Ultrasonic image ovarian tumor intelligent classification method, device, equipment and medium
By using an intelligent classification method for ovarian tumors based on ultrasound images, tumor features are extracted and merged to obtain an O-RADS score. This solves the problem of imprecise assessment of the severity of ovarian tumor lesions in existing technologies and achieves a more accurate assessment of the severity of tumor lesions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-22
- Publication Date
- 2026-03-24
AI Technical Summary
Current technologies have low accuracy in assessing the extent of ovarian tumor lesions in ultrasound images, and cannot provide precise quantification.
An intelligent classification method for ovarian tumors using ultrasound images was adopted. Tumor features from multiple ultrasound images were extracted, merged, and then the O-RADS score was obtained based on the deredundancy features to quantify the degree of tumor lesions.
It improves the accuracy of tumor lesion assessment, refines the assessment results, allows the use of more ultrasound images for assessment, and reduces computational burden.
Smart Images
Figure CN121724965A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image processing technology, specifically to a method, device, equipment, and medium for intelligent classification of ovarian tumors using ultrasound images. Background Technology
[0002] Ultrasound imaging is a medical imaging technique that uses high-frequency sound waves to acquire information about tissue structure and blood flow. Due to its advantages such as being radiation-free, easy to operate, low-cost, and capable of real-time dynamic imaging, it has wide applications in tumor diagnosis. Current technology uses image processing techniques to classify ultrasound images (including ovarian tumor ultrasound images) and determines the degree of tumor lesion based on the classification results. However, the classification results are limited to benign, malignant, and borderline cases between benign and malignant. In other words, current technology only roughly classifies tumors into three categories and cannot precisely quantify the degree of tumor lesion.
[0003] In summary, current technologies have low accuracy in assessing the degree of tumor lesions.
[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention
[0005] To address the aforementioned technical problems, this invention provides a method, device, equipment, and medium for intelligent classification of ovarian tumors using ultrasound images, which solves the problem of low accuracy in assessing the degree of tumor lesions in existing technologies.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent classification method for ovarian tumors using ultrasound images, comprising: Tumor features were extracted from multiple ultrasound images separately. Based on the similarity between multiple tumor features, the multiple tumor features are merged to obtain redundant features; Based on the deduplication features, the O-RADS score of the tumor is obtained, which is used to quantify the degree of tumor lesion.
[0007] In one implementation, multiple ultrasound images are grouped into a single ultrasound image set. The grouping methods for ultrasound images include: Acquire raw ultrasound images; The pre-selected ultrasound images are obtained by filtering out ultrasound images containing only one tumor from the original ultrasound images; Identify the tumor type on the pre-selected ultrasound images; Based on the tumor type, the pre-selected ultrasound images are grouped to obtain ultrasound image groups, and the tumor types on each ultrasound image within the same ultrasound image group are the same.
[0008] In one implementation, multiple ultrasound images are grouped into a single ultrasound image set. The grouping methods for ultrasound images include: Acquire raw ultrasound images; Segmenting individual tumor images from the original ultrasound images; Identify the tumor type in a single tumor image; Based on the tumor type, individual tumor images are grouped to obtain ultrasound image groups, and the tumor types on each individual tumor image within the same ultrasound image group are the same.
[0009] In one implementation, tumor features are extracted from multiple ultrasound images, including: For each ultrasound image, extract its local and global features respectively; The local features and the global features are linearly transformed to remove redundant features, thereby obtaining the tumor features of each ultrasound image.
[0010] In one implementation, based on the similarity between multiple tumor features, the multiple tumor features are merged to obtain redundant features, including: Multiple tumor features are clustered to merge similar features, resulting in clustered features, which are then used as deduplication features.
[0011] In one implementation, the O-RADS score of the tumor is obtained based on the deduplication features, including: The deduplication features are then fused pairwise to obtain the fused features; The O-RADS score of the tumor is obtained based on the fused features using a multilayer perceptron.
[0012] In one implementation, the ultrasound image is an ultrasound image of an ovarian tumor.
[0013] Secondly, embodiments of the present invention also provide an intelligent classification device for ovarian tumors using ultrasound images, wherein the device comprises the following components: The feature extraction module is used to extract tumor features from multiple ultrasound images separately. The redundancy removal module is used to merge multiple tumor features based on their similarity to each other to obtain redundancy-removed features; The scoring module is used to obtain the O-RADS score of the tumor based on the deduplication features, and the O-RADS score is used to quantify the degree of tumor lesion.
[0014] Thirdly, embodiments of the present invention also provide a terminal device, wherein the terminal device includes a memory, a processor, and an intelligent classification program for ovarian tumors in ultrasound images stored in the memory and executable on the processor. When the processor executes the intelligent classification program for ovarian tumors in ultrasound images, it implements the steps of the intelligent classification method for ovarian tumors in ultrasound images described above.
[0015] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing an intelligent classification program for ovarian tumors in ultrasound images. When the intelligent classification program for ovarian tumors in ultrasound images is executed by a processor, it implements the steps of the intelligent classification method for ovarian tumors in ultrasound images described above.
[0016] Beneficial Effects: This invention merges multiple tumor features based on the similarity between them in multiple ultrasound images to obtain redundant features. Then, an O-RADS score is derived based on these redundant features. Since the O-RADS score includes five values, it can more finely refine the degree of tumor lesion, thereby improving the assessment results of tumor lesion severity. Furthermore, this invention extracts tumor features from each ultrasound image separately and then obtains the final O-RADS score based on the tumor features from multiple ultrasound images. Regardless of the number of ultrasound images acquired, features of each ultrasound image can be extracted individually, and an O-RADS score can be obtained regardless of the number of tumor features. Therefore, this invention, based on O-RADS scoring, does not limit the number of acquired ultrasound images for assessing tumor lesion severity, thus allowing the use of more ultrasound images to assess tumor lesion severity and further improve the accuracy of tumor lesion severity assessment. Attached Figure Description
[0017] Figure 1 This is an overall flowchart of the present invention; Figure 2 This is a flowchart of the feature extraction process in an embodiment of the present invention; Figure 3 This is a flowchart of the feature clustering process in an embodiment of the present invention; Figure 4 A structural diagram of the intelligent classification device for ovarian tumors using ultrasound images provided by this invention; Figure 5 This is a block diagram illustrating the internal structure of a terminal device provided in an embodiment of the present invention. Detailed Implementation
[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0019] Research has shown that ultrasound imaging is a medical imaging technique that uses high-frequency sound waves to acquire information about tissue structure and blood flow. Due to its advantages such as being radiation-free, easy to operate, low-cost, and capable of real-time dynamic imaging, it has wide applications in tumor diagnosis. Current technology uses image processing techniques to classify ultrasound images and determines the degree of tumor lesion based on the classification results. However, the classification results are limited to benign, malignant, and borderline cases between benign and malignant. In other words, current technology only roughly classifies tumors into three categories and cannot precisely quantify the degree of tumor lesion.
[0020] To address the aforementioned technical problems, this invention provides a method, device, equipment, and medium for intelligent classification of ovarian tumors using ultrasound images, which solves the problem of low accuracy in assessing the degree of tumor lesions in existing technologies.
[0021] The intelligent classification method for ovarian tumors using ultrasound images in this embodiment can be applied to a terminal device, which can be a terminal product with image processing capabilities, such as a computer. In this embodiment, as... Figure 1 As shown, the intelligent classification method for ovarian tumors in ultrasound images specifically includes the following steps: S100, extracts tumor features from multiple ultrasound images respectively; S200, based on the similarity between the multiple tumor features, the multiple tumor features are merged to obtain redundant features; S300, Based on the deduplication features, the O-RADS score of the tumor is obtained, and the O-RADS score is used to quantify the degree of tumor lesion.
[0022] In step S100, multiple ultrasound images are grouped together, and each group of ultrasound images corresponds to an O-RADS score. In this embodiment, the grouping of ultrasound images includes two independent methods.
[0023] The first grouping method includes the following specific steps: acquiring original ultrasound images; selecting ultrasound images with only one tumor from the original ultrasound images to obtain pre-selected ultrasound images; identifying the tumor type on the pre-selected ultrasound images; grouping the pre-selected ultrasound images according to the tumor type to obtain ultrasound image groups, wherein the tumor types on each ultrasound image within the same ultrasound image group are the same.
[0024] The original ultrasound image is an ultrasound image acquired from the largest section of the tumor. The number of tumors in the original ultrasound image may be more than one, and the tumor type may not be limited to one type. First, ultrasound images with only one tumor are selected from the original ultrasound images as pre-selected ultrasound images. Then, the tumor types in multiple pre-selected ultrasound images are identified, and multiple pre-selected ultrasound images are grouped according to tumor type, so that tumors of the same type are grouped into the same ultrasound image group.
[0025] The reason this embodiment limits the number of tumors on each ultrasound image within a group to one and ensures that the tumor type on each ultrasound image is the same is to guarantee that the final O-RADS score applies only to the same tumor. This allows the physician to determine whether the tumor needs to be removed based on the O-RADS score. If the same ultrasound image group includes multiple different tumors, and each ultrasound image group corresponds to only one O-RADS score, it would lead to uncertainty about which tumor the O-RADS score applies to, thus making it difficult for the physician to determine which tumor to remove.
[0026] The reason why this embodiment first filters the number of tumors and then groups them based on tumor type is that the algorithm for filtering the number of tumors is relatively simple and requires very little computing power. Therefore, it can quickly filter out the pre-selected ultrasound images of individual tumors, reducing the computing power pressure of subsequent tumor type identification.
[0027] The second grouping method includes the following specific steps: acquiring the original ultrasound image; segmenting individual tumor images from the original ultrasound image; identifying the tumor type of the individual tumor image; grouping the individual tumor images according to the tumor type to obtain ultrasound image groups, wherein the tumor types on each individual tumor image within the same ultrasound image group are the same.
[0028] The method for obtaining the original ultrasound images using the second grouping method is the same as that using the first grouping method, and will not be described in detail here.
[0029] Since the acquired raw ultrasound images contain not only tumors but also other human tissues, and the presence of other human tissues can affect the subsequent extraction of tumor features, it is necessary to segment the tumor image from the raw ultrasound image in order to remove interference from other human tissues in the extraction of tumor features.
[0030] Step S100, which involves extracting tumor features from multiple ultrasound images, includes the following specific steps: for each ultrasound image, extracting local and global features; performing a linear transformation on the local and global features to remove redundant features, thereby obtaining the tumor features of each ultrasound image.
[0031] like Figure 2 As shown, each ultrasound image in the same ultrasound image group (where the number of tumors in each ultrasound image within the group is unique and the tumors are identical) is sequentially input into a CNN branch. This allows the CNN branch to output a local feature corresponding to each ultrasound image. For example, if there are ten ultrasound images in the same ultrasound image group, then inputting any one ultrasound image into the CNN branch will result in the CNN branch outputting a local feature, with each of the ten ultrasound images corresponding to one of the ten local features.
[0032] The CNN branch can be either ResNet50 or EfficientNet.
[0033] Similarly, each ultrasound image within the same ultrasound image group is input into the transformer branch sequentially, and the transformer branch outputs the global features corresponding to each ultrasound image.
[0034] Using vectors Representing local features using vectors Representing global features, vector sum vector All dimensions are 1×768. Then... and Concatenating along the second dimension yields a vector (1×1536 dimensions). This vector is then subjected to a linear transformation to obtain tumor features. The vector is then used to... Representing tumor features, vector The dimension is 1×768. The number of tumor features is the same as the number of ultrasound images; that is, the number of ultrasound images and the number of tumor features are the same.
[0035] Step S200, which involves merging multiple tumor features based on their similarity to each other to obtain redundant features, includes the following specific steps: clustering multiple tumor features to merge similar features, obtaining clustered features, and using the clustered features as redundant features.
[0036] like Figure 3 As shown, this involves K-means clustering of the tumor features corresponding to multiple ultrasound images. By clustering, similar image features are merged to remove redundant information. Figure 3 The N-instance feature learning module extracts tumor features from each of the N ultrasound images to obtain N tumor features, and then clusters the N tumor features to obtain M cluster features. In this embodiment, M is less than N.
[0037] The purpose of clustering is to remove similar redundant features. Similar redundant features represent the same tumor features, and the same tumor features have the same impact on the final score. Removing redundant features in sequence will not affect the accuracy of the final score, and can also reduce the computational pressure on subsequent tumor feature-based scoring.
[0038] Step S300, which involves obtaining the O-RADS score of the tumor based on the deduplication features, includes the following specific steps: fusing the deduplication features in pairs to obtain fused features; and obtaining the O-RADS score of the tumor based on the fused features using a multilayer perceptron.
[0039] That is, to combine the M clustering features in pairs to obtain The features of each group are input into the Mamba network so that the Mamba network fuses the two cluster features in each group to obtain the fused features corresponding to each group. Then, the fused features are flattened and input into three multilayer perceptron layers for dimensionality reduction. The O-RADS score of the tumor is obtained through the linear layer.
[0040] The O-RADS score is a standardized imaging assessment system specifically designed to assess the malignancy risk of lesions in the ovarian and fallopian tube regions of women.
[0041] When the O-RADS score is 1, it means that there are no abnormalities in the ovaries, or that the ovaries only have physiological cysts that do not affect health, and that these physiological cysts will theoretically not develop into tumors.
[0042] When the O-RADS score is 2, it means that the tumor is extremely likely to be benign, which means that the probability of the tumor being malignant is extremely low, that is, the probability of the tumor being malignant is less than 1%.
[0043] An O-RADS score of 3 indicates low risk, meaning the probability of the tumor being malignant is greater than 1% but less than 10%.
[0044] An O-RADS score of 4 indicates intermediate risk, meaning the probability of the tumor being malignant is greater than 10% but less than 50%.
[0045] An O-RADS score of 5 indicates a high risk, meaning the probability of the tumor being malignant is greater than 50%.
[0046] When the O-RADS score is 1, 2, or 3, no immediate intervention is needed for the tumor; follow-up observation is sufficient. When the O-RADS score is 4 or 5, further examination is required to rule out malignancy.
[0047] This embodiment also provides an intelligent classification device for ovarian tumors using ultrasound images, such as... Figure 4 As shown, the device comprises the following components: Feature extraction module 01 is used to extract tumor features from multiple ultrasound images respectively; The redundancy removal module 02 is used to merge multiple tumor features based on their similarity to each other to obtain redundancy-removed features; Scoring module 03 is used to obtain the O-RADS score of the tumor based on the deduplication features, and the O-RADS score is used to quantify the degree of tumor lesion.
[0048] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which can be as follows: Figure 5 As shown, the terminal device includes a processor, memory, network interface, and display screen connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an intelligent classification method for ovarian tumors from ultrasound images. The display screen can be a liquid crystal display (LCD) or an e-ink display.
[0049] Those skilled in the art will understand that Figure 5 The schematic diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.
[0050] In one embodiment, a terminal device is provided, comprising a memory, a processor, and an intelligent classification program for ovarian tumors based on ultrasound images stored in the memory and executable on the processor. When the processor executes the intelligent classification program for ovarian tumors based on ultrasound images, it implements the following operation instructions: Tumor features were extracted from multiple ultrasound images separately. Based on the similarity between multiple tumor features, the multiple tumor features are merged to obtain redundant features; Based on the deduplication features, the O-RADS score of the tumor is obtained, which is used to quantify the degree of tumor lesion.
[0051] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for intelligent classification of ovarian tumors using ultrasound images, characterized in that, include: Tumor features were extracted from multiple ultrasound images separately. Based on the similarity between multiple tumor features, the multiple tumor features are merged to obtain redundant features; Based on the deduplication features, the O-RADS score of the tumor is obtained, which is used to quantify the degree of tumor lesion.
2. The intelligent classification method for ovarian tumors using ultrasound images as described in claim 1, characterized in that, Multiple ultrasound images are grouped into a single ultrasound image. The grouping methods for ultrasound images include: Acquire raw ultrasound images; The pre-selected ultrasound images are obtained by filtering out ultrasound images containing only one tumor from the original ultrasound images; Identify the tumor type on the pre-selected ultrasound images; Based on the tumor type, the pre-selected ultrasound images are grouped to obtain ultrasound image groups, and the tumor types on each ultrasound image within the same ultrasound image group are the same.
3. The intelligent classification method for ovarian tumors using ultrasound images as described in claim 1, characterized in that, Multiple ultrasound images are grouped into a single ultrasound image. The grouping methods for ultrasound images include: Acquire raw ultrasound images; Segmenting individual tumor images from the original ultrasound images; Identify the tumor type in a single tumor image; Based on the tumor type, individual tumor images are grouped to obtain ultrasound image groups, and the tumor types on each individual tumor image within the same ultrasound image group are the same.
4. The intelligent classification method for ovarian tumors using ultrasound images as described in claim 1, characterized in that, Tumor features were extracted from multiple ultrasound images, including: For each ultrasound image, extract its local and global features respectively; The local features and the global features are linearly transformed to remove redundant features, thereby obtaining the tumor features of each ultrasound image.
5. The intelligent classification method for ovarian tumors using ultrasound images as described in claim 1, characterized in that, Based on the similarity between multiple tumor features, the multiple tumor features are merged to obtain redundant features, including: Multiple tumor features are clustered to merge similar features, resulting in clustered features, which are then used as deduplication features.
6. The intelligent classification method for ovarian tumors using ultrasound images as described in claim 1, characterized in that, Based on the aforementioned redundancy-removing features, the O-RADS score for the tumor is obtained, including: The deduplication features are then fused pairwise to obtain the fused features; The O-RADS score of the tumor is obtained based on the fused features using a multilayer perceptron.
7. The intelligent classification method for ovarian tumors using ultrasound images as described in any one of claims 1-6, characterized in that, The ultrasound image is an ultrasound image of an ovarian tumor.
8. An intelligent classification device for ovarian tumors using ultrasound images, characterized in that, The device comprises the following components: The feature extraction module is used to extract tumor features from multiple ultrasound images separately. The redundancy removal module is used to merge multiple tumor features based on their similarity to each other to obtain redundancy-removed features; The scoring module is used to obtain the O-RADS score of the tumor based on the deduplication features, and the O-RADS score is used to quantify the degree of tumor lesion.
9. A terminal device, characterized in that, The terminal device includes a memory, a processor, and an intelligent classification program for ovarian tumors in ultrasound images stored in the memory and executable on the processor. When the processor executes the intelligent classification program for ovarian tumors in ultrasound images, it implements the steps of the intelligent classification method for ovarian tumors in ultrasound images as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an intelligent classification program for ovarian tumors in ultrasound images. When the intelligent classification program for ovarian tumors in ultrasound images is executed by a processor, it implements the steps of the intelligent classification method for ovarian tumors in ultrasound images as described in any one of claims 1-7.