Methods, devices, and electronic equipment for identifying ultrasound images in vascular pathology
By using deep learning methods, the backbone feature extraction network of the U-Net network was replaced with the VGG16 network. Combined with residual structure and multi-scale convolutional structure, an image recognition model was trained, which solved the problem of difficult ultrasound image recognition of infantile hemangiomas and venous malformations, and achieved efficient and accurate image recognition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGHAN UNIVERSITY
- Filing Date
- 2022-12-01
- Publication Date
- 2026-06-30
AI Technical Summary
Infantile hemangiomas and venous malformations have similar ultrasound image characteristics, with rich lesion texture and blurred boundaries, making ultrasound detection and identification difficult and prone to misdiagnosis.
A deep learning approach was adopted, replacing the backbone feature extraction network of the U-Net network with a VGG16 network. Combined with residual structures and multi-scale convolutional structures, first and second image recognition models were trained. Feature extraction and recognition were performed using labeled masks to achieve automatic detection and recognition of vascular pathology ultrasound images.
It improves the recognition accuracy of ultrasound images for vascular pathology, and achieves efficient identification of infantile hemangiomas and venous malformations, with an accuracy rate of over 90%.
Smart Images
Figure CN115965587B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image recognition technology, and specifically to a method, apparatus, and electronic device for recognizing vascular pathology ultrasound images. Background Technology
[0002] The similar ultrasound features of infantile hemangiomas and venous malformations, the unique blood flow ultrasound image acquisition method of venous malformations, the rich texture of lesions in ultrasound images, the wide affected area, and the blurred boundaries all contribute to the difficulty in ultrasound detection and identification. If these ultrasound images are manually identified, the complex and difficult-to-master image characteristics can lead to misjudgments if the personnel lack sufficient experience and skills.
[0003] Therefore, there is a need to provide a method that can accurately and efficiently identify the category and location of lesions in vascular pathology ultrasound images. Summary of the Invention
[0004] In view of this, it is necessary to provide a method, device and electronic device for identifying vascular pathology ultrasound images, so as to accurately and efficiently identify the category and lesion location of vascular pathology ultrasound images.
[0005] To achieve the above objectives, the present invention provides a method for recognizing vascular pathology ultrasound images, comprising:
[0006] Obtain ultrasound image samples of the target size and the corresponding annotation masks of the ultrasound image samples; the ultrasound image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples.
[0007] The first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset first network model for training to obtain a first image recognition model and a first weight. The second type of vascular pathology ultrasound image sample after compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model and a second weight.
[0008] Based on the first image recognition model and the first weight, the ultrasound image to be tested is recognized to obtain the first recognition result;
[0009] If it is determined that there are ultrasound images of the second type of vascular pathology before compression and normal vascular ultrasound images in the first identification result, based on the second image recognition model and the second weight, the ultrasound images of the second type of vascular pathology after compression and the normal vascular ultrasound images corresponding to the ultrasound images of the second type of vascular pathology before compression in the first identification result are identified to obtain the second identification result.
[0010] The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network consists of convolutional layers and pooling layers in the VGG16 network.
[0011] Furthermore, the step of acquiring the ultrasound image sample of the target size and the corresponding annotation mask of the ultrasound image sample includes:
[0012] Acquire multiple types of blood flow ultrasound images; the multiple types of blood flow ultrasound images include: first type of vascular pathology ultrasound images, second type of vascular pathology ultrasound images, and normal ultrasound images;
[0013] The second type of vascular pathology ultrasound images were cropped into second type pre-compression ultrasound images and second type post-compression ultrasound images.
[0014] The first type of vascular pathology ultrasound image, the second type of ultrasound image before compression, the second type of ultrasound image after compression, and the normal ultrasound image are adjusted to the target size to obtain the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image before compression, the second type of vascular pathology ultrasound image after compression, and the normal vascular ultrasound image sample.
[0015] The ultrasound image samples of the first type of vascular pathology, the ultrasound image samples of the second type of vascular pathology before compression, the ultrasound image samples of the second type of vascular pathology after compression, and the ultrasound image samples of the normal blood vessels are labeled to obtain the corresponding labeling masks.
[0016] Furthermore, the step of annotating the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal vascular ultrasound image sample to obtain a corresponding annotation mask includes:
[0017] Based on a preset graphic frame, the lesion range in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal range in the normal vascular ultrasound image sample are labeled to obtain the corresponding labeling mask.
[0018] Furthermore, the step of annotating the lesion range in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal range in the normal vascular ultrasound image sample based on a preset graphic frame, to obtain a corresponding annotation mask, including:
[0019] Based on preset graphic frames of different colors, the range of different types of lesions in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, and the second type of vascular pathology ultrasound image sample after compression, as well as the normal range in the normal vascular ultrasound image sample, are marked to obtain the corresponding marking mask.
[0020] Further, the step of inputting the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, and the normal vascular ultrasound image samples, along with the corresponding labeled masks for each type of ultrasound image sample, into a preset first network model for training to obtain a first image recognition model and first weights includes:
[0021] The first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, and the normal vascular ultrasound image samples, along with the corresponding labeled masks for each type of ultrasound image sample, are input into a preset first network model for training to obtain a first image recognition model.
[0022] Based on the backbone feature extraction network of the first network model, downsampling features are extracted from the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, the normal vascular ultrasound image samples, and the labeled masks corresponding to various types of ultrasound image samples to obtain the preliminary effective feature layer of the first network model.
[0023] Based on the enhanced feature extraction network of the first network model, the preliminary effective feature layer of the first network model is upsampled and feature fused to obtain the target effective feature layer of the first network model;
[0024] The first weight is obtained by performing a convolution operation between the target effective feature layer of the first network model and a convolutional layer with the same number of recognition categories.
[0025] Furthermore, the step of inputting the ultrasound image samples of the second type of vascular pathological compression, the ultrasound image samples of normal blood vessels, and the corresponding labeled masks of various ultrasound image samples into a preset second network model for training, to obtain a second image recognition model and a second weight, includes:
[0026] The ultrasound image samples of the second type of vascular pathology after compression, the ultrasound image samples of normal vascular diseases, and the corresponding labeled masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model.
[0027] Based on the backbone feature extraction network of the second network model, downsampling feature extraction is performed on the ultrasound image samples after compression of the second type of vascular pathology, the ultrasound image samples of normal vascular diseases, and the labeled masks corresponding to various ultrasound image samples to obtain the preliminary effective feature layer of the second network model.
[0028] Based on the enhanced feature extraction network of the second network model, the preliminary effective feature layer of the second network model is upsampled and feature fused to obtain the target effective feature layer of the second network model;
[0029] The second weights are obtained by convolving the target effective feature layer of the second network model with a convolutional layer having the same number of recognition categories.
[0030] Furthermore, both the backbone feature extraction network and the enhanced feature extraction network include residual structures and multi-scale convolutional structures.
[0031] The present invention also provides a device for recognizing vascular pathology ultrasound images, comprising:
[0032] The acquisition module is used to acquire ultrasound image samples of the target size and the corresponding annotation mask of the ultrasound image samples; the ultrasound image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples.
[0033] The training module is used to input the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, the normal vascular ultrasound image samples, and the corresponding annotation masks of various ultrasound image samples into a preset first network model for training to obtain a first image recognition model and a first weight; and to input the second type of vascular pathology ultrasound image samples after compression, the normal vascular ultrasound image samples, and the corresponding annotation masks of various ultrasound image samples into a preset second network model for training to obtain a second image recognition model and a second weight.
[0034] The first recognition module is used to recognize the ultrasound image to be tested based on the first image recognition model and the first weight, and obtain a first recognition result.
[0035] The second identification module is used to identify, based on the second image recognition model and the second weight, the ultrasound image after compression of the second type of vascular pathology and the normal vascular ultrasound image corresponding to the ultrasound image before compression of the second type of vascular pathology in the first identification result, and the ultrasound image of the normal vascular pathology, when it is determined that there are pre-compression images of the second type of vascular pathology and normal vascular ultrasound images in the first identification result, to obtain the second identification result.
[0036] The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network consists of convolutional layers and pooling layers in the VGG16 network.
[0037] The present invention also provides an electronic device, including a memory and a processor, wherein,
[0038] The memory is used to store programs;
[0039] The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the method for recognizing vascular pathology ultrasound images as described above.
[0040] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for recognizing vascular pathology ultrasound images as described in any of the preceding claims.
[0041] The beneficial effects of the above implementation are as follows: The vascular pathology ultrasound image recognition method, device, and electronic device provided by the present invention, by inputting the first type of vascular pathology ultrasound image samples, the second type of vascular pathology pre-compression ultrasound image samples, and normal vascular ultrasound image samples, as well as the corresponding labeled masks of each type, into a preset first network model for training, a first image recognition model and a first weight are obtained. The second type of vascular pathology post-compression ultrasound image samples, the normal vascular ultrasound image samples, and the corresponding labeled masks of each type of ultrasound image sample are input into a preset second network model for training, a second image recognition model and a second weight are obtained. After the second type of vascular pathology pre-compression ultrasound image has its features extracted by the first image recognition model, the output is the recognition result of the second type of vascular pathology pre-compression image. The corresponding second type of vascular pathology post-compression ultrasound image is found through the second recognition result of the second type of vascular pathology pre-compression ultrasound image, and then recognized by the second image recognition model. By continuously increasing the number of ultrasound image samples for training and optimizing the weights of the image recognition model, the recognition accuracy of ultrasound images can be improved. Moreover, by using deep learning model training and detection, the automatic detection and recognition of target lesion features in ultrasound images can be achieved, thereby improving the efficiency of ultrasound image recognition. Attached Figure Description
[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0043] Figure 1 A flowchart illustrating an embodiment of the method for recognizing vascular pathology ultrasound images provided by the present invention;
[0044] Figure 2 The present invention provides a processing flow for various types of blood flow ultrasound images in one embodiment;
[0045] Figure 3 A schematic diagram of the process for generating weights for an image recognition model in one embodiment of the present invention;
[0046] Figure 4 A schematic diagram of the process for ultrasound image recognition based on an image recognition model and corresponding weights in one embodiment of the present invention;
[0047] Figure 5 This is a schematic diagram of an embodiment of the vascular pathology ultrasound image recognition device provided by the present invention;
[0048] Figure 6 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation
[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0050] In the description of the embodiments of this application, unless otherwise stated, "a plurality of" means two or more.
[0051] In this embodiment of the invention, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, apparatus, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product or device.
[0052] The naming or numbering of steps in the embodiments of the present invention does not mean that the steps in the method flow must be executed in the time / logical order indicated by the naming or numbering. The execution order of the named or numbered process steps can be changed according to the technical purpose to be achieved, as long as the same or similar technical effect can be achieved.
[0053] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0054] This invention provides a method, device, and electronic device for recognizing vascular pathology ultrasound images, which will be described below.
[0055] like Figure 1 As shown, the present invention provides a method for recognizing vascular pathology ultrasound images, comprising:
[0056] Step 110: Obtain ultrasound image samples of the target size and the corresponding annotation mask of the ultrasound image samples; the ultrasound image samples include ultrasound image samples of the first type of vascular pathology, ultrasound image samples of the second type of vascular pathology before compression, ultrasound image samples of the second type of vascular pathology after compression, and ultrasound image samples of normal blood vessels.
[0057] It is understandable that the first type of vascular pathology ultrasound image samples, the second type of vascular pathology pre-compression ultrasound image samples, the vascular pathology ultrasound image samples, the second type of vascular pathology post-compression ultrasound image samples, and normal vascular ultrasound image samples are all ultrasound image samples. The target size can be a preset size of 512x512 pixels. The first type of vascular pathology ultrasound image sample can be an ultrasound image sample of infantile hemangioma, the second type of vascular pathology pre-compression ultrasound image sample can be an ultrasound image sample of venous malformation pre-compression, the second type of vascular pathology post-compression ultrasound image sample can be an ultrasound image sample of venous malformation post-compression, and the normal vascular ultrasound image sample is also a normal blood flow ultrasound image sample. The annotation mask is the sample label required for the model training process.
[0058] Step 120: Input the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, the normal vascular ultrasound image samples, and the corresponding labeled masks of each type of ultrasound image sample into a preset first network model for training to obtain a first image recognition model and a first weight. Input the second type of vascular pathology ultrasound image samples after compression, the normal vascular ultrasound image samples, and the corresponding labeled masks of each type of ultrasound image sample into a preset second network model for training to obtain a second image recognition model and a second weight. Both the first and second network models include a backbone feature extraction network and a reinforcement feature extraction network. The backbone feature extraction network consists of convolutional layers and pooling layers in a VGG16 network.
[0059] It is understandable that both the first and second network models are obtained by replacing the backbone feature extraction network in the U-Net network with the convolutional and pooling layers in the VGG16 network.
[0060] The U-Net network can be divided into three parts:
[0061] The first part is the backbone feature extraction network, which is used to obtain multiple feature layers. Here, the backbone feature extraction network is replaced by the convolutional and pooling layers in the VGG16 network. The VGG16 network has a total of 16 layers, namely 13 convolutional layers, 5 pooling layers and 3 fully connected layers. This invention only uses the convolutional and pooling layers. After two convolutions of 64 convolutional layers in the first step, a pooling operation is performed. After two convolutions of 128 convolutional layers in the second step, a pooling operation is performed. After three convolutions of 256 convolutional layers in the third step, a pooling operation is performed. After repeating the convolutions of 512 convolutional layers in the third step, the preliminary effective feature layers of the network model are obtained.
[0062] The second part is to strengthen the feature extraction network. The last of the five preliminary effective feature layers obtained by the backbone feature extraction network is upsampled sequentially. The feature layer obtained by each upsampling is then fused with the corresponding preliminary effective feature layer extracted by the backbone feature extraction network to obtain a final target effective feature layer that integrates all features.
[0063] The third part is the recognition network, which is used to classify each feature point of the last target effective feature layer, which is equivalent to classifying each pixel.
[0064] This invention replaces the backbone feature extraction network of the U-Net network with convolutional and pooling layers from the VGG16 network for the following reasons:
[0065] Replacing the feature extraction network with convolutional and pooling layers from the VGG16 network enables the complete extraction of target features (lesion features) from both local and global perspectives. Because the U-Net backbone feature extraction network is trained on large medical images, each convolution, downsampling, and upsampling operation involves center cropping of the image, resulting in the loss of some image features. However, the images used in this invention are of standard size and do not require cropping. Inputting the complete image into the model allows for the extraction of both local and global features.
[0066] Replacing the feature extraction network with convolutional and pooling layers from the VGG16 network allows for more detailed feature extraction. The first and second types of vascular pathology exhibit both differences and similarities in their image features. Therefore, to fully extract features from each image, the backbone feature extraction network was replaced with convolutional and pooling layers from the VGG16 network. The feature extraction structure of the VGG16 network's convolutional and pooling layers is similar to that of the U-Net network. However, using the VGG16's convolutional and pooling layer structure for convolutional operations increases the receptive field, thus enabling more detailed extraction of image features.
[0067] Replacing the backbone feature extraction network with convolutional and pooling layers from the VGG16 network allows the use of pre-trained weights provided by the official VGG16 documentation. Since the ultrasound image features of Type I and Type II vascular pathologies are unknown, adding pre-trained weights allows for the transfer of feature characteristics from currently known features. Through transfer learning and continuous adjustment of network parameters, weights that better fit the pathological ultrasound images are derived, rather than randomly generated parameters for training. This improves the network's feature extraction performance and reduces training time.
[0068] Step 130: Based on the first image recognition model and the first weight, the ultrasound image to be tested is recognized to obtain the first recognition result.
[0069] It is understandable that the ultrasound image to be tested may include one or more of the following: Type I vascular pathology ultrasound image, Type II vascular pathology pre-compression ultrasound image, and normal vascular ultrasound image. Normal vascular ultrasound image is also known as normal ultrasound image.
[0070] The first type of vascular pathology ultrasound image can be an ultrasound image of an infantile hemangioma; the second type of vascular pathology ultrasound image before compression can be an ultrasound image of a venous malformation before compression; and the normal vascular ultrasound image can be an ultrasound image of normal blood flow.
[0071] The first type of vascular pathology ultrasound image, the second type of vascular pathology ultrasound image before compression, and the normal vascular ultrasound image are input into the first image recognition model trained above. The first image recognition model, combined with the first weight, recognizes the first type of vascular pathology ultrasound image, the second type of vascular pathology ultrasound image before compression, and the normal vascular ultrasound image to obtain the first recognition result.
[0072] Step 140: If it is determined that there are ultrasound images of the second type of vascular pathology before compression and normal vascular ultrasound images in the first identification result, based on the second image recognition model and the second weight, the ultrasound images of the second type of vascular pathology after compression and the normal vascular ultrasound images corresponding to the ultrasound images of the second type of vascular pathology before compression in the first identification result are identified to obtain the second identification result.
[0073] It is understandable that the ultrasound image after compression of the second type of vascular pathology can be an ultrasound image after compression of the venous malformation to be tested. If it is determined that there are ultrasound images before compression of the second type of vascular pathology and normal vascular ultrasound images in the first identification result, the corresponding ultrasound image after compression of the second type of vascular pathology can be found based on the ultrasound image before compression of the second type of vascular pathology in the first identification result, and then the ultrasound image after compression of the second type of vascular pathology can be identified using the second image recognition model.
[0074] The ultrasound images of the second type of vascular pathology after compression and the ultrasound images of normal blood vessels are input into the second image recognition model. The second image recognition model, combined with the second weight, identifies the ultrasound images of the second type of vascular pathology after compression and the ultrasound images of normal blood vessels, and obtains the final recognition result.
[0075] The final identification result can be a blood flow ultrasound image of the lesion area labeled with its corresponding category name, serving as a basis for clinicians to assist in diagnosis. Experimental results show that the method provided by this invention can achieve an accuracy rate of over 90% in identifying infantile hemangiomas and venous malformations.
[0076] In some embodiments, acquiring the ultrasound image sample of the target size and the corresponding annotation mask of the ultrasound image sample includes:
[0077] Acquire multiple types of blood flow ultrasound images; the multiple types of blood flow ultrasound images include: first type of vascular pathology ultrasound images, second type of vascular pathology ultrasound images, and normal ultrasound images;
[0078] The second type of vascular pathology ultrasound images were cropped into second type pre-compression ultrasound images and second type post-compression ultrasound images.
[0079] The first type of vascular pathology ultrasound image, the second type of ultrasound image before compression, the second type of ultrasound image after compression, and the normal ultrasound image are adjusted to the target size to obtain the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image before compression, the second type of vascular pathology ultrasound image after compression, and the normal vascular ultrasound image sample.
[0080] The ultrasound image samples of the first type of vascular pathology, the ultrasound image samples of the second type of vascular pathology before compression, the ultrasound image samples of the second type of vascular pathology after compression, and the ultrasound image samples of the normal blood vessels are labeled to obtain the corresponding labeling masks.
[0081] Understandably, the processing flow for various types of blood flow ultrasound images is as follows: Figure 2 As shown, after acquiring multiple types of blood flow ultrasound images, all blood flow ultrasound images are preprocessed to hide the privacy information of the personnel on the blood flow ultrasound images. The second type of vascular pathology ultrasound images, namely venous malformation ultrasound images, are cropped into the second type of pre-compression ultrasound images (i.e., venous malformation pre-compression ultrasound images) and the second type of post-compression ultrasound images (i.e., venous malformation post-compression ultrasound images).
[0082] The ultrasound images were then uniformly cropped to a size of 512x512 to obtain ultrasound image samples of the first type of vascular pathology, ultrasound image samples of the second type of vascular pathology before compression, ultrasound image samples of the second type of vascular pathology after compression, and normal vascular ultrasound image samples.
[0083] Finally, the ultrasound image samples of the first type of vascular pathology, the ultrasound image samples of the second type of vascular pathology before compression, the ultrasound image samples of the second type of vascular pathology after compression, and the ultrasound image samples of normal blood vessels were labeled to mark the lesion areas and obtain the corresponding labeling masks.
[0084] In some embodiments, the step of annotating the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal vascular ultrasound image sample to obtain a corresponding annotation mask includes:
[0085] Based on a preset graphic frame, the lesion range in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal range in the normal vascular ultrasound image sample are labeled to obtain the corresponding labeling mask.
[0086] It is understandable that the preset graphic frame can be a rectangle, a circle, an ellipse, or a polygon. The lesion area is outlined and labeled using the graphic frame to obtain the labeling mask for the ultrasound image sample of the first type of vascular pathology, the ultrasound image sample of the second type of vascular pathology before compression, and the ultrasound image sample of the second type of vascular pathology after compression. The normal vascular ultrasound image sample is outlined and labeled using the graphic frame to obtain the labeling mask.
[0087] In some embodiments, the step of annotating the lesion range in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal range in the normal vascular ultrasound image sample based on a preset graphic frame to obtain a corresponding annotation mask includes:
[0088] Based on preset graphic frames of different colors, the range of different types of lesions in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, and the second type of vascular pathology ultrasound image sample after compression, as well as the normal range in the normal vascular ultrasound image sample, are marked to obtain the corresponding marking mask.
[0089] Understandably, to obtain the entire blood flow response to the lesion, a rectangle can be used to enclose the entire lesion area containing blood flow in the IH and VM ultrasound images, that is, to enclose the entire lesion area containing some normal texture within the rectangle. IH lesions are marked with a red rectangle, VM lesions with a green rectangle, and normal areas with a blue rectangle.
[0090] In some embodiments, the step of inputting the first type of vascular pathology ultrasound image samples, the second type of vascular pathology pre-compression ultrasound image samples, and the normal vascular ultrasound image samples, along with the corresponding annotation masks for each type of ultrasound image sample, into a preset first network model for training to obtain a first image recognition model and first weights includes:
[0091] The first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, and the normal vascular ultrasound image sample are input into a preset first network model for training to obtain a first image recognition model.
[0092] Based on the backbone feature extraction network of the first network model, downsampling features are extracted from the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, and the normal vascular ultrasound image samples to obtain the preliminary effective feature layer of the first network model.
[0093] Based on the enhanced feature extraction network of the first network model, the preliminary effective feature layer of the first network model is upsampled and feature fused to obtain the target effective feature layer of the first network model;
[0094] The first weight is obtained by performing a convolution operation between the target effective feature layer of the first network model and a convolutional layer with the same number of recognition categories.
[0095] Understandably, the process for generating the first weight is as follows: Figure 3 As shown, based on the backbone feature extraction network, downsampling features are extracted from the ultrasound image samples of the first type of vascular pathology, the ultrasound image samples of the second type of vascular pathology before compression, and the ultrasound image samples of normal blood vessels. For each type of ultrasound image sample, five preliminary effective feature layers of the first network model are extracted.
[0096] For each ultrasound image sample, the preliminary effective feature layer of the 5th first network model is upsampled by a factor of 2 and then concatenated with the preliminary effective feature layer of the 4th first network model. After multiple convolutions, it is upsampled again, and the upsampled result is concatenated with the preliminary effective feature layer of the 3rd first network model. This process is repeated, with all the preliminary effective feature layers of the first network models inputting residual structures and multi-scale convolutional structures to complete feature fusion. Finally, the fused feature layers are convolved using convolutional layers with a kernel size of 1X1, which is the same as the number of recognition categories, and the weights of each category, i.e., the first weights, are output.
[0097] Specifically, based on different acquisition methods for venous malformations, the blood flow ultrasound image samples of infantile hemangiomas, blood flow ultrasound image samples before venous malformation compression, normal blood vessel ultrasound image samples, and their corresponding labeled masks are input into the first network model for training and to obtain the first weights.
[0098] In some embodiments, the step of inputting the ultrasound image samples of the second type of vascular pathological compression, the ultrasound image samples of normal blood vessels, and the corresponding annotation masks of various ultrasound image samples into a preset second network model for training to obtain a second image recognition model and a second weight includes:
[0099] The ultrasound image samples of the second type of vascular pathology compression and the ultrasound image samples of the normal vascular system are input into a preset second network model for training to obtain a second image recognition model.
[0100] Based on the backbone feature extraction network of the second network model, downsampling feature extraction is performed on the ultrasound image samples after compression of the second type of vascular pathology and the ultrasound image samples of normal vascular tissue to obtain the preliminary effective feature layer of the second network model.
[0101] Based on the enhanced feature extraction network of the second network model, the preliminary effective feature layer of the second network model is upsampled and feature fused to obtain the target effective feature layer of the second network model;
[0102] The second weights are obtained by convolving the target effective feature layer of the second network model with a convolutional layer having the same number of recognition categories.
[0103] Understandably, the flowchart for generating the second weight is as follows: Figure 3 As shown, based on the backbone feature extraction network, downsampling features are extracted from ultrasound image samples after compression of the second type of vascular pathology and ultrasound image samples of normal vascularity. For each type of ultrasound image sample, five preliminary effective feature layers of the second network model are extracted.
[0104] For each ultrasound image sample, the preliminary effective feature layer of the 5th second network model is upsampled by a factor of 2 and then concatenated with the preliminary effective feature layer of the 4th second network model. After multiple convolutions, it is upsampled again, and the upsampled result is concatenated with the preliminary effective feature layer of the 3rd second network model. This process is repeated, with all the preliminary effective feature layers of the second network models inputting residual structures and multi-scale convolutional structures to complete feature fusion. Finally, the fused feature layers are convolved using convolutional layers with a kernel size of 1X1, which is the same as the number of recognition categories, and the weights of each category, i.e., the second weights, are output.
[0105] Specifically, the ultrasound images of blood flow after venous malformation compression and normal blood flow, along with their corresponding labeled masks, are input into the second network model for training and to obtain the second weights.
[0106] After obtaining the first and second weights, the process of image recognition using the first and second weights is as follows: Figure 4 As shown.
[0107] In some embodiments, both the backbone feature extraction network and the enhanced feature extraction network include residual structures and multi-scale convolutional structures.
[0108] Understandably, after the backbone feature extraction network with residual structure and multi-scale convolution structure extracts features from the input image samples, five preliminary feature layers are obtained. The fifth preliminary feature layer is upsampled by a factor of two and concatenated with the fourth preliminary feature layer. After multiple convolutions, it is upsampled again. The upsampled result is then concatenated with the third preliminary feature layer, and so on. All preliminary feature layers are input into the complete feature fusion with residual structure and multi-scale convolution structure. Finally, the fused feature layers are convolved using a 1x1 convolutional layer with the same kernel size as the number of recognition categories, and the weights of each category are output.
[0109] In summary, the method for recognizing vascular pathological ultrasound images provided by the present invention includes:
[0110] Obtain ultrasound image samples of the target size and the corresponding annotation masks of the ultrasound image samples; the ultrasound image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples.
[0111] The first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset first network model for training to obtain a first image recognition model and a first weight. The second type of vascular pathology ultrasound image sample after compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model and a second weight.
[0112] Based on the first image recognition model and the first weight, the ultrasound image to be tested is recognized to obtain the first recognition result;
[0113] If it is determined that there are ultrasound images of the second type of vascular pathology before compression and normal vascular ultrasound images in the first identification result, based on the second image recognition model and the second weight, the ultrasound images of the second type of vascular pathology after compression and the normal vascular ultrasound images corresponding to the ultrasound images of the second type of vascular pathology before compression in the first identification result are identified to obtain the second identification result.
[0114] The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network consists of convolutional layers and pooling layers in the VGG16 network.
[0115] In the method for recognizing vascular pathology ultrasound images provided by this invention, a first type of vascular pathology ultrasound image sample, a second type of vascular pathology ultrasound image sample before compression, a normal vascular ultrasound image sample, and the corresponding labeled masks of various ultrasound image samples are input into a preset first network model for training to obtain a first image recognition model and a first weight. A second type of vascular pathology ultrasound image sample after compression, a normal vascular ultrasound image sample, and the corresponding labeled masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model and a second weight. After the first image recognition model extracts features from the second type of vascular pathology ultrasound image before compression, the output is the recognition result of the second type of vascular pathology ultrasound image before compression. The corresponding second type of vascular pathology ultrasound image after compression is found through the first recognition result of the second type of vascular pathology ultrasound image before compression, and then recognized by the second image recognition model. The accuracy of ultrasound image recognition can be improved by continuously increasing the number of ultrasound image samples for training and optimizing the weights of the image recognition model. Furthermore, the use of deep learning model training and detection enables automatic detection and recognition of target lesion features in ultrasound images, improving the efficiency of ultrasound image recognition.
[0116] like Figure 5 As shown, the present invention also provides a device 500 for recognizing vascular pathology ultrasound images, comprising:
[0117] The acquisition module 510 is used to acquire ultrasound image samples of the target size and the corresponding annotation mask of the ultrasound image samples; the ultrasound image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples.
[0118] Training module 520 is used to input the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, the normal vascular ultrasound image samples, and the corresponding annotation masks of various ultrasound image samples into a preset first network model for training to obtain a first image recognition model and a first weight; and to input the second type of vascular pathology ultrasound image samples after compression, the normal vascular ultrasound image samples, and the corresponding annotation masks of various ultrasound image samples into a preset second network model for training to obtain a second image recognition model and a second weight.
[0119] The first recognition module 530 is used to recognize the ultrasound image to be tested based on the first image recognition model and the first weight, and obtain a first recognition result.
[0120] The second identification module 540 is used to identify, based on the second image recognition model and the second weight, the ultrasound image after compression of the second type of vascular pathology and the normal vascular ultrasound image corresponding to the ultrasound image before compression of the second type of vascular pathology in the first identification result, when it is determined that there are ultrasound images before compression of the second type of vascular pathology and normal vascular ultrasound images in the first identification result, so as to obtain the second identification result.
[0121] The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network consists of convolutional layers and pooling layers in the VGG16 network.
[0122] The vascular pathology ultrasound image recognition device provided in the above embodiments can realize the technical solutions described in the above vascular pathology ultrasound image recognition method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above vascular pathology ultrasound image recognition method embodiments, which will not be repeated here.
[0123] like Figure 6 As shown, the present invention also provides an electronic device 600. The electronic device 600 includes a processor 601, a memory 602, and a display 603. Figure 6 Only some components of the electronic device 600 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.
[0124] In some embodiments, memory 602 may be an internal storage unit of electronic device 600, such as a hard disk or memory of electronic device 600. In other embodiments, memory 602 may also be an external storage device of electronic device 600, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 600.
[0125] Furthermore, the memory 602 may include both internal storage units of the electronic device 600 and external storage devices. The memory 602 is used to store application software and various types of data installed on the electronic device 600.
[0126] In some embodiments, processor 601 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 602 or process data, such as the method for recognizing vascular pathology ultrasound images in this invention.
[0127] In some embodiments, display 603 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 603 is used to display information from electronic device 600 and to display a visual user interface. Components 601-603 of electronic device 600 communicate with each other via a system bus.
[0128] In some embodiments of the present invention, when the processor 601 executes the recognition program for the vascular pathology ultrasound images in the memory 602, the following steps can be implemented:
[0129] Obtain ultrasound image samples of the target size and the corresponding annotation masks of the ultrasound image samples; the ultrasound image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples.
[0130] The first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset first network model for training to obtain a first image recognition model and a first weight. The second type of vascular pathology ultrasound image sample after compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model and a second weight.
[0131] Based on the first image recognition model and the first weight, the ultrasound image to be tested is recognized to obtain the first recognition result;
[0132] If it is determined that there are ultrasound images of the second type of vascular pathology before compression and normal vascular ultrasound images in the first identification result, based on the second image recognition model and the second weight, the ultrasound images of the second type of vascular pathology after compression and the normal vascular ultrasound images corresponding to the ultrasound images of the second type of vascular pathology before compression in the first identification result are identified to obtain the second identification result.
[0133] The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network consists of convolutional layers and pooling layers in the VGG16 network.
[0134] It should be understood that when the processor 601 executes the recognition program for vascular pathology ultrasound images in the memory 602, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.
[0135] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 600 mentioned. Electronic device 600 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 600 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).
[0136] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the XXXX method provided by the methods described above, the method comprising:
[0137] Obtain image samples of the target size and the corresponding labeled masks of the image samples; the image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples.
[0138] The first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset first network model for training to obtain a first image recognition model and a first weight. The second type of vascular pathology ultrasound image sample after compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model and a second weight.
[0139] Based on the first image recognition model and the first weight, the ultrasound image to be tested is recognized to obtain the first recognition result;
[0140] If it is determined that there are ultrasound images of the second type of vascular pathology before compression and normal vascular ultrasound images in the first identification result, based on the second image recognition model and the second weight, the ultrasound images of the second type of vascular pathology after compression and the normal vascular ultrasound images corresponding to the ultrasound images of the second type of vascular pathology before compression in the first identification result are identified to obtain the second identification result.
[0141] The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network consists of convolutional layers and pooling layers in the VGG16 network.
[0142] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0143] The above provides a detailed description of the identification of vascular pathological ultrasound images provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for recognizing vascular pathology ultrasound images, characterized in that, include: Obtain ultrasound image samples of the target size and the corresponding annotation masks of the ultrasound image samples; the ultrasound image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples. The first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset first network model for training to obtain a first image recognition model and a first weight. The second type of vascular pathology ultrasound image sample after compression, the normal vascular ultrasound image sample, and the corresponding label masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model and a second weight. Based on the first image recognition model and the first weight, the ultrasound image to be tested is recognized to obtain the first recognition result; If it is determined that there is a second type of vascular pathology compression pre-extraction ultrasound image in the first identification result, based on the second image recognition model and the second weight, the second type of vascular pathology compression post-extraction ultrasound image corresponding to the second type of vascular pathology compression pre-extraction ultrasound image in the first identification result is identified to obtain the second identification result. The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network is a convolutional layer and a pooling layer in the VGG16 network. The first type of vascular pathology ultrasound image sample is an ultrasound image sample of infantile hemangioma; the second type of vascular pathology ultrasound image sample before compression is an ultrasound image sample of venous malformation before compression; and the second type of vascular pathology ultrasound image sample after compression is an ultrasound image sample of venous malformation after compression.
2. The method for recognizing vascular pathology ultrasound images according to claim 1, characterized in that, The process of acquiring the ultrasound image sample of the target size and the corresponding annotation mask for the ultrasound image sample includes: Acquire multiple types of blood flow ultrasound images; the multiple types of blood flow ultrasound images include: first type of vascular pathology ultrasound images, second type of vascular pathology ultrasound images, and normal ultrasound images; The second type of vascular pathology ultrasound images were cropped into second type pre-compression ultrasound images and second type post-compression ultrasound images. The first type of vascular pathology ultrasound image, the second type of ultrasound image before compression, the second type of ultrasound image after compression, and the normal ultrasound image are adjusted to the target size to obtain the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image before compression, the second type of vascular pathology ultrasound image after compression, and the normal vascular ultrasound image sample. The ultrasound image samples of the first type of vascular pathology, the ultrasound image samples of the second type of vascular pathology before compression, the ultrasound image samples of the second type of vascular pathology after compression, and the ultrasound image samples of the normal blood vessels are labeled to obtain the corresponding labeling masks.
3. The method for recognizing vascular pathology ultrasound images according to claim 2, characterized in that, The step of annotating the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, the second type of vascular pathology ultrasound image samples after compression, and the normal vascular ultrasound image samples to obtain corresponding annotation masks includes: Based on a preset graphic frame, the lesion range in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal range in the normal vascular ultrasound image sample are labeled to obtain the corresponding labeling mask.
4. The method for recognizing vascular pathology ultrasound images according to claim 3, characterized in that, The method uses a preset graphical frame to annotate the lesion range in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, the second type of vascular pathology ultrasound image sample after compression, and the normal range in the normal vascular ultrasound image sample, to obtain a corresponding annotation mask, including: Based on preset graphic frames of different colors, the range of different types of lesions in the first type of vascular pathology ultrasound image sample, the second type of vascular pathology ultrasound image sample before compression, and the second type of vascular pathology ultrasound image sample after compression, as well as the normal range in the normal vascular ultrasound image sample, are marked to obtain the corresponding marking mask.
5. The method for recognizing vascular pathology ultrasound images according to claim 1, characterized in that, The step of inputting the first type of vascular pathology ultrasound image samples, the second type of vascular pathology pre-compression ultrasound image samples, and the normal vascular ultrasound image samples, along with the corresponding labeled masks for each type of ultrasound image sample, into a preset first network model for training, to obtain a first image recognition model and a first weight, includes: The first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, and the normal vascular ultrasound image samples, along with the corresponding labeled masks for each type of ultrasound image sample, are input into a preset first network model for training to obtain a first image recognition model. Based on the backbone feature extraction network of the first network model, downsampling features are extracted from the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, the normal vascular ultrasound image samples, and the labeled masks corresponding to various types of ultrasound image samples to obtain the preliminary effective feature layer of the first network model. Based on the enhanced feature extraction network of the first network model, the preliminary effective feature layer of the first network model is upsampled and feature fused to obtain the target effective feature layer of the first network model; The first weight is obtained by performing a convolution operation between the target effective feature layer of the first network model and a convolutional layer with the same number of recognition categories.
6. The method for recognizing vascular pathology ultrasound images according to claim 1, characterized in that, The process involves inputting the ultrasound image samples of the second type of vascular pathological compression, the ultrasound image samples of normal blood vessels, and the corresponding labeled masks of various ultrasound image samples into a preset second network model for training, thereby obtaining a second image recognition model and second weights, including: The ultrasound image samples of the second type of vascular pathology after compression, the ultrasound image samples of normal vascular diseases, and the corresponding labeled masks of various ultrasound image samples are input into a preset second network model for training to obtain a second image recognition model. Based on the backbone feature extraction network of the second network model, downsampling feature extraction is performed on the ultrasound image samples after compression of the second type of vascular pathology, the ultrasound image samples of normal vascular diseases, and the labeled masks corresponding to various ultrasound image samples to obtain the preliminary effective feature layer of the second network model. Based on the enhanced feature extraction network of the second network model, the preliminary effective feature layer of the second network model is upsampled and feature fused to obtain the target effective feature layer of the second network model; The second weights are obtained by convolving the target effective feature layer of the second network model with a convolutional layer having the same number of recognition categories.
7. The method for recognizing vascular pathology ultrasound images according to any one of claims 1-6, characterized in that, Both the backbone feature extraction network and the enhanced feature extraction network contain residual structures and multi-scale convolutional structures.
8. A device for recognizing ultrasound images of vascular pathology, characterized in that, include: The acquisition module is used to acquire ultrasound image samples of the target size and the corresponding annotation mask of the ultrasound image samples; the ultrasound image samples include first type of vascular pathology ultrasound image samples, second type of vascular pathology ultrasound image samples before compression, second type of vascular pathology ultrasound image samples after compression, and normal vascular ultrasound image samples. The training module is used to input the first type of vascular pathology ultrasound image samples, the second type of vascular pathology ultrasound image samples before compression, the normal vascular ultrasound image samples, and the corresponding annotation masks of various ultrasound image samples into a preset first network model for training to obtain a first image recognition model and a first weight; and to input the second type of vascular pathology ultrasound image samples after compression, the normal vascular ultrasound image samples, and the corresponding annotation masks of various ultrasound image samples into a preset second network model for training to obtain a second image recognition model and a second weight. The first recognition module is used to recognize the ultrasound image to be tested based on the first image recognition model and the first weight, and obtain a first recognition result. The second identification module is used to identify, based on the second image recognition model and the second weight, the second type of ultrasound image after compression of the second type of vascular pathology corresponding to the ultrasound image before compression of the second type of vascular pathology in the first identification result when it is determined that there is an ultrasound image before compression of the second type of vascular pathology in the first identification result, and to obtain the second identification result. The first network model and the second network model both include a backbone feature extraction network and an enhanced feature extraction network. The backbone feature extraction network is a convolutional layer and a pooling layer in the VGG16 network. The first type of vascular pathology ultrasound image sample is an ultrasound image sample of infantile hemangioma; the second type of vascular pathology ultrasound image sample before compression is an ultrasound image sample of venous malformation before compression; and the second type of vascular pathology ultrasound image sample after compression is an ultrasound image sample of venous malformation after compression.
9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the method for recognizing vascular pathology ultrasound images according to any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for recognizing vascular pathology ultrasound images as described in any one of claims 1 to 7.