature map are ).

[0052] It should be noted that, in actual applications, the number of coding units included in the coding module and the image scale of the coding feature map extracted by each coding unit can be set according to actual needs, and no specific restrictions are made here.

[0053] S20: Input the smallest-scale coding feature map among the multi-scale coding feature maps into the Radon projection module, and output a fine-grained feature map through the Radon projection module.

[0054] Specifically, the Radon projection module is used to learn spatial features from the smallest-scale coded feature map, so that the resulting fine-grained feature map contains geometric structure information, improving the accuracy of ovarian lesion segmentation results determined by decoding the fine-grained feature map. The Radon projection module is configured with a Radon transform and an inverse Radon transform, which learn spatial features while preserving the complete information of the smallest-scale coded feature map.

[0055] In one implementation, Figure 3 As shown, the Radon projection module includes a latent feature projection unit, a feature extraction unit, and a back projection unit; the latent feature projection unit is connected to the feature extraction unit, which is in turn connected to the back projection unit. The latent feature projection unit is used to perform a Radon transform to obtain a mapping feature map; the feature extraction unit is used to extract features from the mapping feature map in the channel dimension and the angle dimension to obtain a channel feature map and an angle feature map; the back projection unit is used to perform a reverse Radon transform on the channel feature map and the angle feature map to obtain a fine-grained feature map. Based on this, the output of the fine-grained feature map by the Radon projection module specifically includes:

[0056] S21, performing a Radon transform on the smallest-scale encoding feature map at a preset angle by the latent feature projection unit to obtain a mapping feature map;

[0057] S22. Performing feature extraction on the mapping feature map from a channel dimension and an angle dimension by the feature extraction unit to output a channel feature map and an angle feature map;

[0058] S23. Perform a reverse Radon transform on the channel feature map and the angle feature map through the reverse projection unit to output a fine-grained feature map.

[0059] Specifically, in step S21, the mapping feature map is obtained by performing a Radon transform on the smallest-scale encoding feature map at a preset angle. That is, compared to the existing Radon transform at the image level, the embodiment of the present application implements the Radon transform at the feature level. This can intervene in the expression form of the features during the feature learning process and can learn fine-grained features, thereby improving the accuracy of the ovarian lesion segmentation results. At the same time, after implementing the Radon transform at the feature level, the present application will also perform an inverse Radon transform on the learned features to back-project the features back to the spatial dimension of the smallest-scale encoding feature map.

[0060] Furthermore, the preset angle is pre-set and is used to determine the projection direction of the coding feature map with the smallest scale, wherein the preset angle can be multiple, for example, 0°, 30°, 60°, 90°, 120°, 150° and 180°.

[0061] In addition, the preset angle can be set according to the ultrasound image or user needs. For example, interactive information can be obtained by interacting with the user. The interactive information may include multiple preset angles or the angle interval between two adjacent preset angles (such as 15°, 30°, etc.). The preset angle is then determined according to the interactive information, and the preset angle is configured in the latent feature projection unit, so that the latent feature projection unit can map the smallest-scale encoding feature map at the preset angle to obtain a mapping feature map.

[0062] Exemplarily, performing a Radon transform on the smallest-scale encoding feature map at a preset angle by the latent feature projection unit to obtain a mapping feature map specifically includes:

[0063] The latent feature projection unit performs a Radon transform on the smallest-scale encoding feature map at each preset angle to obtain a mapping feature map.

[0064] Specifically, the mapping feature map is formed by accumulating the projections corresponding to each preset angle, and the projection corresponding to each preset angle is obtained by performing Radon transform on the preset angle of the smallest-scale coding feature map. The mapping feature map includes channel dimension, image width dimension and angle dimension. The image width of the mapping feature map is the same as the image width of the smallest-scale coding feature map, the number of channels of the mapping feature map is the same as the number of channels of the smallest-scale coding feature map, and the image height of the mapping feature map is equal to the number of preset angles. For example, Figure 3 As shown, the number of channels of the smallest-scale encoding feature map is , the image width of the smallest scale encoding feature map is , the preset number of angles is , then the image scale of the mapped feature map is .

[0065] Furthermore, in step S22, after obtaining the mapping feature map, feature extraction can be performed on the mapping feature map to extract finer-grained features. When extracting features from the mapping feature map, feature extraction is performed from the angle dimension and the channel dimension respectively to extract richer spatial feature information.

[0066] Exemplarily, the extracting features from the mapping feature map from the channel dimension and the angle dimension by the feature extraction unit to output the channel feature map and the angle feature map specifically includes:

[0067] Splitting the mapping feature map into a plurality of channel dimension features according to the channel dimension and into a plurality of angle dimension features according to the angle dimension by the feature extraction unit;

[0068] Input several angle dimension features into the transformer, and determine the angle feature map through the transformer;

[0069] Several channel dimension features are input into the transformer, and the channel feature map is determined by the transformer.

[0070] Specifically, the number of channel dimension features is equal to the number of channels of the mapping feature map, and the number of angle dimension features is equal to the preset number of angles corresponding to the mapping feature map. Of course, in practical applications, the mapping feature map can also be split in other ways, for example, the channel dimension and angle dimension of the mapping feature map are obtained, and then the channel dimension of each channel dimension feature is determined based on the obtained channel dimension, and the angle dimension of each angle dimension feature is determined based on the angle dimension. Specifically, the correspondence between the channel dimension and the channel splitting cardinality, as well as the correspondence between the angle dimension and the angle splitting cardinality, can be pre-set. After obtaining the channel dimension and the angle dimension of the mapping feature map, the channel splitting cardinality and the angle splitting cardinality corresponding to the mapping feature map are determined according to the two preset correspondences, and then the mapping feature map is split according to the channel splitting cardinality and the angle splitting cardinality, respectively, so that the number of channels of the channel dimension feature obtained by splitting is equal to the channel splitting cardinality, and the angle dimension of the angle dimension feature is equal to the angle splitting cardinality, wherein, when the number of channels of the last channel dimension feature can be less than the channel splitting cardinality, the angle dimension of the last angle dimension feature can be less than the angle splitting cardinality.

[0071] Furthermore, after obtaining a number of channel dimension features and a number of angle dimension features, the channel dimension features and the angle dimension features can be respectively input into a transformer, and the transformer is used to extract features of the channel dimension features and the angle dimension features to obtain a channel feature map and an angle feature map. Specifically, after splitting to obtain a number of channel dimension features, the channel dimension features are input into a transformer, and the transformer outputs the channel features corresponding to each channel dimension feature, and then all the channel features are cascaded according to the channel dimension to obtain a channel feature map. Similarly, after splitting to obtain a number of angle dimension features, the angle dimension features are input into a transformer, and the transformer outputs the angle features corresponding to each angle dimension feature, and then all the angle features are cascaded according to the angle dimension to obtain an angle feature map. Among them, the image scale of the channel feature map and the angle feature map is the same as the image scale of the mapping feature map. In addition, the transformer can adopt a Transformer network, which includes multi-head attention, layer normalization, multi-layer perceptron and layer normalization cascaded in sequence, and the input item of the multi-head attention is fused with the output item of the multi-head attention through an adder as the input item of the layer normalization of the multi-head attention, and the input item of the multi-layer perceptron is fused with the output item of the multi-layer perceptron through an adder as the input item of the layer normalization of the multi-head perceptron.

[0072] Further, in step S23, after obtaining the channel feature map and the angular feature map, the channel feature map and the angular feature map can be respectively subjected to a reverse Radon transform by a reverse projection unit to restore the channel feature map and the angular feature map to spatial features having the same size as the smallest-scale encoding feature map. Exemplarily, the reverse Radon transform of the channel feature map and the angular feature map by the reverse projection unit to output a fine-grained feature map specifically includes:

[0073] Performing a reverse Radon transform on the channel feature map and the angle feature map respectively by the reverse projection unit to obtain a reverse channel feature map and a reverse angle feature map;

[0074] The reverse channel feature map, the reverse angle feature map and the encoding feature map with the smallest scale are cascaded through the reverse projection unit to obtain a fine-grained feature map.

[0075] Specifically, if Figure 3 As shown, the reverse projection unit performs a reverse Radon transform on the channel feature map and the angle feature map to transform the channel feature map into a reverse channel feature map and transform the angle feature map into a reverse angle feature map, wherein the angle for implementing the reverse Radon transform is the same as the angle for implementing the Radon transform, that is, the channel feature map and the angle feature map are respectively subjected to a reverse Radon transform according to the preset angle corresponding to the Radon transform.

[0076] Furthermore, after obtaining the channel feature map and the angle feature map, the reverse channel feature map, the reverse angle feature map, and the smallest-scale encoding feature map are concatenated according to the channel dimension, and a convolution operation is performed on the concatenated feature map to obtain a fine-grained feature map. The image scale of the fine-grained feature map is the same as the image scale of the smallest-scale encoding feature map.

[0077] S30. Inputting each coding feature map except the coding feature map with the smallest scale and the fine-grained feature map into the decoding module, and outputting the ovarian lesion segmentation result through the decoding module.

[0078] Specifically, the fine-grained feature map is used to compare the encoding features of other scales except the smallest encoding feature map. Figure 1 The decoding module is used to decode the features so that the features are gradually restored to the size of the ultrasound image. Then, the ovarian lesion segmentation result is determined based on the decoding features output by the last decoding unit in the decoding module. For example, the decoding features output by the last decoding module are input into the sigmoid activation function to obtain a probability map. The probability map is converted to the image by taking the threshold. Figure 2 The ovarian lesion segmentation result can be a contour segmentation map of the lesion area or a bounding box of the lesion area.

[0079] In summary, this embodiment provides a method, system, device and medium for segmenting ovarian lesions. The method includes inputting an ultrasound image containing an ovarian area into a coding module, extracting a multi-scale coding feature map corresponding to the ultrasound image through the coding module; inputting the smallest-scale coding feature map in the multi-scale coding feature map into the Radon projection module, outputting a fine-grained feature map through the Radon projection module; inputting each coding feature map except the smallest-scale coding feature map and the fine-grained feature map into the decoding module, and outputting the ovarian lesion segmentation result through the decoding module. After obtaining the multi-scale coding features through the coding module, the present application performs spatial feature mining on the smallest-scale coding feature map through the Radon projection module to obtain fine-grained features. This not only learns the spatial feature information, but also allows the fine-grained features to retain the integrity of the smallest-scale coding feature domain information, which can significantly improve the accuracy of the ovarian lesion segmentation results.

[0080] Based on the above-mentioned ovarian lesion segmentation method, this embodiment provides an ovarian lesion segmentation system, which uses a trained ovarian lesion segmentation model, wherein the ovarian lesion segmentation model includes an encoding module, a Radon projection module, and a decoding module; Figure 4 As shown, the ovarian lesion segmentation system specifically includes:

[0081] An acquisition module 100 is used to acquire an ultrasound image including the ovarian region;

[0082] The control module 200 is configured to input an ultrasound image containing an ovarian region into an encoding module, extract a multi-scale encoding feature map corresponding to the ultrasound image through the encoding module, input the encoding feature map with the smallest scale among the multi-scale encoding feature maps into the Radon projection module, output a fine-grained feature map through the Radon projection module, and input each encoding feature map except the encoding feature map with the smallest scale and the fine-grained feature map into the decoding module, and output an ovarian lesion segmentation result through the decoding module.

[0083] Based on the above-mentioned method for segmenting ovarian lesions, this embodiment provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in the method for segmenting ovarian lesions as described in the above-mentioned embodiment.

[0084] Based on the above ovarian lesion segmentation method, the present application also provides a terminal device, such as Figure 5 As shown, it includes at least one processor 20; a display screen 21; and a memory 22. It may also include a communications interface 23 and a bus 24. The processor 20, display screen 21, memory 22, and communications interface 23 can communicate with each other via bus 24. The display screen 21 is configured to display a preset user guidance interface in the initial setup mode. The communications interface 23 can transmit information. The processor 20 can invoke logic instructions in the memory 22 to execute the method described in the above embodiment.

[0085] In addition, the logic instructions in the memory 22 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0086] The memory 22, as a computer-readable storage medium, can be configured to store software programs or computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes the software programs, instructions, or modules stored in the memory 22 to perform functional applications and data processing, thereby implementing the methods in the above embodiments.

[0087] The memory 22 may include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data created based on the use of the terminal device. In addition, the memory 22 may include high-speed random access memory and non-volatile memory. For example, various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, may also be transient storage media.

[0088] In addition, the specific process of loading and executing the multiple instructions in the storage medium and the processor in the terminal device has been described in detail in the above method and will not be described here one by one.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for segmenting ovarian lesions, characterized in that: A trained ovarian lesion segmentation model is applied, wherein the ovarian lesion segmentation model includes an encoding module, a Radon projection module, and a decoding module; and the ovarian lesion segmentation method specifically includes: Inputting an ultrasound image containing an ovarian region into a coding module, and extracting a multi-scale coding feature map corresponding to the ultrasound image through the coding module; Inputting the smallest-scale coded feature map among the multi-scale coded feature maps into the Radon projection module, and outputting a fine-grained feature map through the Radon projection module; Inputting each encoding feature map except the encoding feature map with the smallest scale and the fine-grained feature map into the decoding module, and outputting the ovarian lesion segmentation result through the decoding module; The Radon projection module includes a latent feature projection unit, a feature extraction unit, and a back projection unit; the preset angle includes multiple preset angles, and outputting a fine-grained feature map through the Radon projection module specifically includes: The latent feature projection unit performs a Radon transform on the smallest-scale encoding feature map at each preset angle to obtain a mapping feature map, wherein the mapping feature map includes a channel dimension, an angle dimension, and a width dimension; The feature extraction unit performs feature extraction on the mapping feature map from the channel dimension and the angle dimension respectively to output a channel feature map and an angle feature map; Performing a reverse Radon transform on the channel feature map and the angle feature map through the reverse projection unit to output a fine-grained feature map; The extracting features of the mapping feature map from the channel dimension and the angle dimension by the feature extraction unit to output the channel feature map and the angle feature map specifically includes: Splitting the mapping feature map into a plurality of channel dimension features according to the channel dimension and into a plurality of angle dimension features according to the angle dimension by the feature extraction unit; Input several angle dimension features into the transformer, and determine the angle feature map through the transformer; Input several channel dimension features into the transformer, and determine the dimension feature map through the transformer; The performing a reverse Radon transform on the channel feature map and the angle feature map by the reverse projection unit to output a fine-grained feature map specifically includes: Performing a reverse Radon transform on the channel feature map and the angle feature map respectively by the reverse projection unit to obtain a reverse channel feature map and a reverse angle feature map; A fine-grained feature map is determined by the reverse projection unit based on the reverse channel feature map, the reverse angle feature map, and the encoding feature map with the smallest scale.

2. The ovarian lesion segmentation method according to claim 1, characterized in that: The angular dimension of the mapping feature map is equal to the number of the preset angles.

3. The method for segmenting ovarian lesions according to claim 1, characterized in that: The determining of the fine-grained feature map based on the reverse channel feature map, the reverse angle feature map, and the minimum-scale encoding feature map by the reverse projection unit specifically includes: Cascading along the channel dimension based on the reverse channel feature map, the reverse angle feature map, and the encoding feature map with the smallest scale by the reverse projection unit to form a cascade feature map; The convolution calculation is performed on the cascade feature map through the back projection unit to obtain a fine-grained feature map, and the image scale of the fine-grained feature map is the same as the image scale of the smallest-scale encoding feature map.

4. A segmentation system for ovarian lesions, characterized in that: A trained ovarian lesion segmentation model is applied, wherein the ovarian lesion segmentation model includes an encoding module, a Radon projection module, and a decoding module; the ovarian lesion segmentation system specifically includes: an acquisition module, configured to acquire an ultrasound image including an ovarian region; A control module is configured to input an ultrasound image containing an ovarian region into an encoding module, extract a multi-scale encoding feature map corresponding to the ultrasound image through the encoding module; input the encoding feature map with the smallest scale among the multi-scale encoding feature maps into the Radon projection module, output a fine-grained feature map through the Radon projection module; and input each encoding feature map except the encoding feature map with the smallest scale and the fine-grained feature map into the decoding module, and output an ovarian lesion segmentation result through the decoding module; The Radon projection module includes a latent feature projection unit, a feature extraction unit, and a back projection unit; the preset angle includes multiple preset angles, and outputting a fine-grained feature map through the Radon projection module specifically includes: The latent feature projection unit performs a Radon transform on the smallest-scale encoding feature map at each preset angle to obtain a mapping feature map, wherein the mapping feature map includes a channel dimension, an angle dimension, and a width dimension; The feature extraction unit performs feature extraction on the mapping feature map from the channel dimension and the angle dimension respectively to output a channel feature map and an angle feature map; Performing a reverse Radon transform on the channel feature map and the angle feature map through the reverse projection unit to output a fine-grained feature map; The extracting features of the mapping feature map from the channel dimension and the angle dimension by the feature extraction unit to output the channel feature map and the angle feature map specifically includes: Splitting the mapping feature map into a plurality of channel dimension features according to the channel dimension and into a plurality of angle dimension features according to the angle dimension by the feature extraction unit; Input several angle dimension features into the transformer, and determine the angle feature map through the transformer; Input several channel dimension features into the transformer, and determine the dimension feature map through the transformer; The performing a reverse Radon transform on the channel feature map and the angle feature map by the reverse projection unit to output a fine-grained feature map specifically includes: Performing a reverse Radon transform on the channel feature map and the angle feature map respectively by the reverse projection unit to obtain a reverse channel feature map and a reverse angle feature map; A fine-grained feature map is determined by the reverse projection unit based on the reverse channel feature map, the reverse angle feature map, and the encoding feature map with the smallest scale.

5. A computer-readable storage medium, characterized in that The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the ovarian lesion segmentation method according to any one of claims 1 to 3.

6. A terminal device, characterized in that: include: processor and memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the processor implements the steps of the ovarian lesion segmentation method according to any one of claims 1 to 3.

Images