Method and device for automatically detecting focus and computer storage medium
By clustering and screening the prediction labels and observation labels of ultrasound images, the lesion detection model is optimized, and the problem of inaccurate lesion detection caused by label noise is solved, and the accuracy of diagnostic results is improved.
Patent Information
- Application Number
- CN202411933992.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-04-25
AI Technical Summary
In the existing automatic detection methods of lesions, label noise will lead to inaccurate model identification and diagnosis of lesions, affecting the patient's diagnosis results.
By obtaining the prediction labels and observation labels of ultrasound images, clustering processing is performed, the noise data subset and the denoising data subset are filtered out, the noise data subset is proofreaded, and the lesion detection model is optimized.
It improves the accuracy of the lesion detection model, enhances the accuracy of the diagnostic results, and significantly improves the accuracy of screening noise labels.
Smart Images
Figure CN119887662A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of medical diagnosis, and in particular to a method and device for automatically detecting lesions, and a computer storage medium. Background Art
[0002] With the development of modern science and technology, artificial intelligence technology has been widely used in medical diagnosis of organisms. Doctors input ultrasound images into the detection model to quickly detect lesions and segment and measure the lesions to obtain lesion information, thereby diagnosing patients, greatly improving diagnostic efficiency.
[0003] The above model needs a large amount of ultrasound image data for training before use. The ultrasound image data is the data with lesion labels. The lesion labels include both the lesion frame used to frame the lesion and the lesion category. However, there may be label noise in the ultrasound image labels, such as missing labels, suspicious labels, mispositioned labels, and misclassified labels. Missing labels means that there are no labels in the labels where there should be labels, suspicious labels means that labels are given where there should not be labels, mispositioned labels means that there is a large error between the given label position and the actual position, and misclassified labels means that the given label category is wrong. Label noise will cause the model to identify and diagnose lesions inaccurately, affecting the patient's diagnosis results. Summary of the invention
[0004] The purpose of this application is to overcome the defect in existing automatic lesion detection methods that label noise can lead to inaccurate model identification and diagnosis of lesions, affecting the patient's diagnosis results.
[0005] To this end, the present application provides a method for automatically detecting lesions, comprising:
[0006] Obtain predicted labels and observed labels for ultrasound images;
[0007] Clustering of predicted labels and observed labels;
[0008] Screening the label samples to obtain a first data subset and a second data subset, the first data subset is a noise data subset, and the second data subset is a denoised data subset;
[0009] Review and proofread the first data subset;
[0010] Optimizing the lesion detection model based on the second data subset and the reviewed and proofread first data subset;
[0011] The ultrasound image to be diagnosed is input into the optimized lesion detection model to automatically detect the lesion.
[0012] Furthermore, clustering the predicted labels and the observed labels includes:
[0013] Perform hierarchical clustering on predicted and observed labels;
[0014] Transform the clustering results using a subtractive algorithm.
[0015] Furthermore, the screening of the label samples to obtain the first data subset and the second data subset includes:
[0016] The confident learning method is used to calculate the quality scores of the labels converted by the subtraction algorithm;
[0017] Corresponding noise samples are screened out through a threshold to obtain a first data subset and a second data subset.
[0018] Furthermore, the subtraction algorithm converts the clustering result into two matrices by converting the hierarchical clustering result.
[0019] Furthermore, the hierarchical clustering is to classify the predicted labels and the observed labels using 1-IOU(a,b) as a distance metric.
[0020] Furthermore, the lesion detection model automatically detects the lesion and frames the lesion.
[0021] Furthermore, the method for automatically detecting lesions also includes the step of inputting the output result of the lesion detection model into a segmentation model, wherein the segmentation model is used to segment the lesion and identify the lesion information.
[0022] The present application also provides a device for automatically detecting lesions, comprising:
[0023] A label acquisition module, which is used to obtain predicted labels and observed labels of ultrasound images;
[0024] The label clustering module is used to cluster the predicted labels and observed labels;
[0025] A screening and proofreading module, which is used to screen the label samples to obtain the first data subset and the second data subset and to review and proofread the first data subset;
[0026] A model optimization module, which is used to optimize the lesion detection model based on the second data subset and the first data subset after review and proofreading;
[0027] The lesion detection module is used to input the ultrasound image to be diagnosed into the optimized lesion detection model to automatically detect the lesion.
[0028] The present application also provides a device for automatically detecting lesions, including a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the aforementioned method for automatically detecting lesions is implemented.
[0029] The present application also provides a computer storage medium, on which a program is stored. When the program is executed by a processor, it is used to implement the aforementioned method for automatically detecting lesions.
[0030] The technical solution of this application has at least the following advantages:
[0031] 1. The method for automatic lesion detection of the present application can accurately screen out the first data subset and the second data subset of the ultrasound image by acquiring the predicted labels and observation labels of the ultrasound image and clustering the labels, and then optimize the lesion detection model by calibrating the first data subset, thereby improving the accuracy of the lesion detection model in detecting lesions and improving the accuracy of the diagnosis results.
[0032] 2. This application uses hierarchical clustering, subtraction algorithm and confidence learning algorithm to screen the label noise of ultrasound images, which greatly improves the accuracy of screening noise labels of ultrasound images. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0034] Figure 1 is a flow chart of a method for automatically detecting lesions in Example 1 of the present application;
[0035] Figure 2 This is a structural block diagram of a device for automatically detecting lesions in Try 2 of this application;
[0036] Figure 3 This is a structural block diagram of a device for automatically detecting lesions in Example 3 of the present application. DETAILED DESCRIPTION
[0037] The following will be combined with the accompanying drawings to clearly and completely describe the technical solutions in this application. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0038] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.
[0039] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, it can also be the internal connection of two components, it can be a wireless connection, or it can be a wired connection. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0040] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.
[0041] Example 1
[0042] Embodiment 1 provides a method for automatically detecting lesions. Figure 1 For a flowchart of automatic lesion detection, refer to Figure 1 , the methods for automatically detecting lesions include:
[0043] S1, obtain the predicted label and observed label of the ultrasound image;
[0044] Specifically, the observed label is a manually annotated label, that is, the label obtained by manually annotating the ultrasound image. The observed label of the ultrasound image can be obtained by experts annotating the lesions in the ultrasound image. The predicted label is the label output after the ultrasound image is input into the trained detection model.
[0045] When training the detection model, the annotated ultrasound images are divided into multiple groups, and several groups are selected to train the detection model. The ultrasound images that are not involved in the training are input into the trained detection model as the test group to obtain the predicted label of the test group. The training method is used in turn to use any several groups of multiple groups of ultrasound images as training groups, and the remaining groups as test groups to obtain the predicted labels of each group of annotated ultrasound images in turn. For example, there are 5 groups of annotated ultrasound images, and 4 groups are selected as training groups to train the detection model, and the remaining group is used as the test group to obtain its predicted label, and then 4 groups of ultrasound images different from the previous group are selected in turn as training groups, and the remaining group is used as the test group to obtain the predicted label of the test group until the predicted label of each group of ultrasound images is obtained. According to the above method, the observation label and predicted label of the ultrasound image are obtained, and the detection model can be any detection model that can identify the label of the ultrasound image.
[0046] S2, clustering the predicted labels and observed labels;
[0047] Specifically, clustering is performed on the predicted labels and observed labels, including:
[0048] Perform hierarchical clustering on the predicted labels and observed labels; hierarchical clustering aims to reduce the number of clusters. In essence, it uses 1-IOU(a,b) as the distance metric to classify the predicted labels and observed labels, where IOU(a,b) is the intersection-over-union ratio of the observed label and the predicted label, a is the coordinate point of the lesion box of the observed label in the ultrasound image, and b is the coordinate point of the lesion box of the predicted label in the ultrasound image; specifically, the output result of hierarchical clustering is represents the observation label in the kth cluster, is the predicted label in the kth cluster.
[0049] The clustering results are transformed using the subtractive algorithm. The subtractive algorithm transforms the clustering results into two matrices. Specifically, the hierarchical clustering results are transformed into As input, the output Y is passed through the subtraction algorithm k ′ and P k ′ , where Y k ′ It is a binary matrix, whose value is either 0 or 1. Specifically, it determines whether the predicted label is the observed label of a certain category. If so, it is 1, and if not, it is 0; P k ′ is the probability of predicting the label, P k ′ ,m To predict the maximum probability of a certain class prediction label.
[0050] S3, screening the label samples to obtain a first data subset and a second data subset, the first data subset is a noise data subset, the second data subset is a denoised data subset, the noise data subset is a set of ultrasound images with label noise, and the denoised data subset is a set of ultrasound images without label noise;
[0051] Specifically, screening the label samples to obtain the first data subset and the second data subset includes:
[0052] The confidence learning algorithm is used to calculate the quality score of the observation label; specifically, the output result Y k ′ and P k ′ Input the confidence learning algorithm to calculate the quality score of the observation label;
[0053] The corresponding noise samples are screened out through a threshold to obtain the first data subset and the second data subset. The threshold is an empirical value. The observation labels with low quality scores are filtered according to the threshold, and then the observation labels with low quality scores are divided into four categories through a classifier, namely, missing labels, suspicious labels, mislocated labels and misclassified labels. Missing labels mean that there are no labels where there should be labels, suspicious labels mean that labels are given where there should not be labels, mislocated labels mean that there is a large error between the given label position and the true position, and misclassified labels mean that the given label category is incorrect. The set of ultrasound images with missing labels, suspicious labels, mislocated labels and misclassified labels is the first data subset, and the set of ultrasound images with remaining high-quality labels is the second data subset.
[0054] The present application screens the label noise of ultrasound images by cooperating with hierarchical clustering, subtraction algorithm and confidence learning algorithm, thereby greatly improving the accuracy of screening the noise labels of ultrasound images.
[0055] S4, reviewing and proofreading the first data subset;
[0056] Specifically, the first data subset is manually proofread by an expert, that is, the expert modifies the labels of the ultrasound images in the first data subset, or re-labels the labels of the ultrasound images in the first data subset through a deep learning algorithm; specifically, the operations of the expert manually reviewing and proofreading the first data subset are: modifying labels with positioning errors, modifying misclassified labels, adjusting suspicious labels, and re-labeling missing labels;
[0057] S5, optimizing the lesion detection model based on the second data subset and the first data subset after review and proofreading;
[0058] S6. Input the ultrasound image to be diagnosed into the optimized lesion detection model to automatically detect the lesion.
[0059] Specifically, the lesion detection model automatically detects the lesion and outputs the lesion frame, and then the output result of the lesion detection model is input into the segmentation model to segment the lesion to obtain the contour information of the lesion and identify the lesion information according to the contour information. Further, the image output by the lesion detection model is cropped first, and then the cropped ultrasound image is input into the segmentation model to segment the lesion. The identified lesion information may be the long and short diameters of the identified lesion.
[0060] The above-mentioned method for automatic lesion detection can accurately screen out the first data subset and the second data subset of the ultrasound image by obtaining the predicted labels and observation labels of the ultrasound image and clustering the labels, and then optimize the lesion detection model by calibrating the first data subset, thereby improving the accuracy of the lesion detection model in detecting lesions and improving the accuracy of the diagnosis results.
[0061] Example 2
[0062] Embodiment 2 provides a device for automatically detecting lesions. Figure 2 For the structural block diagram of the device for automatically detecting lesions, refer to Figure 2 The device for automatically detecting lesions includes a label acquisition module 201, a label clustering module 202, a screening and proofreading module 203, a model optimization module 204, and a lesion detection module 205. The label acquisition module 201 is used to acquire the predicted label and the observed label of the ultrasound image; the label clustering module 202 is used to perform clustering processing on the predicted label and the observed label; the screening and proofreading module 203 is used to screen the label samples to obtain the first data subset and the second data subset and to review and proofread the first data subset; the model optimization module 204 is used to optimize the lesion detection model based on the second data subset and the first data subset after review and proofreading; the lesion detection module 205 is used to input the ultrasound image to be diagnosed into the optimized lesion detection model to automatically detect the lesion. The present application cooperates with the label acquisition module 201, the label clustering module 202, and the screening and proofreading module 203 to accurately screen out the first data subset and the second data subset and proofread the first data subset, and further optimizes the lesion detection model through the model optimization module 204, thereby improving the accuracy of lesion detection by the lesion detection module 205, thereby improving the accuracy of subsequent lesion segmentation and measurement of lesion information, and improving the accuracy of biological disease diagnosis based on ultrasound images.
[0063] Example 3
[0064] refer to Figure 3 The present application also provides a vehicle-mounted control device, which includes a processor 310 and a memory 320.
[0065] The processor 310 may include one or more processing cores. The processor 310 uses various interfaces and lines to connect various parts of the entire device, and executes various functions of the device and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 320, and calling data stored in the memory 320. Optionally, the processor 310 can be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 310 can integrate one or a combination of a central processing unit (CPU) and a modem. Among them, the CPU mainly processes the operating system and application programs; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 310, but implemented separately through a chip.
[0066] Optionally, the processor 310 implements the method for automatically detecting lesions provided in the above-mentioned embodiment 1 when executing the program instructions in the memory 320 .
[0067] The memory 320 may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory 320 includes a non-transitory computer-readable storage medium. The memory 320 may be used to store instructions, programs, codes, code sets, or instruction sets. The memory 320 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the above-mentioned embodiment 1, etc.; the data storage area may store data created according to the use of the device, etc.
[0068] Optionally, the present application also provides a computer storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the method for automatically detecting lesions provided in the above-mentioned embodiment 1.
[0069] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection created by this application.
Claims
1. A method for automatically detecting lesions, characterized in that: include: Obtain predicted labels and observed labels for ultrasound images; Clustering of predicted labels and observed labels; Screening the label samples to obtain a first data subset and a second data subset, wherein the first data subset is a noise data subset and the second data subset is a denoised data subset; Review and proofread the first data subset; Optimizing the lesion detection model based on the second data subset and the reviewed and proofread first data subset; The ultrasound image to be diagnosed is input into the optimized lesion detection model to automatically detect the lesion.
2. The method for automatically detecting lesions according to claim 1, characterized in that: The clustering process of the predicted labels and the observed labels includes: Perform hierarchical clustering on predicted and observed labels; Transform the clustering results using a subtractive algorithm.
3. The method for automatically detecting lesions according to claim 2, characterized in that: The screening of the label samples to obtain a first data subset and a second data subset includes: The confidence learning algorithm is used to calculate the quality score of the observation label; Corresponding noise samples are screened out through a threshold to obtain a first data subset and a second data subset.
4. The method for automatically detecting lesions according to claim 3, characterized in that: The subtraction algorithm converts the clustering results into two matrices.
5. The method for automatically detecting lesions according to claim 2, characterized in that: The hierarchical clustering is to classify the predicted labels and observed labels using 1-IOU(a,b) as the distance metric.
6. The method for automatically detecting lesions according to claim 1, characterized in that: The lesion detection model automatically detects lesions and frames them.
7. The method for automatically detecting lesions according to claim 6, characterized in that: The method also includes a step of inputting the output result of the lesion detection model into a segmentation model, wherein the segmentation model is used to segment the lesion and identify the lesion information.
8. A device for automatically detecting lesions, characterized in that: include: A label acquisition module, which is used to obtain predicted labels and observed labels of ultrasound images; The label clustering module is used to cluster the predicted labels and observed labels; A screening and proofreading module, which is used to screen the label samples to obtain the first data subset and the second data subset and to review and proofread the first data subset; A model optimization module, which is used to optimize the lesion detection model based on the second data subset and the first data subset after review and proofreading; The lesion detection module is used to input the ultrasound image to be diagnosed into the optimized lesion detection model to automatically detect the lesion.
9. A device for automatically detecting lesions, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the method for automatically detecting lesions according to any one of claims 1 to 7 is implemented.
10. A computer storage medium having a program stored thereon, wherein when the program is executed by a processor, the program is used to implement the method for automatically detecting lesions according to any one of claims 1 to 7.