Image detection method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application disclose an image detection method and device, electronic equipment and storage medium. The method comprises: obtaining an ultrasonic image to be detected; using at least two algorithm sub-modules to correspondingly detect at least two different types of abnormal features of the ultrasonic image to be detected to obtain at least two groups of abnormal feature detection results; determining an abnormal detection result of the ultrasonic image to be detected according to the at least two groups of abnormal feature detection results, and the abnormal detection result is used at least to indicate whether the ultrasonic image to be detected is an abnormal image. This scheme can detect different types of abnormal features of the ultrasonic image to be detected by using multiple preset algorithms, thereby saving manpower and being relatively efficient.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more specifically, to an image detection method, apparatus, electronic device, storage medium, and computer program product. Background Technology

[0002] During the process of using an ultrasound imaging system, there may be problems such as probe array element damage, current signal interference, software failure or defects in the ultrasound imaging system, which may cause abnormalities in the obtained ultrasound images.

[0003] In the software development process of ultrasound imaging systems, testers usually need to manually inspect a large number of repetitive ultrasound images acquired by the ultrasound imaging system to find abnormal ultrasound images in order to analyze software problems. This method is time-consuming and inefficient. In addition, the software of ultrasound imaging systems is updated very quickly, and the speed of manual inspection of ultrasound images usually cannot keep up with the speed of software updates. Summary of the Invention

[0004] The present invention was proposed in view of the above-mentioned problems. Embodiments of the present invention provide an image detection method, apparatus, electronic device, storage medium, and computer program product. This solution can utilize multiple preset algorithms to detect different types of abnormal features in the ultrasound image under test, saving manpower and achieving high efficiency.

[0005] According to one aspect of the present invention, an image detection method is provided. The method includes: acquiring an ultrasound image to be tested; detecting at least two different types of abnormal features in the ultrasound image to be tested using at least two algorithm submodules in a one-to-one correspondence, to obtain at least two sets of abnormal feature detection results; and, based on the abnormal detection results of the at least two sets of abnormal feature detection results, the abnormal detection results are used at least to indicate whether the ultrasound image to be tested is an abnormal image.

[0006] Optionally, the anomalous features include at least two of the following: edge white bar features, noise block features, texture edge features, and layered features; the at least two algorithm sub-modules include at least two of the following: a first algorithm sub-module, a second algorithm sub-module, a third algorithm sub-module, and a fourth algorithm sub-module; wherein, at least two different types of anomalous features of the ultrasound image under test are detected one-to-one using the at least two algorithm sub-modules to obtain at least two sets of anomalous feature detection results, including at least the following at least two operations: detecting edge white bar features in the ultrasound image under test using the first algorithm sub-module; detecting noise block features in the ultrasound image under test using the second algorithm sub-module; detecting texture edge features in the ultrasound image under test using the third algorithm sub-module, where texture edge features include grid features and / or artifact features; and detecting layered features in the ultrasound image under test using the fourth algorithm sub-module.

[0007] Optionally, the third algorithm submodule is a residual network; wherein the third algorithm submodule includes some continuous residual blocks in the preset residual network, and the level of the some continuous residual blocks in the preset residual network is not the deepest; and / or, at least some convolutional layers in the third algorithm submodule use a preset operator for the convolution kernel, wherein the preset operator is used to detect texture edge features in the horizontal direction and texture edge features in the vertical direction in the input image.

[0008] Optionally, the number of channels corresponding to each convolutional layer of a partial continuous residual block in the residual network is less than the number of channels corresponding to the preset residual network.

[0009] Optionally, detecting edge white bar features in the ultrasound image under test using the first algorithm submodule includes: performing image segmentation on the ultrasound image under test using the first algorithm submodule to obtain corresponding image segmentation results, the image segmentation results being used to indicate the image region where the edge white bar features in the ultrasound image under test are located, and a set of abnormal feature detection results corresponding to the edge white bar features including the image segmentation results; and / or detecting noise block features in the ultrasound image under test using the second algorithm submodule includes: performing target detection on the ultrasound image under test using the second algorithm submodule to obtain corresponding target detection results, the target detection results including the location information of the noise block features in the ultrasound image under test and a first confidence level, the first confidence level being used to indicate the probability that the image region represented by the location information contains the noise block features, and a set of abnormal features corresponding to the noise block features including the image segmentation results; and / or detecting noise block features using the second algorithm submodule to obtain corresponding target detection results, the second algorithm submodule being used ... The feature detection results include target detection results; and / or, using a third algorithm submodule to detect texture edge features in the ultrasound image under test, including: using the third algorithm submodule to perform image classification on the ultrasound image under test to obtain corresponding image classification results, the image classification results including a second confidence level, the second confidence level being used to indicate the probability that the ultrasound image under test contains texture edge features, and a set of abnormal feature detection results corresponding to the texture edge features including the image classification results; and / or, using a fourth algorithm submodule to detect layered features in the ultrasound image under test, including: using the fourth algorithm submodule to determine at least two image blocks in the ultrasound image under test, calculating the similarity between every two image blocks in the at least two image blocks to obtain corresponding image processing results, and a set of abnormal feature detection results corresponding to the layered features including the image processing results.

[0010] Optionally, calculating the similarity between any two image patches in at least two image patches includes: calculating at least two hash values ​​that correspond one-to-one with the at least two image patches, and calculating the Hamming distance between any two hash values ​​in the at least two hash values, wherein the Hamming distance is used to indicate the similarity between the corresponding two image patches.

[0011] Optionally, the third algorithm submodule is trained as follows: an initial training set is obtained, which includes at least one positive sample image and at least one negative sample image, where the positive sample image is an abnormal image and the negative sample image is a normal image; an iterative training operation is performed on the third algorithm submodule, each training operation including: image enhancement of at least a portion of the at least one positive sample image to obtain at least one enhanced image; adding the at least one enhanced image to the initial training set to obtain a new training set; and training the third algorithm submodule using the new training set.

[0012] Optionally, the image enhancement degree of the enhanced images used in the earlier training operations is greater than that of the enhanced images used in the later training operations, and each training operation includes one or more consecutive training operations.

[0013] Optionally, the abnormal detection result of the ultrasound image under test is determined based on at least two sets of abnormal feature detection results, including: performing a weighted summation on the normalized results corresponding to each of the at least two sets of abnormal feature detection results; comparing the weighted summation result with a preset threshold; if the weighted summation result is greater than the preset threshold, the ultrasound image under test is determined to be an abnormal image; if the weighted summation result is less than or equal to the preset threshold, the ultrasound image under test is determined to be a normal image.

[0014] Optionally, the second confidence level is the normalized result corresponding to a set of abnormal feature detection results corresponding to the texture edge features; and / or, before performing a weighted summation of the normalized results corresponding to each of the at least two sets of abnormal feature detection results, determining the abnormal detection result of the ultrasound image to be tested based on the at least two sets of abnormal feature detection results further includes: determining the area ratio between the image region where the edge white bar feature indicated by the image segmentation result is located and the overall image region of the ultrasound image to be tested as the normalized result corresponding to a set of abnormal feature detection results corresponding to the edge white bar feature; and / or, determining the maximum value in the first confidence level as the normalized result corresponding to a set of abnormal feature detection results corresponding to the noise block feature; and / or, determining the normalized result corresponding to a set of abnormal feature detection results corresponding to the layered feature based on the ratio between the minimum and maximum values ​​in the similarity.

[0015] According to another aspect of the present invention, an image detection apparatus is also provided, comprising: an acquisition module for acquiring an ultrasound image to be tested; a detection module for detecting at least two different types of abnormal features of the ultrasound image to be tested using at least two algorithm sub-modules in a one-to-one correspondence to obtain at least two sets of abnormal feature detection results; and a determination module for determining whether the ultrasound image to be tested is an abnormal image based on the at least two sets of abnormal feature detection results to obtain an abnormality detection result.

[0016] According to another aspect of the present invention, an electronic device is also provided, comprising: a processor and a memory, wherein the memory stores computer program instructions, which are executed by the processor to perform the image detection method described above.

[0017] According to another aspect of the present invention, a storage medium is also provided, on which program instructions are stored, which are used to execute the above-described image detection method when running.

[0018] According to another aspect of the present invention, a computer program product is also provided, including computer program instructions, which, when executed, are used to perform the image detection method as described above.

[0019] The aforementioned technical solution utilizes multiple algorithm sub-modules to detect various types of abnormal features in the ultrasound image under test, enabling automated ultrasound image detection and reducing manual operations. Furthermore, this approach allows for targeted detection of each type of abnormal feature based on the algorithmic characteristics of each sub-module, thus improving the accuracy of detection for each type of abnormal feature. On the other hand, since the anomaly detection results are obtained by combining multiple sets of anomaly feature detection results, they can comprehensively represent the overall situation of abnormal features contained in the ultrasound image under test, providing strong reference value. This solution effectively reduces the workload of testing personnel in manually inspecting a large number of repetitive ultrasound images, is time-efficient, and highly effective.

[0020] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0021] The above and other objects, features, and advantages of the present invention will become more apparent from the more detailed description of the embodiments of the invention in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same parts or steps.

[0022] Figure 1 A schematic flowchart of an image detection method according to an embodiment of the present invention is shown;

[0023] Figure 2 A comparative schematic diagram of normal and abnormal images according to an embodiment of the present invention is shown;

[0024] Figure 3A schematic flowchart of a fourth algorithm submodule according to an embodiment of the present invention is shown;

[0025] Figure 4 A schematic flowchart of a method for determining abnormal images according to an embodiment of the present invention is shown;

[0026] Figure 5 A schematic block diagram of an image detection apparatus according to an embodiment of the present invention is shown;

[0027] Figure 6 A schematic block diagram of an electronic device according to an embodiment of the present invention is shown. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0029] To at least partially solve the aforementioned technical problems, embodiments of the present invention provide an image detection method, apparatus, electronic device, storage medium, and computer program product. This solution can utilize multiple preset algorithms to detect different types of abnormal features in the ultrasound image under test, saving manpower and achieving high efficiency.

[0030] Please see Figure 1 The diagram shown is a schematic flowchart of an image detection method according to an embodiment of the present invention. According to one aspect of the present invention, an image detection method is provided. The method includes steps S110, S120, and S130.

[0031] In step S110, the ultrasound image to be tested is acquired.

[0032] For example, the ultrasound image to be tested can be an ultrasound image acquired by an ultrasound imaging system in any of the following imaging modes: brightness mode (B-mode), motion mode (M-mode), color Doppler mode, power Doppler mode, contrast imaging, elastography, etc. Ultrasound images acquired in brightness mode are preferred. For ease of description and understanding, this embodiment of the invention uses an ultrasound image acquired in brightness mode as an example.

[0033] In step S120, at least two different types of abnormal features in the ultrasound image to be tested are detected one-to-one using at least two algorithm sub-modules to obtain at least two sets of abnormal feature detection results.

[0034] For example, the abnormal feature types of the ultrasound image to be tested may include edge white bar features, noise block features, texture edge features, layered features, etc. For different types of abnormal features, multiple algorithm sub-modules can be used to detect multiple types of abnormal features one-to-one. At least two algorithm sub-modules may include, for example, an image segmentation algorithm sub-module, specifically such as a U-shaped network (Unet), a segmentation network (SegNet), or a fully convolutional densely connected network (Fully Convolutional DenseNets). At least two algorithm sub-modules may also include, for example, an object detection algorithm sub-module, specifically such as a Region-based Convolutional Neural Network (R-CNN), a You Only Look Once (YOLO) network, or a Single Shot MultiBox Detector (SSD). At least two algorithm sub-modules may also include, for example, an image classification algorithm sub-module, specifically such as a ResNet or a Densely Connected Convolutional Network (DenseNet). At least two algorithm submodules may also include an algorithm submodule for calculating the Hamming distance between the hash values ​​of each image patch within the image. Each algorithm submodule can be used to obtain a set of abnormal feature detection results for the ultrasound image under test.

[0035] In step S130, the abnormality detection result of the ultrasound image under test is determined based on at least two sets of abnormal feature detection results. The abnormality detection result is used to indicate at least whether the ultrasound image under test is an abnormal image.

[0036] For example, it can be determined whether the ultrasound image under test is an abnormal image based on at least two sets of abnormal feature detection results. In some embodiments, each set of abnormal feature detection results can at least indicate whether the ultrasound image under test contains the corresponding abnormal feature, or at least indicate the probability that the ultrasound image under test contains the corresponding abnormal feature. When any set of abnormal feature detection results indicates that the ultrasound image under test contains the corresponding abnormal feature, or when the probability that the ultrasound image under test contains the corresponding abnormal feature is greater than or equal to a specific probability threshold, the abnormal detection result can at least indicate that the ultrasound image under test is an abnormal image, and can also optionally indicate the type of abnormal feature contained in the ultrasound image under test, and can also optionally indicate the probability that the ultrasound image under test contains a certain type of abnormal feature. In other embodiments, each set of abnormal feature detection results can be normalized to obtain their respective corresponding normalized results. Based on the comparison between the weighted sum / average / maximum / minimum value of each normalized result and a preset threshold, the abnormal detection result of the ultrasound image under test can be determined. Accordingly, the abnormal detection result can at least indicate whether the ultrasound image under test is an abnormal image.

[0037] The aforementioned technical solution utilizes multiple algorithm sub-modules to detect various types of abnormal features in the ultrasound image under test, enabling automated ultrasound image detection and reducing manual operations. Furthermore, this approach allows for targeted detection of each type of abnormal feature based on the algorithmic characteristics of each sub-module, thus improving the accuracy of detection for each type of abnormal feature. On the other hand, since the anomaly detection results are obtained by combining multiple sets of anomaly feature detection results, they can comprehensively represent the overall situation of abnormal features contained in the ultrasound image under test, providing strong reference value. This solution effectively reduces the workload of testing personnel in manually inspecting a large number of repetitive ultrasound images, is time-efficient, and highly effective.

[0038] Optionally, the anomalous features include at least two of the following: edge white bar features, noise block features, texture edge features, and layered features; the at least two algorithm sub-modules include at least two of the following: a first algorithm sub-module, a second algorithm sub-module, a third algorithm sub-module, and a fourth algorithm sub-module; wherein, at least two different types of anomalous features of the ultrasound image under test are detected one-to-one using the at least two algorithm sub-modules to obtain at least two sets of anomalous feature detection results, including at least the following at least two operations: detecting edge white bar features in the ultrasound image under test using the first algorithm sub-module; detecting noise block features in the ultrasound image under test using the second algorithm sub-module; detecting texture edge features in the ultrasound image under test using the third algorithm sub-module, where texture edge features include grid features and / or artifact features; and detecting layered features in the ultrasound image under test using the fourth algorithm sub-module.

[0039] For example, in step S120, a first algorithm submodule can be used to detect edge white bar features in the ultrasound image to be tested. The first algorithm submodule can be, for example, the image segmentation algorithm submodule or the object detection algorithm submodule in the aforementioned embodiments. It is preferable to use the image segmentation algorithm submodule to detect edge white bar features, and the image segmentation algorithm submodule is preferably a U-shaped network. A second algorithm submodule can also be used to detect noise block features in the ultrasound image to be tested. Similarly, the second algorithm submodule can be, for example, the image segmentation algorithm submodule or the object detection algorithm submodule in the aforementioned embodiments. It is preferable to use the object detection algorithm submodule, and the object detection algorithm submodule is preferably a single detection network version 8 (YOLOv8). A third algorithm submodule can also be used to detect texture edge features in the ultrasound image to be tested. Texture edge features can include, for example, one or both of grid features and artifact features. The third algorithm submodule can be, for example, the image classification algorithm submodule in the aforementioned embodiments. The image classification algorithm submodule is preferably a residual network-50 layers (ResNet50) or a ResNet50 with network architecture adjusted and / or convolution kernel adjusted. A fourth algorithm submodule can also be used to detect layered features in the ultrasound image under test. For example, the fourth algorithm submodule can calculate the Hamming distance between the hash values ​​of each image patch in the image to obtain the layered feature detection results. Anomalous features can include at least two of the following: edge white bar features, noise block features, texture edge features, and layered features.

[0040] Please see Figure 2 As shown, it is a schematic diagram comparing a normal image and an abnormal image according to an embodiment of the present invention. Figure 2 (a) shows a normal ultrasound image. Within the image area (fan-shaped area) of a normal ultrasound image, the grayscale values ​​of this ultrasound image are all within a specific grayscale range. Edge white bars are typically located at the left and right edges of the ultrasound image, with white or black borders, such as... Figure 2 (b) shows an abnormal ultrasound image containing edge white bars (a "white bar" is present at the right edge of the fan-shaped area). Noise blocks are typically located in the center of the ultrasound image and are accompanied by white or black acoustic noise points, such as... Figure 2 (c) shows an abnormal ultrasound image containing noise block features. Layering features are typically located in the middle of the image, forming dark or light layers, such as... Figure 2 (d) shows an abnormal ultrasound image containing layered features (one or more stripes exist in the fan-shaped region, the gray values ​​of which are outside a specific gray range and the gray value range is within a preset difference range). The grid features are mostly located in the center of the image, forming disordered patterns, such as... Figure 2(e) The ultrasound image shown is an abnormal image containing grid features (multiple textures, such as disordered patterns, exist in the fan-shaped region; at least some textures have gray values ​​outside a specific gray range, and the gray value range is outside a preset difference range). Artifact features are generally located in the center of the image and have snowflake-like ripples, such as... Figure 2 (f) shows an abnormal image containing artifact features (at least a portion of the sub-regions in the fan-shaped region have gray values ​​outside a specific gray range, and the image contains snowflake-like ripples).

[0041] The above technical solution utilizes at least two algorithm sub-modules to detect at least two abnormal features among edge white bar features, noise block features, texture edge features, and layering features. This can combine the characteristics of ultrasound images themselves to specifically detect possible abnormal features in ultrasound images. This detection method can be well adapted to the image characteristics of ultrasound images, thus obtaining anomaly detection results with strong reference value.

[0042] Optionally, the third algorithm submodule is a residual network; wherein the third algorithm submodule includes some continuous residual blocks in the preset residual network, and the level of the some continuous residual blocks in the preset residual network is not the deepest; and / or, at least some convolutional layers in the third algorithm submodule use a preset operator for the convolution kernel, wherein the preset operator is used to detect texture edge features in the horizontal direction and texture edge features in the vertical direction in the input image.

[0043] For example, the third algorithm submodule can be a residual network, and the residual blocks in the residual network can be ordinary residual blocks or bottleneck residual blocks. The preset residual network can be, for example, a ResNet34 network, a ResNet50 network, or a ResNet101 network, with a ResNet50 network being preferred. The third algorithm submodule may include some residual blocks from the preset residual network. The layers of these residual blocks are continuous, and the deepest residual block in these residual blocks is not the deepest layer in the preset residual network. In a specific embodiment, the preset residual network can be a ResNet50 network, and the third algorithm submodule may not include the last residual block (i.e., the conv5_x layer) in the ResNet50. For example, since the third algorithm submodule is used to detect texture edge features, which belong to shallow features of the image, the number of layers in the third algorithm submodule can be less than the number of layers in the preset residual network. Specifically, the last x consecutive residual blocks of the preset residual network can be deleted in ascending order of depth. x can be defined by the user based on actual conditions. The retained residual blocks can be used as residual blocks in the third algorithm submodule. For example, at least some convolutional layers in the residual network can use preset operators as convolution kernels. These preset operators can be operators used to detect texture edge features in the horizontal and vertical directions of the input image. For example, preset operators can be Sobel, Prewitt, Scharr, etc., with Sobel being preferable. Taking Sobel as an example, the Sobel operator can include two convolution kernels in the horizontal and vertical directions, respectively used to detect texture edge features in the horizontal and vertical directions of the input image. By performing convolution operations with these two kernels, two edge intensity maps in the horizontal and vertical directions of the image can be obtained. Fusing the two edge intensity maps yields the image's edge feature map. Based on the edge feature map, the probability that the image contains texture edge features can be obtained.

[0044] The above technical solution is based on the premise that texture edge features are shallow features. It uses a residual network with a relatively small number of layers and a relatively shallow layer depth to detect texture edge features. The detection speed is fast and the computational resources required are small.

[0045] Optionally, the number of channels corresponding to each convolutional layer of a partial continuous residual block in the residual network is less than the number of channels corresponding to the preset residual network.

[0046] For example, the number of channels in each convolutional layer of the third algorithm submodule is less than the number of channels in the corresponding convolutional layer of the preset residual network. In a specific embodiment, the preset residual network can be a ResNet50 network, and each residual block of the third algorithm submodule can include three convolutional layers: 1×1, 3×3, and 1×1. For each residual block, the number of channels in the three convolutional layers contained in the residual block can be reduced; for example, the number of channels can be reduced to 50% of the initial number.

[0047] The third algorithm submodule used in the above technical solution has fewer parameters, lower computational complexity for the ultrasound image under test, and higher efficiency in detecting texture edge features.

[0048] Optionally, detecting edge white bar features in the ultrasound image under test using the first algorithm submodule includes: performing image segmentation on the ultrasound image under test using the first algorithm submodule to obtain corresponding image segmentation results. The image segmentation results are used to indicate the image region where the edge white bar features in the ultrasound image under test are located. A set of abnormal feature detection results corresponding to the edge white bar features includes the image segmentation results.

[0049] For example, the first algorithm submodule can segment the ultrasound image to be tested. The resulting image segmentation result can indicate the location of the edge white bar feature in the ultrasound image to be tested, that is, the image region where the edge white bar feature is located in the ultrasound image to be tested. Correspondingly, a set of abnormal feature detection results corresponding to the edge white bar feature can include the image segmentation result. In a specific embodiment, the first algorithm submodule is a multi-scale feature representation model obtained by constructing a Feature Pyramid Network (FPN) based on a U-shaped network. It can extract multi-layer features of the input image, and after upsampling and fusion, it can form a multi-scale feature representation (e.g., a feature map). Based on the feature map, a segmented image with the same size as the input image can be obtained. Each pixel in the segmented image can correspond to a category label, which can indicate whether the corresponding pixel is located in the image region where the edge white bar feature is located. In this embodiment, the image segmentation result can include the segmented image and the category label corresponding to each pixel within the segmented pixel. More specifically, before using the first algorithm submodule to detect edge white bar features in the ultrasound image to be tested, the first algorithm submodule can be trained using sample ultrasound images containing edge white bar features. The label of each sample ultrasound image can indicate the location of the region of interest in the sample ultrasound image, that is, the image region where the edge white bar features are located in the sample ultrasound image.

[0050] Optionally, the second algorithm submodule is used to detect noise block features in the ultrasound image under test, including: using the second algorithm submodule to perform target detection on the ultrasound image under test to obtain the corresponding target detection result. The target detection result includes the location information of the noise block features in the ultrasound image under test and a first confidence level. The first confidence level is used to indicate the probability that the image region represented by the location information contains the noise block features. A set of abnormal feature detection results corresponding to the noise block features includes the target detection result.

[0051] For example, the second algorithm submodule can perform target detection on the ultrasound image under test. The corresponding target detection result can include location information indicating the location of the noise block feature in the ultrasound image under test (i.e., the image region where the noise block feature is located in the ultrasound image under test), and can also include a first confidence level indicating the probability that the image region represented by the location information contains the noise block feature. Accordingly, a set of abnormal feature detection results corresponding to the noise block feature can include the target detection result. In a specific embodiment, the second algorithm submodule can be a YOLOv8 network. The sample ultrasound images used to train the second algorithm submodule can include positive sample images and negative sample images. Among them, the sample ultrasound images containing the noise block feature can be used as positive samples, and the sample ultrasound images not containing abnormal features can be used as negative samples. The second algorithm submodule can use an anchor-free detection head, which has a faster detection speed and higher detection accuracy for features of the noise block feature type.

[0052] Optionally, the third algorithm submodule is used to detect texture edge features in the ultrasound image under test, including: using the third algorithm submodule to perform image classification on the ultrasound image under test to obtain the corresponding image classification result, the image classification result including a second confidence level, the second confidence level being used to indicate the probability that the ultrasound image under test contains texture edge features, and a set of abnormal feature detection results corresponding to the texture edge features including the image classification result.

[0053] For example, the third algorithm submodule can perform image classification on the ultrasound image under test, and the corresponding image classification result can include a second confidence score indicating the probability that the ultrasound image under test contains texture edge features. Correspondingly, a set of abnormal feature detection results corresponding to the texture edge features can include the image classification result. In a specific embodiment, the third algorithm submodule can be a ResNet50 network, which is a binary classification model. The sample ultrasound images used to train the third algorithm submodule can include positive sample images and negative sample images. Similarly, sample ultrasound images containing grid features and / or artifact features can be used as positive sample images, and sample ultrasound images not containing abnormal features can be used as negative sample images. The label corresponding to a positive sample image can be, for example, "1", and the label corresponding to a negative sample image can be, for example, "0".

[0054] Optionally, the fourth algorithm submodule is used to detect the layered features in the ultrasound image to be tested, including: using the fourth algorithm submodule to determine at least two image blocks in the ultrasound image to be tested, calculating the similarity between each pair of image blocks in the at least two image blocks to obtain the corresponding image processing results, and the set of abnormal feature detection results corresponding to the layered features includes the image processing results.

[0055] For example, please continue reading Figure 3 As shown, the fan-shaped region is the image region used to display ultrasound images. The fourth algorithm submodule can determine multiple image blocks within the image region of the ultrasound image to be tested. These multiple image blocks have the same shape and size, but their positions within the image region differ. The image processing result can be obtained based on the similarity between any two image blocks. Specifically, the gray-level histogram of each pair of image blocks can be calculated, and a distance metric (e.g., Euclidean distance, chi-square distance) can be used to compare the two histograms. The distance between the two image blocks represents their similarity. The mean square error between each pair of image blocks can also be calculated; a smaller mean square error indicates a greater similarity. The distance between the feature vectors of each pair of image blocks (e.g., cosine similarity, Euclidean distance) can also be calculated to determine the similarity between the two image blocks. No specific limitations are made here regarding the method for determining similarity. Correspondingly, the set of abnormal feature detection results corresponding to the hierarchical features can include the image processing result.

[0056] In the above technical solution, since each type of abnormal feature can have a corresponding distribution pattern, each algorithm submodule that can adapt to the corresponding distribution pattern is used to detect various types of abnormal features one by one, and the matching between the detection algorithm and the abnormal features is high.

[0057] Optionally, calculating the similarity between any two image patches in at least two image patches includes: calculating at least two hash values ​​that correspond one-to-one with the at least two image patches, and calculating the Hamming distance between any two hash values ​​in the at least two hash values, wherein the Hamming distance is used to indicate the similarity between the corresponding two image patches.

[0058] For example, in the fourth algorithm submodule, the hash value of each image patch can be calculated, and the Hamming distance between the two hash values ​​of each pair of image patches can be calculated. The similarity between the two image patches can be determined based on the Hamming distance. The smaller the Hamming distance, the greater the similarity between the two image patches. See also... Figure 3 As shown, it is a schematic flowchart of the fourth algorithm submodule according to an embodiment of the present invention. Figure 3 In the illustrated embodiment, the image region of the ultrasound image to be tested is a fan-shaped region in the image. After extracting the fan-shaped region, N rectangular image blocks can be determined. In this embodiment, N=3, and these three rectangular image blocks are located in... Figure 3 The ultrasound image to be tested is arranged in a triangular pattern. The fourth algorithm submodule can calculate the hash value of each of the three rectangular image blocks, and can calculate the Hamming distance between the hash values ​​of each pair of rectangular image blocks. Those skilled in the art will understand that the hash value of an image block refers to the image features of the image block being converted into a fixed-length string (i.e., a hash value) using a specific hash algorithm. The Hamming distance between hash values ​​can be used to determine the similarity between two image blocks; the smaller the Hamming distance, the more similar the two image blocks are. Based on the Hamming distance, it can be determined whether the ultrasound image to be tested contains layered features. It can be understood that the larger the Hamming distance, the greater the difference between the two image blocks. This difference can be considered as being due to at least one of the two image blocks containing layered features. Therefore, by using the Hamming distance, it can be determined whether the ultrasound image to be tested contains layered features, thereby determining whether the ultrasound image to be tested is abnormal.

[0059] The above technical solution determines the similarity between two image blocks by calculating the Hamming distance between the hash values ​​of every two image blocks in the ultrasound image to be tested. This method of calculating similarity has high robustness to some negligible small noise features and is fast in calculation.

[0060] Optionally, the third algorithm submodule is trained as follows: an initial training set is obtained, which includes at least one positive sample image and at least one negative sample image, where the positive sample image is an abnormal image and the negative sample image is a normal image; an iterative training operation is performed on the third algorithm submodule, each training operation including: image enhancement of at least a portion of the at least one positive sample image to obtain at least one enhanced image; adding the at least one enhanced image to the initial training set to obtain a new training set; and training the third algorithm submodule using the new training set.

[0061] For example, the training process of the third algorithm submodule may specifically include: obtaining an initial training set, which may include multiple sample ultrasound images, including at least one normal image without abnormal features and at least one abnormal image containing texture edge features, wherein the normal image can be used as a negative sample image and the abnormal image can be used as a positive sample image. An iterative training operation can be performed on the third algorithm submodule. In each training operation, at least a portion of the at least one positive sample image can be enhanced. The image enhancement method may include any one or more of the following: random cropping, horizontal flipping, color jittering, rotation, and Gaussian noise. The resulting enhanced image can be directly added to the initial training set, and / or replace the initial image in the initial training set to obtain a new training set. This training operation can then use the new training set to train the third algorithm submodule.

[0062] The above technical solution can expand the training set and improve the diversity of data in the training set by performing image enhancement on at least some positive sample images. In particular, there are few ultrasound images containing texture edge features, and it is relatively difficult to obtain positive sample images. Usually, the ratio of negative sample images to positive sample images in the initial training set is unbalanced. Image enhancement on positive sample images helps to solve the problem of the imbalance between the number of positive and negative sample images.

[0063] Optionally, the image enhancement degree of the enhanced images used in the earlier training operations is greater than that of the enhanced images used in the later training operations, and each training operation includes one or more consecutive training operations.

[0064] For example, the image enhancement level can be dynamically adjusted based on the number of training operations. For instance, several sets of training operations can be preset, each set including one or more consecutive training operations. The number of training operations in each set can be the same or different. Arranging the training operations in the order they occur during the training cycle, the image enhancement level of the enhanced images used in each training operation outside the first set can be reduced, and the image enhancement level corresponding to later training operations can be less than that of earlier training operations. Similarly, the image enhancement level of the enhanced images used in some training operations outside the first set can be reduced; similarly, in these training operations, the image enhancement level corresponding to later training operations can be less than that of earlier training operations. It should be noted that image enhancement level can refer to image enhancement intensity; for example, the contrast enhancement intensity of the enhanced images used in earlier training operations is greater than that of the enhanced images used in later training operations. Image enhancement level can also refer to image enhancement frequency, for example, the number of enhanced images used in earlier training operations is greater than that used in later training operations.

[0065] In the above technical solution, as the number of training iterations increases, by gradually reducing the image enhancement level of abnormal images, the third algorithm submodule can obtain a more suitable learning environment at different training stages, which is beneficial to improving the stability and convergence speed of the third algorithm submodule.

[0066] Optionally, the abnormal detection result of the ultrasound image under test is determined based on at least two sets of abnormal feature detection results, including: performing a weighted summation on the normalized results corresponding to each of the at least two sets of abnormal feature detection results; comparing the weighted summation result with a preset threshold; if the weighted summation result is greater than the preset threshold, the ultrasound image under test is determined to be an abnormal image; if the weighted summation result is less than or equal to the preset threshold, the ultrasound image under test is determined to be a normal image.

[0067] For example, each group of abnormal feature detection results can have its own corresponding normalized result. These normalized results are then weighted and summed. The resulting weighted sum can be compared with a preset threshold. If the weighted sum is greater than the preset threshold, the ultrasound image under test can be identified as an abnormal image; otherwise, it is identified as an abnormal image. Please refer to [link to relevant documentation]. Figure 4 The diagram shown is a schematic flowchart of a method for determining abnormal images according to an embodiment of the present invention. Figure 4In the illustrated embodiment, the first, second, third, and fourth algorithm submodules can be denoted as module A, module B, module C, and module D, respectively. Modules A, B, C, and D are used to detect corresponding abnormal features in the ultrasound image under test, resulting in four sets of abnormal feature detection results. The normalized results of these four sets of abnormal feature detection results are then weighted and summed, and the weighted sum is compared with a preset threshold. Based on the comparison with the preset threshold, it can be determined whether the ultrasound image is abnormal. It should be noted that this embodiment of the invention does not limit the weights of each normalized result, and these weights can be defined by the user according to actual detection needs.

[0068] The above technical solution performs a weighted summation of the normalized results corresponding to the abnormal feature detection results, and determines whether the ultrasound image under test is an abnormal image based on the comparison between the weighted summation result and the preset threshold. This can comprehensively measure the existence of various abnormal features in the ultrasound image under test and make a comprehensive judgment on whether the ultrasound image under test is an abnormal image. The judgment result of whether the ultrasound image under test is an abnormal image has strong comprehensiveness.

[0069] Optionally, the second confidence level is the normalized result corresponding to a set of abnormal feature detection results corresponding to the texture edge features; and / or, before performing a weighted summation of the normalized results corresponding to each of the at least two sets of abnormal feature detection results, determining the abnormal detection result of the ultrasound image to be tested based on the at least two sets of abnormal feature detection results further includes: determining the area ratio between the image region where the edge white bar feature indicated by the image segmentation result is located and the overall image region of the ultrasound image to be tested as the normalized result corresponding to a set of abnormal feature detection results corresponding to the edge white bar feature; and / or, determining the maximum value in the first confidence level as the normalized result corresponding to a set of abnormal feature detection results corresponding to the noise block feature; and / or, determining the normalized result corresponding to a set of abnormal feature detection results corresponding to the layered feature based on the ratio between the minimum and maximum values ​​in the similarity.

[0070] For example, for a set of abnormal feature detection results corresponding to texture edge features, the normalized result corresponding to this set of abnormal feature detection results is the second confidence level in the aforementioned embodiment. For example, if the second confidence level is equal to 0.8, then the normalized result can be 0.8. For a set of abnormal feature detection results corresponding to edge white bar features, the normalized result corresponding to this set of abnormal feature detection results is the area ratio between the image region where the edge white bar features indicated by the image segmentation result are located and the overall image region of the ultrasound image to be tested. For example, if the area of ​​the image region where the edge white bar features indicated by the image segmentation result are located is 16, and the area of ​​the overall image region of the ultrasound image to be tested is 128, then the normalized result can be 0.125. For a set of abnormal feature detection results corresponding to noise block features, the normalized result corresponding to this set of abnormal feature detection results is the maximum value of the first confidence level in the aforementioned embodiment. Specifically, the second algorithm submodule can output one or more bounding boxes for the ultrasound image under test, along with a first confidence score for each box. The first confidence score represents the probability that the image region corresponding to the predicted box in the ultrasound image under test contains noise block features. When outputting multiple first confidence scores for multiple predicted boxes, the maximum value among the first confidence scores can be used as the normalized result corresponding to the set of anomaly detection results. For example, if the confidence score for the first predicted box output by the second algorithm submodule is 0.6, the confidence score for the second predicted box is 0.2, and the confidence scores for the third predicted box are 0 and 3, then the normalized result can be 0.6. For a set of anomaly detection results corresponding to hierarchical features, the normalized result corresponding to this set of anomaly detection results can be the difference between 1 and the ratio between the minimum and maximum similarity values. Taking the Hamming distance between the hash values ​​of corresponding image patches in the ultrasound image under test as an example, the less obvious the layering features of the ultrasound image under test, the smaller the Hamming distances, the smaller the range of the Hamming distances, and the closer the ratio of the minimum to the maximum value is to 1. Conversely, the more obvious the layering features of the ultrasound image under test, the more discrete the numerical distribution of the Hamming distances, the larger the range of the Hamming distances, and the closer the ratio of the minimum to the maximum value is to 0. Let 'a' denote the ratio between the minimum and the maximum value, then the normalization result can be equal to 1-a. The smaller 'a' is, the larger the normalization result is. For example, if the Hamming distances are 0.5, 0.9, and 2, then the normalization result can be 0.75. It can be understood that the numerical range of each normalization result can be between 0 and 1.

[0071] The above technical solution can convert the detection results of each group of abnormal features to a specific numerical range, thereby making different types of abnormal features comparable. Then, mathematical operations can be performed on each type of abnormal feature to obtain quantitative abnormal detection results. These abnormal detection results are more intuitive and easier for users to analyze.

[0072] Please see Figure 5 The diagram shown is a schematic block diagram of an image detection apparatus according to an embodiment of the present invention. According to another aspect of the present invention, an image detection apparatus 500 is also provided, comprising:

[0073] The acquisition module 510 is used to acquire the ultrasound image to be tested;

[0074] The detection module 520 is used to detect at least two different types of abnormal features in the ultrasound image under test by using at least two algorithm sub-modules in a one-to-one correspondence, so as to obtain at least two sets of abnormal feature detection results.

[0075] The determination module 530 is used to determine the abnormality detection result of the ultrasound image under test based on at least two sets of abnormal feature detection results. The abnormality detection result is used to indicate at least whether the ultrasound image under test is an abnormal image.

[0076] Please see Figure 6 As shown, it is a schematic block diagram of an electronic device according to an embodiment of the present invention. According to another aspect of the present invention, an electronic device 600 is also provided. The electronic device 600 includes a processor 610 and a memory 620, wherein the memory 620 stores computer program instructions, which are executed by the processor 610 to perform the above-described image detection method.

[0077] According to another aspect of the present invention, a storage medium is also provided, on which program instructions are stored. When the program instructions are executed by a computer or processor, the computer or processor performs corresponding steps of the image detection method described in the embodiments of the present invention, and is used to implement corresponding modules in the image detection apparatus described in the embodiments of the present invention or corresponding modules in the image detection apparatus described above. The storage medium may, for example, include a memory card of a smartphone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. A computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0078] According to another aspect of the present invention, a computer program product is also provided, including computer program instructions, which, when executed, are used to perform the image detection method as described above.

[0079] Those skilled in the art can understand the specific implementation and beneficial effects of the above-described image detection device, electronic device, storage medium, and computer program product by reading the detailed description of the image detection method above, and for the sake of brevity, they will not be described in detail here.

[0080] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of the invention. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of the invention. All such changes and modifications are intended to be included within the scope of the invention as claimed in the appended claims.

[0081] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0083] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0084] Similarly, it should be understood that, in order to streamline the invention and aid in understanding one or more of the various aspects of the invention, features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the description of exemplary embodiments of the invention. However, this approach should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the corresponding claims, its inventive point lies in solving the corresponding technical problem with fewer features than all of those in a single disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of the invention.

[0085] Those skilled in the art will understand that, apart from the mutual exclusion of features, all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or apparatus so disclosed can be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0086] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the claims, any of the claimed embodiments can be used in any combination.

[0087] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some modules in the image detection apparatus according to embodiments of the present invention. The present invention can also be implemented as an apparatus program (e.g., a computer program and computer program product) for performing some or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0088] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0089] The above are merely specific embodiments or descriptions of the present invention, and the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An image detection method, characterized in that, The method includes: Acquire the ultrasound image to be tested; At least two different types of abnormal features in the ultrasound image under test are detected one-to-one using at least two algorithm sub-modules to obtain at least two sets of abnormal feature detection results; The abnormality detection result of the ultrasound image under test is determined based on the at least two sets of abnormal feature detection results, and the abnormality detection result is used at least to indicate whether the ultrasound image under test is an abnormal image.

2. The method of claim 1, wherein, The abnormal features include at least two of the following: edge white bar features, noise block features, texture edge features, and layered features; the at least two algorithm sub-modules include at least two of the following: a first algorithm sub-module, a second algorithm sub-module, a third algorithm sub-module, and a fourth algorithm sub-module; the step of using at least two algorithm sub-modules to detect at least two different types of abnormal features in the ultrasound image under test to obtain at least two sets of abnormal feature detection results includes the following at least two operations: The first algorithm submodule is used to detect edge white bar features in the ultrasound image under test; The second algorithm submodule is used to detect noise block features in the ultrasound image under test; The third algorithm submodule is used to detect texture edge features in the ultrasound image under test, the texture edge features including grid features and / or artifact features; The fourth algorithm submodule is used to detect the layered features in the ultrasound image to be tested.

3. The method of claim 2, wherein, The third algorithm submodule is a residual network; wherein... The third algorithm submodule includes a portion of continuous residual blocks in a preset residual network, and the portion of continuous residual blocks is not at the deepest level in the preset residual network; and / or, at least some of the convolutional layers in the third algorithm submodule use a preset operator for the convolution kernel, wherein the preset operator is used to detect texture edge features in the horizontal direction and texture edge features in the vertical direction in the input image.

4. The method of claim 3, wherein, The number of channels corresponding to each convolutional layer of the partial continuous residual block in the residual network is less than the number of channels corresponding to the preset residual network.

5. The method according to claim 2, characterized in that, The step of detecting edge white bar features in the ultrasound image under test using the first algorithm submodule includes: performing image segmentation on the ultrasound image under test using the first algorithm submodule to obtain corresponding image segmentation results, wherein the image segmentation results are used to indicate the image region where the edge white bar features are located in the ultrasound image under test, and a set of abnormal feature detection results corresponding to the edge white bar features include the image segmentation results; And / or, The step of detecting noise block features in the ultrasound image under test using the second algorithm submodule includes: performing target detection on the ultrasound image under test using the second algorithm submodule to obtain a corresponding target detection result. The target detection result includes the location information of the noise block feature in the ultrasound image under test and a first confidence level. The first confidence level is used to indicate the probability that the image region represented by the location information contains the noise block feature. A set of abnormal feature detection results corresponding to the noise block feature includes the target detection result. And / or, The step of detecting texture edge features in the ultrasound image under test using the third algorithm submodule includes: performing image classification on the ultrasound image under test using the third algorithm submodule to obtain a corresponding image classification result, wherein the image classification result includes a second confidence level, the second confidence level being used to indicate the probability that the ultrasound image under test contains the texture edge features, and a set of abnormal feature detection results corresponding to the texture edge features including the image classification result; And / or, The step of detecting layered features in the ultrasound image under test using the fourth algorithm submodule includes: using the fourth algorithm submodule to determine at least two image blocks in the ultrasound image under test, calculating the similarity between each pair of image blocks in the at least two image blocks to obtain the corresponding image processing results, and the set of abnormal feature detection results corresponding to the layered features includes the image processing results.

6. The method of claim 5, wherein, The calculation of the similarity between any two image patches in the at least two image patches includes: Calculate at least two hash values ​​that correspond one-to-one with the at least two image patches, and calculate the Hamming distance between every two hash values, the Hamming distance being used to indicate the similarity between the corresponding two image patches.

7. The method of claim 5, wherein, The third algorithm submodule is trained in the following manner: Obtain an initial training set, which includes at least one positive sample image and at least one negative sample image, wherein the positive sample image is an abnormal image and the negative sample image is a normal image; Perform iterative training operations on the third algorithm submodule, each training operation including: Image enhancement is performed on at least a portion of the at least one positive sample image to obtain at least one enhanced image; The at least one enhanced image is added to the initial training set to obtain a new training set; The third algorithm submodule is trained using the new training set.

8. The method of claim 7, wherein, The image enhancement used in the earlier training operations is more extensive than that used in the later training operations. Each training operation consists of one or more consecutive training operations.

9. The method according to any one of claims 1 to 7, characterized in that, The step of determining the abnormality detection result of the ultrasound image under test based on the at least two sets of abnormal feature detection results includes: The normalized results corresponding to each of the at least two sets of abnormal feature detection results are weighted and summed. The weighted summation result is compared with a preset threshold. If the weighted summation result is greater than the preset threshold, the ultrasound image to be tested is determined to be an abnormal image. If the weighted summation result is less than or equal to the preset threshold, the ultrasound image to be tested is determined to be a normal image.

10. The method according to claim 9 of claim 5, characterized in that, The second confidence level is the normalized result corresponding to a set of abnormal feature detection results corresponding to the texture edge features; and / or, Before performing a weighted summation of the normalized results corresponding to each of the at least two sets of abnormal feature detection results, the step of determining the abnormal detection result of the ultrasound image under test based on the at least two sets of abnormal feature detection results further includes: The area ratio between the image region containing the edge white bar feature indicated by the image segmentation result and the overall image region of the ultrasound image under test is determined as the normalized result corresponding to a set of abnormal feature detection results corresponding to the edge white bar feature; and / or, The maximum value in the first confidence score is determined as the normalized result corresponding to a set of abnormal feature detection results corresponding to the noise block feature; and / or, The normalization result corresponding to a set of abnormal feature detection results corresponding to the hierarchical feature is determined based on the ratio between the minimum and maximum values ​​in the similarity.

11. An image detection apparatus characterized by comprising: include: The acquisition module is used to acquire the ultrasound image to be tested; The detection module is used to detect at least two different types of abnormal features in the ultrasound image under test by using at least two algorithm sub-modules in a one-to-one correspondence, so as to obtain at least two sets of abnormal feature detection results; The determination module is used to determine whether the ultrasound image under test is an abnormal image based on the abnormal detection results of the at least two sets of abnormal feature detection results.

12. An electronic device comprising a processor and a memory, characterized in that The memory stores computer program instructions, which, when executed by the processor, are used to perform the image detection method as described in any one of claims 1-10.

13. A storage medium on which program instructions are stored, characterized in that, The program instructions are used to execute the image detection method as described in any one of claims 1-10 when the program is run.

14. A computer program product comprising computer program instructions, characterized in that, The computer program instructions, when executed, are used to perform the image detection method as described in any one of claims 1-10.

Images