Thyroid nodule classification method and system

Abstract

The invention relates to a thyroid nodule classification method and system, and the method comprises the steps: firstly obtaining a plurality of thyroid nodule images, building a class label of each thyroid nodule image, carrying out the preprocessing of all thyroid nodule images, obtaining the preprocessed thyroid nodule images, building a nodule classification model, and carrying out the classification of the thyroid nodule images. The thyroid nodule image recognition method comprises the steps of preprocessing thyroid nodule images, inputting the preprocessed thyroid nodule images into a nodule classification model to obtain label reference parameters corresponding to each category label, and finally training the nodule classification model based on the label reference parameters to obtain a trained nodule classification model. And the problem of misjudgment or missed judgment is reduced.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a thyroid nodule classification method and system. Background Art

[0002] Thyroid nodules are tissue masses that appear in the thyroid tissue after abnormal proliferation of thyroid cells. They may be single or multiple, and may be solid or cystic in texture. The disease has various causes, which may be due to iodine deficiency, excessive proliferation of normal thyroid tissue, degeneration, history of radiation exposure, genetics, thyroid inflammation, etc. There is no age limit for the patients with thyroid nodules, but epidemiological statistics show that the prevalence rate in women is 3.83 times that of men, and the incidence rate is higher in people with a history of radiation exposure and a family history of thyroid nodules.

[0003] Currently, doctors generally diagnose the disease based on the patient's thyroid ultrasound images. However, due to differences in doctors' experience and heavy workload, it has become a trend to introduce artificial intelligence models to assist in the identification of thyroid nodules.

[0004] However, the existing artificial intelligence models do not preprocess the images of thyroid nodules before identifying them, resulting in low recognition accuracy after the images of thyroid nodules are input into the artificial intelligence models. This leads to poor recognition effects and is prone to misjudgment or missed judgments, which may delay the patient's normal treatment. Summary of the invention

[0005] Based on this, it is necessary to provide a thyroid nodule classification method and system to address the problem that the existing artificial intelligence model does not preprocess the image of the thyroid nodule before identifying the nodule, resulting in low recognition accuracy after the image of the thyroid nodule is input into the artificial intelligence model, making the recognition effect poor and prone to misjudgment or missed judgment, which may delay the patient's normal treatment.

[0006] The present application provides a thyroid nodule classification method, comprising:

[0007] Acquire multiple thyroid nodule images, create a category label for each thyroid nodule image, preprocess all thyroid nodule images, and obtain preprocessed thyroid nodule images;

[0008] Create a nodule classification model;

[0009] The preprocessed thyroid nodule images are input into the nodule classification model to obtain the label benchmark parameters corresponding to each category label;

[0010] The nodule classification model is trained based on the label benchmark parameters to obtain a trained nodule classification model.

[0011] On the other hand, the present application also provides a thyroid nodule classification system, the thyroid nodule classification system comprising:

[0012] Acquisition component; the acquisition component is used to acquire thyroid nodule images;

[0013] A processing component is communicatively connected with the acquisition component, and the processing component classifies the thyroid nodules according to the thyroid nodule images input by the acquisition component.

[0014] The present application relates to a thyroid nodule classification method and system. The thyroid nodule classification method first acquires multiple thyroid nodule images, creates a category label for each thyroid nodule image, preprocesses all thyroid nodule images to obtain preprocessed thyroid nodule images, creates a nodule classification model, then inputs the preprocessed thyroid nodule images into the nodule classification model to obtain label benchmark parameters corresponding to each category label, and finally trains the nodule classification model based on the label benchmark parameters to obtain a trained nodule classification model. The present application can effectively improve the recognition accuracy of thyroid nodule images and reduce the problems of misjudgment or missed judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic flow chart of a thyroid nodule classification method provided in one embodiment of the present application.

[0016] Figure 2 A schematic diagram of the structure of a thyroid nodule classification system provided in one embodiment of the present application.

[0017] Reference numerals:

[0018] 100. Collection component; 200. Processing component. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0020] The present application provides a thyroid nodule classification method and system.

[0021] like Figure 1 As shown, in one embodiment of the present application, a thyroid nodule classification method is provided, and the thyroid nodule classification method includes:

[0022] S100, acquiring multiple thyroid nodule images, creating a category label for each thyroid nodule image, and preprocessing all thyroid nodule images to obtain preprocessed thyroid nodule images.

[0023] S200, create a nodule classification model.

[0024] S300, inputting the preprocessed thyroid nodule image into a nodule classification model to obtain label benchmark parameters corresponding to each category label.

[0025] S400, training a nodule classification model based on label benchmark parameters to obtain a trained nodule classification model.

[0026] In this embodiment, first, multiple thyroid nodule images are acquired, a category label is created for each thyroid nodule image, all thyroid nodule images are preprocessed to obtain preprocessed thyroid nodule images, a nodule classification model is created, and then the preprocessed thyroid nodule images are input into the nodule classification model to obtain label reference parameters corresponding to each category label, and finally the nodule classification model is trained based on the label reference parameters to obtain a trained nodule classification model. The present application can effectively improve the recognition accuracy of thyroid nodule images and reduce the problems of misjudgment or missed judgment.

[0027] In one embodiment of the present application, the S100 includes:

[0028] S110, creating an original image database and a pre-processed image database.

[0029] S111, collecting multiple thyroid nodule images and putting them into the original image database.

[0030] S112, randomly selecting a thyroid nodule image from the original image database.

[0031] S113, obtaining the thyroid nodule category corresponding to the thyroid nodule image, and creating a category label for the thyroid nodule image.

[0032] Specifically, thyroid nodules can be divided into four category labels: cystic, solid, cystic-solid, and heterogeneous.

[0033] S114, returning to randomly select a thyroid nodule image from the original image database until the category labels of all thyroid nodule images in the original image database are created.

[0034] In this embodiment, the original image may be an image of a patient with thyroid nodules in a previous hospital. Multiple thyroid nodules are collected and the collected original images are placed in an original image database for use in subsequent model training.

[0035] In one embodiment of the present application, the S100 further includes:

[0036] S120: For each category label, count the number of thyroid nodule images included in the category label.

[0037] S121, sorting the category labels from large to small according to the number of thyroid nodule images, and selecting the category label with the largest number of thyroid nodule images as the benchmark category label.

[0038] S122, selecting a category label other than the baseline category label, and flipping and / or mirroring the thyroid nodule images included in the category label until the number of thyroid nodule images included in the category label is the same as the number of thyroid nodule images included in the baseline category label.

[0039] S123, returning to select another category label except the reference category label, until the number of thyroid nodule images included in all category labels is the same as the number of thyroid nodule images included in the reference category label.

[0040] S124, putting the expanded thyroid nodule image into the original image database.

[0041] In this embodiment, the thyroid nodule images are preprocessed so that the number of thyroid nodule images corresponding to each category remains consistent, thereby avoiding deviation when training the model.

[0042] Since the category label with the largest number of images is selected as the benchmark category label, the total number of images must increase after the preprocessing is completed, thereby expanding the training samples and making the nodule classification model finally trained more reliable.

[0043] In one embodiment of the present application, the S100 further includes:

[0044] S130, randomly selecting a thyroid nodule image from the original image database.

[0045] S131, traverse each pixel point in the thyroid nodule image, and count the pixel value of each pixel point.

[0046] S132, calculate the inter-class variance using Formula 1.

[0047] σ 2 ＝p A( t ) p B( t )( m A( t )-mB( t ))2 Formula 1.

[0048] in, σ2 is the inter-class variance, pA(t ) is the probability that a pixel in the thyroid nodule image belongs to class A, p B (t) is the probability that a pixel in the thyroid nodule image belongs to class B, m A (t) is the average gray value of class A pixels, m B (t) is the average gray value of class B pixels.

[0049] S133, determining whether the pixel value of the pixel point of the thyroid nodule image is greater than or equal to the inter-class variance.

[0050] S134: If the pixel value of the pixel point of the thyroid nodule image is greater than or equal to the inter-class variance, the pixel point is retained.

[0051] S135, collecting all the retained pixels, and the area formed by all the retained pixels is the thyroid nodule area.

[0052] S136, returning to randomly selecting a thyroid nodule image from the original image database until all thyroid nodule images in the original image database are selected, and the thyroid nodule region of each thyroid nodule image is obtained.

[0053] S137, all thyroid nodule images containing thyroid nodule regions are recorded as preprocessed images and put into a preprocessed image database.

[0054] In this embodiment, the thyroid nodule is preprocessed by using the OTSU algorithm, so as to select the thyroid nodule region in the thyroid nodule image.

[0055] The OTSU algorithm, also known as the maximum inter-class variance method (Otsu algorithm), is an algorithm for determining the threshold value. It was proposed by Japanese scholar Hideyuki Otsu in 1979. This method is often used for adaptive threshold calculation when performing binary segmentation of images. It divides the image into two parts, background and object, based on the grayscale distribution characteristics of the image. The basis for segmentation is that the inter-class variance between the two classes is the largest, that is, the difference within the class is minimized.

[0056] In one embodiment of the present application, before S132, the steps include:

[0057] S132-A, calculate the probability that the pixel in the thyroid nodule image is i by formula 2.

[0058]

[0059] Among them, n iis the total number of pixels with grayscale i in the image, and pi is the probability that a pixel in the thyroid nodule image is i.

[0060] S132-B, calculate the probability that the pixel point in the thyroid nodule image belongs to the A-type pixel point by formula 3.

[0061]

[0062] Among them, P A (t) is the probability that a pixel in the thyroid nodule image belongs to class A, pi is the probability that a pixel in the thyroid nodule image is i, and t is the segmentation threshold between class A and class B pixels.

[0063] S132-C, calculate the average pixel grayscale value of the A-type pixels using Formula 4.

[0064]

[0065] Among them, mA(t) is the average pixel gray value of class A pixels, P A (t) is the probability that a pixel in the thyroid nodule image belongs to class A, and pi is the probability that a pixel in the thyroid nodule image is pixel i.

[0066] S132-D, calculate the average pixel grayscale value of the B-type pixels using Formula 5.

[0067]

[0068] Among them, mB(t) is the average pixel gray value of class B pixels, P B (t) is the probability that a pixel in the thyroid nodule image belongs to class B, and pi is the probability that a pixel in the thyroid nodule image is i.

[0069] In this embodiment, the pixel values ​​of the thyroid nodule image are first mapped to the range of 0-255, and then the number of pixels corresponding to each grayscale value is counted, and the probability that the pixel point in the thyroid nodule image is i is calculated. Then, it is assumed that the threshold is known, and the average pixel grayscale values ​​of class A and class B pixels are calculated respectively by Formula 4 and Formula 5.

[0070] In one embodiment of the present application, the S300 includes:

[0071] S310, creating a nodule classification model. The nodule classification model includes multiple nodule regions, each of which corresponds to a category label.

[0072] S320, selecting a category label, and selecting all pre-processed images under the category label from the pre-processed image data.

[0073] S330, inputting the selected preprocessed image into the nodule classification model to train the nodule classification model.

[0074] S340, return to S320, until all category labels are selected, and a trained nodule classification model is obtained. The trained nodule classification model has the ability to determine the nodule category according to the input thyroid nodule image.

[0075] In this embodiment, after the preprocessing of the original image is completed, the preprocessed image is input into the nodule classification model as a training sample to train the nodule classification model, thereby obtaining a trained nodule classification model, so that the trained nodule classification model has the ability to judge the nodule category based on the input thyroid nodule image.

[0076] In one embodiment of the present application, the S300 further includes:

[0077] S350, selecting a pre-processed image.

[0078] S360, input the preprocessed image into a nodule classification model.

[0079] S370, obtaining label parameters output by the nodule classification model.

[0080] S380, return to S350, until all pre-processed images under the category label are selected, and the label parameters corresponding to each pre-processed image under the category label are obtained.

[0081] S390, selecting a central parameter from all label parameters, and using the central parameter as a label reference parameter for labels of this category.

[0082] In this embodiment, thyroid nodules are classified by label parameters.

[0083] In one embodiment of the present application, the S390 includes:

[0084] S391, set the initial radius.

[0085] S392, randomly selecting a label parameter from all label parameters as the initial parameter center.

[0086] S393, calculate the average value of the distances between all other label parameters whose parameter centers are within the initial radius using Formula 2, and record it as a migration vector.

[0087]

[0088] in, is the migration vector, is the initial parameter center, Sc is is the cluster center, the initial radius is the cluster formed by the radius, and k is the number of points contained in Sc.

[0089] S394, combining the migration vector to obtain a new parameter center.

[0090] S395, return to S393, until the parameter center no longer moves, output the final parameter center, and record the parameter center as the label reference parameter of the category label.

[0091] In this embodiment, MeanShift is used to obtain label benchmark parameters, thereby improving the reliability of the label benchmark parameters.

[0092] The reason why the MeanShift mean migration algorithm is used is that the MeanShift mean migration algorithm does not need to specify the number of clusters, and can automatically discover the number of clusters, thereby making the final classification structure more stable.

[0093] In one embodiment of the present application, the S300 further includes:

[0094] S396, select a category label.

[0095] S397, obtaining all label parameters of the category label.

[0096] S398, filtering out the central parameter of the category label from all label parameters under the category label, and using the central parameter as the label reference parameter of the category label.

[0097] S399, return to S396, until all category labels are selected, and obtain the label benchmark parameters of each category label.

[0098] In this embodiment, for each category label, it is necessary to find out the label reference parameter corresponding to the category label, so as to ensure that the nodule classification model will not miss any judgment.

[0099] like Figure 2 As shown, in one embodiment of the present application, a thyroid nodule classification system is provided, and the thyroid nodule classification system includes a collection component 100 and a processing component 200 .

[0100] The acquisition component 100 is used to acquire thyroid nodule images. The processing component 200 is in communication connection with the acquisition component 100 , and the processing component 200 classifies thyroid nodules according to the thyroid nodule images input by the acquisition component 100 .

[0101] In this embodiment, the thyroid nodule images are collected by the acquisition component 100, and the collected thyroid nodule images are input into the processing component 200. After receiving the images, the processing component 200 processes all the original images and performs the thyroid nodule classification method described in the above-mentioned embodiment on the original images.

[0102] The technical features of the above-described embodiments may be arbitrarily combined, and the execution order of the method steps is not limited. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0103] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for classifying thyroid nodules, characterized in that: The thyroid nodule classification method comprises: Acquire multiple thyroid nodule images, create a category label for each thyroid nodule image, preprocess all thyroid nodule images, and obtain preprocessed thyroid nodule images; Create a nodule classification model; The preprocessed thyroid nodule images are input into the nodule classification model to obtain the label benchmark parameters corresponding to each category label; The nodule classification model is trained based on the label benchmark parameters to obtain a trained nodule classification model.

2. The thyroid nodule classification method according to claim 1, characterized in that: The step of acquiring multiple thyroid nodule images and creating a category label for each thyroid nodule image includes: Create original image database and pre-processed image database; Collect multiple thyroid nodule images and put them into the original image database; Randomly select a thyroid nodule image from the original image database; Obtain the thyroid nodule category corresponding to the thyroid nodule image, and create a category label for the thyroid nodule image; return to arbitrarily select a thyroid nodule image from the original image database until the category labels of all thyroid nodule images in the original image database are created.

3. The method for classifying thyroid nodules according to claim 2, characterized in that: The preprocessing of all thyroid nodule images to obtain preprocessed thyroid nodule images includes: For each category label, count the number of thyroid nodule images contained in the category label; Sort the category labels from large to small according to the number of thyroid nodule images, and select the category label with the largest number of thyroid nodule images as the benchmark category label; Select a category label other than the benchmark category label, and perform flipping and / or mirroring processing on the thyroid nodule images included in the category label until the number of thyroid nodule images included in the category label is the same as the number of thyroid nodule images included in the benchmark category label; Return and select another category label except the benchmark category label until the number of thyroid nodule images contained in all category labels is the same as the number of thyroid nodule images contained in the benchmark category label; The expanded thyroid nodule images are placed into the original image database.

4. The method for classifying thyroid nodules according to claim 3, characterized in that: The preprocessing of all thyroid nodule images to obtain preprocessed thyroid nodule images also includes: Randomly select a thyroid nodule image from the original image database; Traverse each pixel point in the thyroid nodule image and count the pixel value of each pixel point; The inter-class variance is calculated by formula 1; σ 2 = p A (t)p B (t)(m A (t) - m B (t)) 2 Equation 1; Among them, σ 2 is the between-class variance, p A (t) is the probability that a pixel in the thyroid nodule image belongs to class A, p B (t) is the probability that a pixel in the thyroid nodule image belongs to class B, m A (t) is the average gray value of class A pixels, m B (t) is the average gray value of class B pixels; Determine whether the pixel value of the pixel point of the thyroid nodule image is greater than or equal to the inter-class variance; If the pixel value of the pixel point of the thyroid nodule image is greater than or equal to the inter-class variance, the pixel point is retained; Collect all the retained pixels, and the area formed by all the retained pixels is the thyroid nodule area; Return to randomly select a thyroid nodule image from the original image database until all thyroid nodule images in the original image database are selected, and obtain the thyroid nodule area of ​​each thyroid nodule image; All thyroid nodule images containing thyroid nodule regions are recorded as preprocessed images and put into a preprocessed image database.

5. The method for classifying thyroid nodules according to claim 4, characterized in that: Before calculating the inter-class variance by formula 1, the thyroid nodule classification method further includes: The probability that a pixel in the thyroid nodule image is i is calculated by formula 2; Among them, n i is the total number of pixels with gray level i in the image, p i is the probability that the pixel in the thyroid nodule image is i; The probability that a pixel in the thyroid nodule image belongs to Class A is calculated by using Formula 3; Among them, P A (t) is the probability that a pixel in the thyroid nodule image belongs to class A, p i is the probability that the pixel in the thyroid nodule image is i, and t is the pixel segmentation threshold between class A pixels and class B pixels; The average pixel grayscale value of class A pixels is calculated by formula 4; Among them, m A (t) is the average pixel grayscale value of class A pixels, P A (t) is the probability that a pixel in the thyroid nodule image belongs to class A, p i is the probability that the pixel in the thyroid nodule image is i; The average pixel grayscale value of class B pixels is calculated by formula 5; Among them, m B (t) is the average pixel gray value of class B pixels, P B (t) is the probability that a pixel in the thyroid nodule image belongs to class B, p i is the probability that the pixel in the thyroid nodule image is i.

6. The method for classifying thyroid nodules according to claim 5, characterized in that: The preprocessed thyroid nodule image is input into the nodule classification model to obtain the label reference parameters corresponding to each category label, including: Creating a nodule classification model; the nodule classification model includes multiple nodule regions, each nodule region corresponds to a category label; Select a category label, and select all preprocessed images under the category label from the preprocessed image data; Inputting the selected preprocessed image into the nodule classification model to train the nodule classification model; Return and select a category label until all category labels are selected to obtain a trained nodule classification model; the trained nodule classification model has the ability to determine the nodule category based on the input thyroid nodule image.

7. The method for classifying thyroid nodules according to claim 6, characterized in that: The step of inputting the preprocessed thyroid nodule image into the nodule classification model to obtain the label reference parameters corresponding to each category label also includes: Select a preprocessed image; The preprocessed image is input into a nodule classification model; Get the label parameters output by the nodule classification model; Return to select a preprocessed image until all preprocessed images under the category label are selected, and obtain the label parameters corresponding to each preprocessed image under the category label; The central parameter is selected from all label parameters and used as the label benchmark parameter for this category of labels.

8. The method for classifying thyroid nodules according to claim 7, characterized in that: The method of selecting a central parameter from all label parameters and using the central parameter as the label reference parameter of this category label includes: Set the initial radius; Randomly select a label parameter from all label parameters as the initial parameter center; The average value of the distances between all other label parameters within the initial radius of the parameter center is calculated by formula 6 and recorded as the migration vector; in, is the migration vector, is the initial parameter center, S c So is the cluster center, the initial radius is the cluster formed by the radius, and k is S c The number of points contained in is the i-th parameter center; Combine the migration vector to get the new parameter center; Return the average value of the distances between the parameter center and all other label parameters within the initial radius calculated by formula 6 until the parameter center stops moving. Output the final parameter center and record the parameter center as the label benchmark parameter of the category label.

9. The method for classifying thyroid nodules according to claim 8, characterized in that: The step of inputting the preprocessed thyroid nodule image into the nodule classification model to obtain the label reference parameters corresponding to each category label also includes: Select a category label; Get all label parameters of this category label; Filter out the central parameter of the category label from all the label parameters under the category label, and use the central parameter as the label benchmark parameter of this category label; Return to select a category label until all category labels are selected, and obtain the label benchmark parameters of each category label.

10. A thyroid nodule classification system, applied to the thyroid nodule classification method according to any one of claims 1 to 9, characterized in that: The thyroid nodule classification system includes: An acquisition component, wherein the acquisition component is used to acquire thyroid nodule images; A processing component is communicatively connected with the acquisition component, and the processing component classifies the thyroid nodules according to the thyroid nodule images input by the acquisition component.

Images