Automatic identification method and system for thyroid nodule ultrasonic image

Abstract

The invention provides an automatic identification method and system for a thyroid nodule ultrasonic image, and the method comprises the steps: obtaining a to-be-identified thyroid nodule ultrasonic image, and converting the to-be-identified thyroid nodule ultrasonic image into a nodule gray image; segmenting the nodule grayscale image into a plurality of sub-image blocks, determining a plurality of calcification regions of thyroid nodules in the sub-image blocks, and determining local calcification features of the sub-image blocks based on convolution kernels of the sub-image blocks in different scale channels and all the calcification regions; determining the recognition depth when thyroid nodule recognition is carried out on the sub-image blocks; determining a significance coefficient based on the recognition depth and the aspect ratio of the thyroid nodules in the sub-image blocks; and performing feature cascading on all the local calcification features based on the significance coefficient to obtain global calcification features of the thyroid nodule ultrasound image to be identified. By adopting the scheme of the application, the calcification part can be subjected to multi-dimensional identification under the interference of various factors, so that accurate description of thyroid calcification features is realized.

Description

Technical Field

[0001] This application relates to the field of image processing technology. More specifically, this application relates to an automatic recognition method and system for thyroid nodule ultrasound images. Background Art

[0002] With the rapid development of medical imaging technologies, such as X-ray detection, Computed Tomography (CT), ultrasound detection, etc., which are widely used, a vast amount of medical image data has been generated. These original medical images often have problems such as blurred imaging and unclear features. In order to improve the quality of medical images for better auxiliary diagnosis, a series of processing technologies have emerged. Among them, image filtering technology is used to enhance image clarity, image segmentation technology can accurately divide the tissue or lesion area of interest, and feature extraction technology can obtain features such as texture, shape, and gray level to help identify different pathological features. The application of medical image processing technology can effectively identify image features and pathological patterns, thereby improving the accuracy and efficiency of diagnosis and promoting the intelligent analysis of medical images.

[0003] In the prior art, the ultrasound image processing technology under medical image processing is a technical field that analyzes digital medical images obtained by medical imaging equipment. Among them, when performing the ultrasound image recognition of thyroid nodules, key technologies such as image preprocessing, lesion localization and segmentation, feature extraction, and classification are involved. Through the above key technologies, the pathological features of the image are automatically extracted and classified to assist doctors in discovering thyroid lesions and improving the diagnostic accuracy, showing great potential. However, in the actual recognition process, the features of thyroid nodules are affected by various factors, such as nodule size, shape, boundary, etc. Among them, the calcification situation of thyroid nodules shows various forms. Microcalcifications may only be tiny points in the ultrasound image, which are difficult to distinguish from image noise. Moreover, for the extraction of calcification features, both the situation of calcification itself and the surrounding tissue information need to be considered, increasing the difficulty of recognition. Therefore, how to perform multi-dimensional recognition of the calcified part under the interference of various factors to accurately describe the thyroid calcification features has become a difficult problem faced by the industry. Summary of the Invention

[0004] This application provides an automatic recognition method and system for thyroid nodule ultrasound images, which can perform multi-dimensional recognition of the calcified part under the interference of various factors, so as to accurately describe the thyroid calcification features.

[0005] In a first aspect, this application provides an automatic recognition method for thyroid nodule ultrasound images, including the following steps: Obtain the thyroid nodule ultrasound image to be recognized, and convert the thyroid nodule ultrasound image to be recognized into a nodule grayscale image; Segment the nodule grayscale image into multiple sub-image blocks, select one sub-image block as the selected sub-image block, determine multiple calcified regions of the thyroid nodule in the selected sub-image block, and determine the local calcification features of the selected sub-image block based on the convolution kernels of the selected sub-image block in different scale channels and all the calcified regions; Determine the recognition depth when recognizing the thyroid nodule for the selected sub-image block; Based on the recognition depth and the aspect ratio of the thyroid nodule in the selected sub-image block, determine the significance coefficient of the selected sub-image block, and continue to determine the local calcification features and significance coefficients of the remaining sub-image blocks; Based on each significance coefficient, perform feature concatenation on all the local calcification features to obtain the global calcification features corresponding to the thyroid nodule ultrasound image to be recognized.

[0006] In some embodiments, segmenting the nodule grayscale image into multiple sub-image blocks specifically includes: Perform normalization processing on the nodule grayscale image to obtain a standard grayscale image; Cut the standard grayscale image to obtain multiple image partitions; Extract multiple sub-image blocks from all the image partitions.

[0007] In some embodiments, determining multiple calcified regions of the thyroid nodule in the selected sub-image block specifically includes: Extract multiple calcification candidate points of the thyroid nodule from the selected sub-image block; Based on all the calcification candidate points, determine multiple suspicious calcified regions of the selected sub-image block; Extract multiple calcified regions of the thyroid nodule from all the suspicious calcified regions.

[0008] In some embodiments, determining the local calcification features of the selected sub-image block based on the convolution kernels of the selected sub-image block in different scale channels and all the calcified regions specifically includes: Set different scale channels and initialize multiple convolution kernels under each scale channel; Perform convolution operations on the selected sub-image block through all the convolution kernels to obtain multiple convolution kernel response maps; For each scale channel, perform feature extraction on all the calcified regions based on the convolution kernel response maps under the scale channel to obtain the feature vectors corresponding to the scale channel, and further obtain the feature vectors corresponding to each scale channel; Fuse all the feature vectors to obtain the multi-dimensional calcification features of the selected sub-image block; Perform feature optimization on the multi-dimensional calcification features to obtain the local calcification features of the selected sub-image block.

[0009] In some embodiments, feature extraction is performed on all calcified regions based on the convolutional kernel response maps under the scale channels to obtain the feature vectors corresponding to the scale channels, which specifically includes: Select a convolutional kernel response map as the selected convolutional kernel response map, and map each calcified region to the selected convolutional kernel response map to obtain multiple mapped regions; Determine the calcification statistical features corresponding to the selected convolutional kernel response map according to all the mapped regions; Continue to determine the corresponding calcification statistical features of the remaining convolutional kernel response maps; Combine all the calcification statistical features to obtain the feature vectors corresponding to the scale channels.

[0010] In some embodiments, feature concatenation is performed on all local calcification features based on each significance coefficient to obtain the global calcification features corresponding to the ultrasound image of the thyroid nodule to be recognized, which specifically includes: Obtain the significance coefficient and local calcification feature corresponding to each sub-image block; Normalize all the significance coefficients to obtain the standard significance coefficient of each significance coefficient; Perform significance weighting on each standard significance coefficient and the corresponding local calcification feature to obtain the significant features of each sub-image block; Combine all the significant features to obtain the global calcification features corresponding to the ultrasound image of the thyroid nodule to be recognized.

[0011] In some embodiments, an ultrasound device is used to scan the thyroid part of a patient to obtain an ultrasound image of the thyroid nodule to be recognized.

[0012] In a second aspect, the present application provides an automatic recognition system for ultrasound images of thyroid nodules, including: An acquisition module, configured to acquire an ultrasound image of a thyroid nodule to be recognized and convert the ultrasound image of the thyroid nodule to be recognized into a nodule grayscale image; A processing module, configured to divide the nodule grayscale image into multiple sub-image blocks, select a sub-image block as the selected sub-image block, determine multiple calcified regions of the thyroid nodule in the selected sub-image block, and determine the local calcification features of the selected sub-image block based on the convolutional kernels of the selected sub-image block at different scale channels and all the calcified regions; The processing module is further configured to determine the recognition depth when recognizing the thyroid nodule for the selected sub-image block; The processing module is further configured to determine the significance coefficient of the selected sub-image block based on the recognition depth and the aspect ratio of the thyroid nodule in the selected sub-image block, and continue to determine the local calcification features and significance coefficients of the remaining sub-image blocks; An execution module, configured to perform feature concatenation on all local calcification features based on respective significance coefficients to obtain global calcification features corresponding to the ultrasound image of the thyroid nodule to be recognized.

[0013] In a third aspect, the present application provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to obtain the code and execute the above-mentioned automatic recognition method for the ultrasound image of the thyroid nodule.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned automatic recognition method for the ultrasound image of the thyroid nodule is implemented.

[0015] The technical solutions provided by the disclosed embodiments of the present application have the following beneficial effects: In the automatic recognition method and system for the ultrasound image of the thyroid nodule provided by the present application, first, an ultrasound image of the thyroid nodule to be recognized is obtained, and the ultrasound image of the thyroid nodule to be recognized is converted into a nodule grayscale image; secondly, the nodule grayscale image is segmented into a plurality of sub-image blocks, and one sub-image block is selected as the selected sub-image block, and a plurality of calcification regions of the thyroid nodule in the selected sub-image block are determined, and local calcification features of the selected sub-image block are determined based on the convolution kernels of the selected sub-image block in different scale channels and all the calcification regions; further, the recognition depth when recognizing the thyroid nodule for the selected sub-image block is determined; then, based on the recognition depth and the aspect ratio of the thyroid nodule in the selected sub-image block, the significance coefficient of the selected sub-image block is determined, and the local calcification features and significance coefficients of the remaining sub-image blocks are continuously determined; finally, feature concatenation is performed on all the local calcification features based on respective significance coefficients to obtain global calcification features corresponding to the ultrasound image of the thyroid nodule to be recognized.

[0016] It can be seen that the present application can perform multi-dimensional identification of calcified sites under the interference of various factors, so as to accurately describe the thyroid calcification characteristics. First, the ultrasonic image of the thyroid nodule to be identified is converted into a nodule grayscale image to prominently reflect the morphology and other pathological characteristics of the thyroid nodule. Secondly, the nodule grayscale image is segmented into multiple sub-image blocks to highlight the processing of local characteristics, thereby improving the accuracy and comprehensiveness of identification. Further, the local calcification characteristics of the sub-image blocks are determined to comprehensively describe the information of the calcified areas in the sub-image blocks, which is beneficial to improving the efficiency of subsequent diagnosis and pathological identification. Then, the identification depth for identifying the thyroid nodule in the sub-image block is determined to describe the degree of excavation of the relevant information of the thyroid nodule in the sub-image block during the subsequent pathological analysis process, so as to dynamically adjust the processing depth when analyzing the sub-image block, thereby improving the efficiency and accuracy of identification. In addition, based on the identification depth and the aspect ratio of the thyroid nodule in the sub-image block, the significance coefficient of the sub-image block is determined, which improves the expression ability of global features and provides higher reliability and effectiveness support for subsequent thyroid nodule identification. Finally, based on each significance coefficient and all local calcification characteristics, the global calcification characteristics corresponding to the ultrasonic image of the thyroid nodule to be identified are obtained, integrating the significance and spatial distribution information of all local thyroid nodule regions, so as to provide a global information description for subsequent calcification classification, pathological detection or diagnosis tasks. In summary, the technical solution provided by the present application can perform multi-dimensional identification of calcified sites under the interference of various factors, so as to accurately describe the thyroid calcification characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is an exemplary flowchart of an automatic identification method for ultrasonic images of thyroid nodules according to some embodiments of the present application; Figure 2 is an exemplary flowchart of determining the identification depth for identifying a thyroid nodule in a selected sub-image block according to some embodiments of the present application; Figure 3 is an exemplary flowchart of determining the significance coefficient of a selected sub-image block according to some embodiments of the present application; Figure 4 is a schematic structural diagram of an automatic identification system for ultrasonic images of thyroid nodules according to some embodiments of the present application; Figure 5 is a schematic structural diagram of a computer device for implementing the automatic identification method for ultrasonic images of thyroid nodules according to some embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0019] Reference Figure 1 , which is an exemplary flowchart of an automatic recognition method for thyroid nodule ultrasound images shown in some embodiments of the present application. The automatic recognition method 100 for thyroid nodule ultrasound images mainly includes the following steps: In step 101, obtain the thyroid nodule ultrasound image to be recognized, and convert the thyroid nodule ultrasound image to be recognized into a nodule grayscale image.

[0020] Specifically, when implementing, obtain the thyroid nodule ultrasound image to be recognized, that is: a doctor uses an ultrasound device, such as an ultrasound probe, etc., to scan the thyroid area of a thyroid nodule patient, and use the obtained high-resolution nodule ultrasound image as the thyroid nodule ultrasound image to be recognized. In addition, in other embodiments, other methods can also be used to obtain the thyroid nodule ultrasound image to be recognized, which is not limited here.

[0021] It should be noted that the thyroid nodule ultrasound image to be recognized in the present application refers to an ultrasound image used for subsequent analysis of thyroid nodule characteristics and pathological diagnosis, including possible lesion areas within the thyroid tissue and its structure.

[0022] Specifically, when implementing, convert the thyroid nodule ultrasound image to be recognized into a nodule grayscale image through image processing software. Specifically, the thyroid nodule ultrasound image to be recognized can be converted into a nodule grayscale image through OpenCV. The specific conversion process will not be elaborated here. In addition, in other embodiments, other image processing software can also be used to convert the thyroid nodule ultrasound image to be recognized into a nodule grayscale image, such as Photoshop, etc., which is not limited here. The nodule grayscale image refers to an image of the thyroid nodule displayed in grayscale in ultrasound imaging, containing the grayscale value of each pixel point. Converting to a nodule grayscale image helps to observe the morphology and other pathological characteristics of the thyroid nodule.

[0023] In step 102, divide the nodule grayscale image into multiple sub-image blocks, select one sub-image block as the selected sub-image block, determine multiple calcification regions of the thyroid nodule in the selected sub-image block, and determine the local calcification characteristics of the selected sub-image block based on the convolution kernels of the selected sub-image block in different scale channels and all the calcification regions.

[0024] In some embodiments, the nodule grayscale image can be divided into multiple sub-image blocks in the following manner, that is: Perform normalization processing on the nodule grayscale image to obtain a standard grayscale image; Cut the standard grayscale image to obtain multiple image partitions; Extract multiple sub-image blocks from all the image partitions.

[0025] In specific implementation, the nodule grayscale image is normalized to obtain a standard grayscale image, that is: the maximum grayscale value and the minimum grayscale value in the nodule grayscale image are extracted, the difference between the maximum grayscale value and the minimum grayscale value is used as the normalization standard, for each pixel point, the grayscale value corresponding to the pixel point is subtracted from the minimum grayscale value, the difference result is divided by the normalization standard, and the calculation result is used as the standard grayscale value of the pixel point, thereby obtaining a standard grayscale image.

[0026] It should be noted that in this embodiment, the standard grayscale image represents the grayscale image obtained after normalizing the nodule grayscale image. Through normalization, the grayscale values of the image can be mapped to the range of 0 to 1, thereby eliminating the influence caused by the brightness difference of the imaging device.

[0027] In specific implementation, the standard grayscale image is sliced by row and column order to obtain multiple image partitions. Specifically, by the fixed grid segmentation method, the standard grayscale image is sliced by row and column order, and the standard grayscale image is equally divided into regular rectangular image partitions, thereby obtaining multiple image partitions. In addition, in other embodiments, other segmentation methods can also be used to slice the standard grayscale image, for example, the sliding window segmentation method, the adaptive block method, etc., which are not limited here.

[0028] In specific implementation, multiple sub-image blocks are extracted from all the image partitions, that is: a valid partition threshold is preset. For each image partition, if the proportion of the valid grayscale area (i.e., non-zero pixel points) of the image partition is greater than the valid partition threshold, the image partition is used as a sub-image block, otherwise it is not processed, thereby obtaining multiple sub-image blocks.

[0029] It should be noted that in this application, the sub-image block represents an image partition that is helpful for subsequent pathological analysis. By determining the sub-image block, the processing of local characteristics can be highlighted, thereby improving the accuracy and comprehensiveness of recognition. In addition, the valid partition threshold can be set according to actual application requirements, which is not limited here.

[0030] In some embodiments, the following method can be used to determine multiple calcified regions of the thyroid nodule in the selected sub-image block, that is: Extract multiple calcification candidate points of the thyroid nodule from the selected sub-image block; Based on all the calcification candidate points, determine multiple suspicious calcified regions of the selected sub-image block; Extract multiple calcified regions of the thyroid nodule from all the suspicious calcified regions.

[0031] In specific implementation, multiple calcification candidate points of the thyroid nodule are extracted from the selected sub-image block through binarization processing. Specifically, a grayscale threshold is preset, and the grayscale values of all pixel points in the selected sub-image block are respectively compared with the grayscale threshold. The grayscale values greater than the grayscale threshold are extracted, and the pixel points corresponding to the extracted grayscale values are used as calcification candidate points, thereby obtaining multiple calcification candidate points of the thyroid nodule.

[0032] It should be noted that in this embodiment, the calcification candidate points represent pixel points that currently have certain calcification site attributes but are not yet clear, and further verification is required for them.

[0033] In specific implementation, multiple suspicious calcification regions of the selected sub-image block are determined based on all the calcification candidate points, that is: the connected components of all the calcification candidate points are extracted through the connected component labeling algorithm, and all the extracted independent regions are used as multiple suspicious calcification regions of the selected sub-image block. The specific implementation process of the connected component labeling algorithm will not be elaborated here. In addition, in other embodiments, other connected algorithms can also be used to determine the suspicious calcification regions, which are not limited here.

[0034] It should be noted that in this embodiment, the suspicious calcification regions represent some regions with calcification characteristic manifestations found in the ultrasonic image examination of the thyroid nodule but cannot be completely determined as typical calcifications. Due to the influence of factors such as equipment and light, the surrounding nodule tissue regions or image noise present characteristics similar to those of the calcification regions. Therefore, further processing and analysis of the suspicious calcification regions are required.

[0035] In specific implementation, multiple calcification regions of the thyroid nodule are extracted from all the suspicious calcification regions, that is: a feature threshold is preset, a suspicious calcification region is selected, the area and perimeter of the suspicious calcification region are extracted through Open CV, and the area and the perimeter are substituted into the circularity calculation formula. If the calculation result is greater than the feature threshold, then the suspicious calcification region is recognized as a calcification region, thereby obtaining multiple calcification regions of the thyroid nodule.

[0036] It should be noted that in this application, the calcification regions represent regions with specific imaging manifestations formed due to calcium salt deposition inside or at the edge of the thyroid nodule, and the calcification regions usually have a relatively high circularity. Accurately identifying the calcification regions is beneficial for subsequent pathological analysis and judgment of the thyroid nodule, thereby improving the accuracy of diagnosis.

[0037] In some embodiments, the local calcification features of the selected sub-image block can be determined based on the convolution kernels of the selected sub-image block in different scale channels and all the calcification regions by the following method, that is: Set different scale channels, and initialize multiple convolution kernels under each scale channel; Perform convolution operations on the selected sub-image patch through all convolutional kernels to obtain multiple convolutional kernel response maps; For each scale channel, perform feature extraction on all calcified regions based on the convolutional kernel response maps under the scale channel to obtain the feature vector corresponding to the scale channel, and then obtain the feature vectors corresponding to each scale channel; Fuse all the feature vectors to obtain the multi-dimensional calcification features of the selected sub-image patch; Optimize the features of the multi-dimensional calcification features to obtain the local calcification features of the selected sub-image patch.

[0038] Specifically, when implementing, set different scale channels and initialize multiple convolutional kernels under each scale channel. That is: First, set multiple scale channels, and the size of the specific scale channel can be set according to actual accuracy requirements. Then, initialize multiple convolutional kernels under each scale channel through the random initialization method. The specific number of convolutional kernels can be adjusted according to actual needs and computing resources. By setting multiple convolutional kernels, feature information can be more comprehensively extracted from different directions and frequencies.

[0039] Specifically, when implementing, perform convolution operations on the selected sub-image patch through all convolutional kernels to obtain multiple convolutional kernel response maps. That is: Select a convolutional kernel, start from the upper left corner of the selected sub-image patch, and each time take a pixel region of the same size as the convolutional kernel and perform the operation of multiplying the corresponding elements and summing them with the convolutional kernel to obtain a new pixel value block as the convolution result. Use a sliding window with a step size of 1 to traverse the entire selected sub-image patch according to the above steps to obtain the convolutional kernel response map of the convolutional kernel, and continue to determine the convolutional kernel response maps of the remaining convolutional kernels.

[0040] Among them, in some embodiments, performing feature extraction on all calcified regions based on the convolutional kernel response maps under the scale channel to obtain the feature vector corresponding to the scale channel can adopt the following method, that is: Select a convolutional kernel response map as the selected convolutional kernel response map, and map each calcified region to the selected convolutional kernel response map to obtain multiple mapped regions; Determine the calcification statistical features corresponding to the selected convolutional kernel response map according to all the mapped regions; Continue to determine the corresponding calcification statistical features of the remaining convolutional kernel response maps; Combine all the calcification statistical features to obtain the feature vector corresponding to the scale channel.

[0041] In specific implementation, each calcified region is correspondingly mapped to the selected convolution kernel response map to obtain multiple mapped regions, that is: each calcified region is mapped to the corresponding position of the selected convolution kernel response map to obtain the mapped regions corresponding to each position, and then multiple mapped regions are obtained. It should be noted that in this embodiment, the mapped region represents the region of the calcified region at the corresponding position in the convolution kernel response map, and the calcification statistical feature in this embodiment represents the feature representation of all pixel points in the calcified region.

[0042] In specific implementation, the calcification statistical feature corresponding to the selected convolution kernel response map is determined according to all the mapped regions, that is: the mean, variance, maximum value, and minimum value of the pixel values of all pixel points in each mapped region are calculated respectively, and all the calculated means, variances, maximum values, and minimum values are combined into the calcification statistical feature of the selected convolution kernel response map.

[0043] It should also be noted that in this embodiment, the numpy component in Python is used to calculate the mean, variance, maximum value, and minimum value of the pixel values of all pixel points in each mapped region. In addition, in other embodiments, other calculation methods can also be used for calculation, which is not limited here. In addition, when combining all the calcification statistical features, it is in accordance with the sequence of convolution operations.

[0044] In specific implementation, all the feature vectors are fused in the order of the size of the scale channels to obtain the multi-dimensional calcification feature of the selected sub-image block. The multi-dimensional calcification feature represents the comprehensive linear expression of the mathematical statistics features of all calcified regions. By determining the multi-dimensional calcification feature, the information of the calcified regions in the selected sub-image block can be comprehensively described, but the data volume is relatively large, so it needs to be optimized to improve the recognition efficiency.

[0045] Among them, in some embodiments, the following method can be used to optimize the multi-dimensional calcification feature to obtain the local calcification feature of the selected sub-image block, that is: The multi-dimensional calcification feature is normalized to obtain the normalized multi-dimensional calcification feature; The local calcification feature of the selected sub-image block is extracted from the normalized multi-dimensional calcification feature.

[0046] In specific implementation, the multi-dimensional calcification feature can be normalized by the maximum-minimum normalization method to obtain the normalized multi-dimensional calcification feature, which will not be elaborated here.

[0047] In specific implementation, the local calcification feature of the selected sub-image block is extracted from the normalized multi-dimensional calcification feature by the principal component analysis method, that is: the principal component analysis method is used to calculate the normalized multi-dimensional calcification feature to obtain multiple eigenvalues and the eigenvectors corresponding to each eigenvalue, and the eigenvectors corresponding to each eigenvalue are combined to obtain the local calcification feature.

[0048] It should also be noted that in the present application, the local calcification feature represents the feature information of the calcified area in the sub-image block. By determining the local calcification feature, the detailed features of the local calcification of the thyroid nodule can be effectively identified, and then the detailed features of the local area can provide conditions for obtaining accurate thyroid nodule calcification features subsequently, thereby improving the efficiency and accuracy of subsequent diagnosis and pathological identification.

[0049] In step 103, determine the recognition depth when identifying the thyroid nodule for the selected sub-image block.

[0050] In some embodiments, referring to Figure 2 As shown, this figure is an exemplary flowchart for determining the recognition depth when identifying the thyroid nodule for the selected sub-image block according to some embodiments of the present application. In this embodiment, the recognition depth when identifying the thyroid nodule for the selected sub-image block can be implemented by the following steps: First, in step 1031, obtain the steady-state recognition interval when identifying the thyroid nodule. Then, in step 1032, determine the regional complexity index of the selected sub-image block based on the gray gradient distribution and calcification density in the selected sub-image block. Furthermore, in step 1033, determine the recognition influence coefficient when identifying the thyroid nodule for the selected sub-image block based on the thyroid nodule area ratio corresponding to the selected sub-image block and the thyroid nodule area ratio corresponding to the remaining sub-image blocks. Finally, in step 1034, determine the recognition depth when identifying the thyroid nodule for the selected sub-image block through the steady-state recognition interval, the regional complexity index, and the recognition influence coefficient.

[0051] It should be noted that in this embodiment, the steady-state recognition interval represents the processing depth range that can be dynamically adjusted in the subsequent feature extraction and analysis process. The lower limit of the interval is used as the minimum depth, and the upper limit of the interval is used as the maximum depth. For regions with more complex features, the processing depth in the subsequent feature extraction and analysis process is closer to the maximum depth, and for regions with simpler features, the processing depth in the subsequent feature extraction and analysis process is closer to the minimum depth. By determining the steady-state recognition interval, the hierarchical complexity of processing the image in the image recognition task can be controlled.

[0052] Among them, in some embodiments, the regional complexity index of the selected sub-image block can be determined based on the gray gradient distribution and calcification density in the selected sub-image block in the following manner, that is: Determine the gray gradient distribution in the selected sub-image block. Obtain multiple calcified regions of the selected sub-image block, and determine the calcification density in the selected sub-image block based on all the calcified regions. Combined with the grayscale gradient distribution and the calcification density, the complexity of the selected sub-image block is evaluated to obtain the regional complexity index of the selected sub-image block.

[0053] In specific implementation, the grayscale gradient distribution in the selected sub-image block is determined, that is, each pixel point in the selected sub-image block is calculated through the Sobel gradient operator, and then the grayscale gradient distribution in the selected sub-image block is obtained. The specific calculation process is not elaborated here. In addition, in other embodiments, other gradient operators can also be used for calculation, such as the Scharr operator, the Prewitt operator, etc., which are not limited here.

[0054] In specific implementation, the calcification density in the selected sub-image block is determined based on all calcified regions, that is, the sum of the areas of all calcified regions is calculated through OpenCV, and the sum of the areas is divided by the area of the selected sub-image block, and the calculation result is used as the calcification density in the selected sub-image block. In addition, in other embodiments, other calculation methods can also be used to calculate the calcification density, which are not limited here.

[0055] It should be noted that in this embodiment, the calcification density represents the concentration degree of the calcified region in the thyroid nodule. By determining the calcification density, the distribution characteristics of the calcified substances in the nodule tissue can be effectively characterized, thus providing data support for subsequent pathological diagnosis.

[0056] In specific implementation, the complexity of the selected sub-image block is evaluated through a preset feature complexity model combined with the grayscale gradient distribution and the calcification density to obtain the regional complexity index of the selected sub-image block. Specifically, the grayscale gradient distribution and the calcification density are used as the input of the model, and the output result is used as the regional complexity index of the selected sub-image block. The feature complexity model represents a model for extracting the complex characteristics of image regions trained by a large amount of historical data, which is not elaborated here.

[0057] It should be noted that in this embodiment, the regional complexity index represents the variability degree of the feature information contained in the selected sub-image block. The larger the regional complexity index, the greater the variability degree of the feature information contained in the selected sub-image block, and the smaller the regional complexity index, the smaller the variability degree of the feature information contained in the selected sub-image block. By determining the regional complexity index, the local feature information content can be measured.

[0058] During specific implementation, the recognition influence coefficient for thyroid nodule recognition of the selected sub-image block is determined based on the thyroid nodule area ratio corresponding to the selected sub-image block and the thyroid nodule area ratio corresponding to the remaining sub-image blocks, that is: the quotient of the thyroid nodule area in the selected sub-image block and the total area of the selected sub-image block is used as the thyroid nodule area ratio of the selected sub-image block, and the quotient of the thyroid nodule area in the remaining sub-image blocks and the total area of the remaining sub-image blocks is used as the thyroid nodule area ratio of the remaining sub-image blocks. Calculate the difference between the thyroid nodule area ratio of the selected sub-image block and the thyroid nodule area ratio of the remaining sub-image blocks, and use the absolute value of the difference result as the recognition influence coefficient of the selected sub-image block.

[0059] It should be noted that in this embodiment, the recognition influence coefficient represents the degree of distribution difference in the area size of the thyroid nodule in the selected sub-image block and the remaining sub-image blocks. The smaller the recognition influence coefficient, the more consistent the processing depth of the selected sub-image block and the remaining sub-image blocks in the subsequent recognition process. The larger the recognition influence coefficient, it indicates that there is a relatively special tissue distribution of the thyroid nodule in the selected sub-image block. For example, most of the nodule is located in the selected sub-image block. At this time, it is necessary to deepen the processing depth of the selected sub-image block. By determining the recognition influence coefficient, the importance of the sub-image block in the subsequent recognition process can be measured, so as to dynamically adjust the processing depth and thus improve the recognition efficiency.

[0060] Among them, in some embodiments, the recognition depth for thyroid nodule recognition of the selected sub-image block can be determined by the following method based on the steady-state recognition interval, the regional complexity index, and the recognition influence coefficient, that is: Use the product of the regional complexity index and the recognition influence coefficient as the dynamic mapping coefficient; Determine the recognition depth for thyroid nodule recognition of the selected sub-image block based on the steady-state recognition interval and the dynamic mapping coefficient.

[0061] It should be noted that in this embodiment, the dynamic mapping coefficient represents a parameter for dynamically controlling the value of the required recognition depth.

[0062] During specific implementation, the recognition depth for thyroid nodule recognition of the selected sub-image block is determined based on the steady-state recognition interval and the dynamic mapping coefficient, that is: extract the interval length and the minimum depth of the steady-state recognition interval, divide the product of the interval length and the dynamic mapping coefficient by the standard mapping coefficient, and add the quotient result to the minimum depth. The added result is the recognition depth for thyroid nodule recognition of the selected sub-image block.

[0063] It should be noted that in this embodiment, the standard mapping coefficient represents the product of the maximum regional complexity index and the maximum recognition influence coefficient obtained in historical experiments. The value range of the recognition depth can be corrected through the standard mapping coefficient.

[0064] It should also be noted that in this application, the recognition depth represents the degree to which the subsequent pathological analysis process mines the information related to thyroid nodules in the selected sub-image block. The smaller the recognition depth, the shallower the degree to which the subsequent pathological analysis process mines the information related to thyroid nodules in the selected sub-image block. The larger the recognition depth, the deeper the degree to which the subsequent pathological analysis process mines the information related to thyroid nodules in the selected sub-image block. By determining the recognition depth, the processing depth for analyzing each sub-image block can be dynamically adjusted, thereby improving the efficiency and accuracy of recognition.

[0065] In step 104, based on the recognition depth and the aspect ratio of the thyroid nodule in the selected sub-image block, the significance coefficient of the selected sub-image block is determined, and then the local calcification feature and the significance coefficient of the remaining sub-image blocks are continuously determined.

[0066] In some embodiments, as shown in Figure 3 The figure is an exemplary flowchart for determining the significance coefficient of the selected sub-image block according to some embodiments of the present application. In this embodiment, the significance coefficient of the selected sub-image block based on the recognition depth and the aspect ratio of the thyroid nodule in the selected sub-image block can be implemented by the following steps: First, in step 1041, based on the recognition depth, the feature sensitivity when the selected sub-image block performs feature recognition on the thyroid nodule is determined; Then, in step 1042, the aspect ratio of the thyroid nodule in the selected sub-image block is determined; Furthermore, in step 1043, based on the aspect ratio, the penalty factor for the thyroid nodule in the selected sub-image block deviating from the ideal state is determined; Finally, in step 1044, based on the feature sensitivity and the penalty factor, the significance coefficient of the selected sub-image block is determined.

[0067] Specifically, when implementing, based on the recognition depth, the feature sensitivity when the selected sub-image block performs feature recognition on the thyroid nodule is determined, that is: the ideal depth in the historical recognition experience is obtained, the difference between the recognition depth and the ideal depth is used as the sensitive difference coefficient, and the product of the sensitive difference coefficient and the correction experience coefficient is used as the feature sensitivity when the selected sub-image block performs feature recognition on the thyroid nodule. Both the ideal depth and the correction experience coefficient are summarized from the historical recognition experience and will not be elaborated here.

[0068] It should be noted that in this embodiment, the feature sensitivity represents the response degree of various features in the selected sub-image block when being recognized. Additionally, the sensitive difference coefficient represents the difference degree between the current recognition depth and the ideal depth. By determining the feature sensitivity, the physical rationality and context correlation adaptability during feature recognition can be ensured.

[0069] Among them, in some embodiments, the aspect ratio of the thyroid nodule in the selected sub-image block can be determined in the following manner, that is: Extract the edge contour of the thyroid nodule in the selected sub-image block; Determine the minimum bounding rectangle of the edge contour; Determine the aspect ratio of the thyroid nodule in the selected sub-image block according to the major axis and the principal axis of the minimum bounding rectangle.

[0070] In specific implementation, the edge contour of the thyroid nodule in the selected sub-image block is extracted through a contour detection algorithm. For example, the findContours algorithm in Open CV can be used to extract the edge contour of the thyroid nodule in the selected sub-image block. In addition, in other embodiments, other methods can also be used to extract the edge contour, which is not limited here.

[0071] It should be noted that in this embodiment, the edge contour represents a closed contour describing the thyroid nodule part in the tile. Additionally, the contour detection algorithm is an existing technology in image processing, and the specific detection process will not be elaborated here.

[0072] In specific implementation, determine the minimum bounding rectangle of the edge contour, that is: traverse to find the smallest rectangle that can completely enclose the edge contour, and use it as the minimum bounding rectangle of the edge contour. The specific process will not be elaborated here.

[0073] In specific implementation, determine the aspect ratio of the thyroid nodule in the selected sub-image block according to the major axis and the principal axis of the minimum bounding rectangle, that is: obtain the major axis and the principal axis of the minimum bounding rectangle through geometric analysis, and use the ratio of the major axis to the principal axis as the aspect ratio of the thyroid nodule in the selected sub-image block. The specific geometric analysis process will not be elaborated here. In addition, in other embodiments, other analysis techniques can also be used to obtain the aspect ratio, which is not limited here.

[0074] It should be noted that in this application, the aspect ratio represents the morphological parameter of the thyroid nodule in the selected sub-image block. The closer the aspect ratio is to 1, the closer the morphology of the thyroid nodule in the selected sub-image block is to a circle. The larger the aspect ratio, the more slender the morphology of the thyroid nodule in the selected sub-image block. By determining the aspect ratio, more comprehensive geometric feature support can be provided for the subsequent analysis of the thyroid nodule.

[0075] In specific implementation, a penalty factor for the deviation of the thyroid nodule in the selected sub-image block from the ideal state is determined based on the aspect ratio, that is: the ideal aspect ratio of the ideal thyroid nodule shape is obtained, and the absolute value of the difference between the aspect ratio and the ideal aspect ratio is used as the penalty factor for the deviation of the thyroid nodule in the selected sub-image block from the ideal state. In addition, in other embodiments, other calculation methods may also be used to determine the penalty factor, which is not limited here.

[0076] It should be noted that in this embodiment, the penalty factor represents the degree of reduction in the response to the features of the selected sub-image block during the subsequent recognition and analysis process. Since the thyroid nodule area deviating from the ideal shape will affect the clarity of the judgment of pathological features, the degree of feature extraction can be corrected and controlled by determining the penalty factor.

[0077] In specific implementation, a significance coefficient of the selected sub-image block is determined based on the feature sensitivity and the penalty factor, that is: the product of the feature sensitivity and the penalty factor is used as the significance coefficient of the selected sub-image block.

[0078] It should be noted that in this application, the significance coefficient represents the contribution degree of the thyroid nodule part and the calcification condition in the selected sub-image block to the subsequent medical recognition. The larger the significance coefficient, the greater the contribution degree of the thyroid nodule part and the calcification condition in the selected sub-image block to the subsequent medical recognition. The smaller the significance coefficient, the smaller the contribution degree of the thyroid nodule part and the calcification condition in the selected sub-image block to the subsequent medical recognition. By determining the significance coefficient, the importance of the calcification feature information at different scales can be dynamically adjusted, improving the expression ability of the global features, thereby providing higher reliability and effectiveness support for the subsequent thyroid nodule recognition.

[0079] It should be noted that the implementation steps of "determining the local calcification feature of the selected sub-image block based on the convolution kernel of the selected sub-image block in different scale channels and all calcification regions" and "determining the significance coefficient of the selected sub-image block based on the recognition depth and the aspect ratio of the thyroid nodule in the selected sub-image block" are continued to determine the local calcification feature and significance coefficient of the remaining sub-image blocks, which will not be elaborated here.

[0080] In step 105, feature concatenation is performed on all local calcification features based on each significance coefficient to obtain the global calcification feature corresponding to the thyroid nodule ultrasound image to be recognized.

[0081] In some embodiments, feature concatenation is performed on all local calcification features based on each significance coefficient to obtain the global calcification feature corresponding to the thyroid nodule ultrasound image to be recognized, which can be implemented in the following manner, that is: Obtain the significance coefficient and local calcification feature corresponding to each sub-image block; Normalize all the significance coefficients to obtain the standard significance coefficient for each significance coefficient; Perform significance weighting on each standard significance coefficient and the corresponding local calcification feature to obtain the significant feature of each sub-image block; Combine all the significant features to obtain the global calcification feature corresponding to the thyroid nodule ultrasound image to be recognized.

[0082] Specifically, when implementing, normalize all the significance coefficients to obtain the standard significance coefficient for each significance coefficient, that is: select a significance coefficient, divide this significance coefficient by the sum of all significance coefficients, and use the quotient result as the standard significance coefficient of this significance coefficient, thereby obtaining the standard significance coefficient for each significance coefficient.

[0083] It should be noted that in this embodiment, the standard significance coefficient represents the importance or contribution weight coefficient of the sub-image block corresponding to this standard significance coefficient in the overall feature representation.

[0084] Specifically, when implementing, perform significance weighting on each standard significance coefficient and the corresponding local calcification feature to obtain the significant feature of each sub-image block, that is: select a standard significance coefficient, multiply this standard significance coefficient by the corresponding local calcification feature, and use the multiplication result as the significant feature of the sub-image block corresponding to this standard significance coefficient, thereby obtaining the significant feature of each sub-image block.

[0085] It should be noted that in this embodiment, the significant feature represents the feature extracted from the sub-image block that can reflect the local information characteristics and has high discrimination ability for the overall recognition task. By determining the significant feature, the feature expression ability of the thyroid nodule ultrasound image can be enhanced, thereby improving the detection accuracy of thyroid nodule recognition.

[0086] Specifically, when implementing, combine all the significant features to obtain the global calcification feature corresponding to the thyroid nodule ultrasound image to be recognized, that is: sort all the standard significance coefficients in descending order, combine the significant features corresponding to all the standard significance coefficients according to this sorting result to obtain a global feature, and use this feature as the global calcification feature corresponding to the thyroid nodule ultrasound image to be recognized.

[0087] It should be noted that in this application, the global calcification feature represents an information set that is highly generalized and can comprehensively reflect the characteristics of the calcified area in the entire thyroid nodule ultrasound image to be recognized. It integrates the significance and spatial distribution information of all local thyroid nodule regions. By determining the global calcification feature, the thyroid nodule ultrasound image to be recognized can be marked, which is beneficial to providing a global information description for subsequent calcification classification, pathological detection, or diagnosis tasks.

[0088] In addition, on the other hand of the present application, in some embodiments, the present application provides an automatic recognition system for thyroid nodule ultrasound images, with reference to Figure 4 , which is a schematic structural diagram of an automatic recognition system for thyroid nodule ultrasound images shown in some embodiments of the present application. The automatic recognition system 200 for thyroid nodule ultrasound images includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described as follows: The acquisition module 201. In the present application, the acquisition module 201 is mainly used to acquire the thyroid nodule ultrasound image to be recognized and convert the thyroid nodule ultrasound image to be recognized into a nodule grayscale image; The processing module 202. In the present application, the processing module 202 is mainly used to divide the nodule grayscale image into multiple sub-image blocks, select one sub-image block as the selected sub-image block, determine multiple calcified regions of the thyroid nodule in the selected sub-image block, and determine the local calcification features of the selected sub-image block based on the convolution kernels of the selected sub-image block in different scale channels and all the calcified regions; The processing module 202 is further used to determine the recognition depth when recognizing the thyroid nodule for the selected sub-image block; In addition, the processing module 202 is further used to determine the significance coefficient of the selected sub-image block based on the recognition depth and the aspect ratio of the thyroid nodule in the selected sub-image block, and continue to determine the local calcification features and significance coefficients of the remaining sub-image blocks; The execution module 203. In the present application, the execution module 203 is mainly used to perform feature concatenation on all the local calcification features based on each significance coefficient to obtain the global calcification features corresponding to the thyroid nodule ultrasound image to be recognized.

[0089] In addition, the present application also provides a computer device. The computer device includes a memory and a processor. The memory stores code, and the processor is configured to obtain the code and execute the above-mentioned automatic recognition method for thyroid nodule ultrasound images.

[0090] In some embodiments, with reference to Figure 5 , which is a schematic structural diagram of a computer device for implementing the automatic recognition method for thyroid nodule ultrasound images shown in some embodiments of the present application. The automatic recognition method for thyroid nodule ultrasound images in the above embodiments can be implemented by Figure 5 The computer device shown includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0091] The processor 301 can be a general - purpose central processing unit (CPU), or an application - specific integrated circuit (ASIC), or one or more are used to control the execution of the automatic recognition method for thyroid nodule ultrasound images in this application.

[0092] The communication bus 302 can be used to transfer information between the above - mentioned components.

[0093] The memory 303 can be a read - only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM), or other types of dynamic storage devices that can store information and instructions. It can also be an electrically erasable programmable read - only memory (EEPROM), a compact disc read - only memory (CDROM), or other optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile discs, Blu - ray discs, etc.), a magnetic disk, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to this. The memory 303 can exist independently and be connected to the processor 301 through the communication bus 302. The memory 303 can also be integrated with the processor 301.

[0094] Among them, the memory 303 is used to store the program code for executing the solution of this application and is controlled by the processor 301 to execute. The processor 301 is used to execute the program code stored in the memory 303. The program code can include one or more software modules. The determination of the automatic recognition method for thyroid nodule ultrasound images in the above - mentioned embodiments can be implemented by one or more software modules in the program code of the processor 301 and the memory 303.

[0095] The communication interface 304, using any device such as a transceiver, is used to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0096] In a specific implementation, as an example, a computer device may include multiple processors, and each of these processors may be a single-core (single CPU) processor or a multi-core (multi CPU) processor. The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions).

[0097] The above computer device may be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the computer device may be a desktop computer, a laptop computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of the computer device.

[0098] In addition, the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the automatic recognition method of the thyroid nodule ultrasound image described above is implemented.

[0099] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0100] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these changes and modifications.

Claims

1. A method for automatically identifying thyroid nodules in ultrasonic images, characterized in that: The steps include: Acquire an ultrasonic image of a thyroid nodule to be identified, and convert the ultrasonic image of the thyroid nodule to be identified into a grayscale image of the nodule; Segmenting the nodule grayscale image into multiple sub-image blocks, selecting a sub-image block as a selected sub-image block, determining multiple calcification regions of the thyroid nodule in the selected sub-image block, and determining local calcification features of the selected sub-image block based on convolution kernels of the selected sub-image block at different scale channels and all calcification regions; Determine the recognition depth when performing thyroid nodule recognition on the selected sub-image block; Determining a significance coefficient of the selected sub-image block based on the identification depth and the aspect ratio of the thyroid nodule in the selected sub-image block, and continuing to determine local calcification features and significance coefficients of the remaining sub-image blocks; All local calcification features are cascaded based on the significance coefficients to obtain the global calcification features corresponding to the ultrasound image of the thyroid nodule to be identified.

2. The method according to claim 1, characterized in that Segmenting the nodule grayscale image into a plurality of sub-image blocks specifically includes: Normalizing the nodule grayscale image to obtain a standard grayscale image; Segmenting the standard grayscale image to obtain a plurality of image partitions; A plurality of sub-image blocks are extracted from all image partitions.

3. The method according to claim 1, characterized in that Determining multiple calcification regions of the thyroid nodule in the selected sub-image block specifically includes: Extracting multiple calcification candidate points of the thyroid nodule from the selected sub-image block; Determine multiple suspicious calcification areas of the selected sub-image block based on all calcification candidate points; Multiple calcification areas of thyroid nodules were extracted from all suspicious calcification areas.

4. The method according to claim 1, characterized in that The local calcification features of the selected sub-image block are determined based on the convolution kernels of the selected sub-image block at different scale channels and all calcification areas, specifically including: Set different scale channels and initialize multiple convolution kernels under each scale channel; Perform convolution operations on the selected sub-image blocks through all convolution kernels to obtain multiple convolution kernel response maps; For each scale channel, feature extraction is performed on all calcified areas based on the convolution kernel response maps under the scale channel to obtain the feature vector corresponding to the scale channel, and then the feature vector corresponding to each scale channel is obtained; All feature vectors are fused to obtain the multi-dimensional calcification features of the selected sub-image block; The multi-dimensional calcification feature is optimized to obtain the local calcification feature of the selected sub-image block.

5. The method according to claim 4, characterized in that Based on the convolution kernel response maps under the scale channel, all calcification areas are feature extracted, and the feature vectors corresponding to the scale channel are obtained, including: A convolution kernel response map is selected as a selected convolution kernel response map, and each calcification area is mapped to the selected convolution kernel response map to obtain a plurality of mapping areas; Determine the calcification statistical features corresponding to the selected convolution kernel response map according to all the mapped areas; Continue to determine the corresponding calcification statistical characteristics of the remaining convolution kernel response map; All calcification statistical features are combined to obtain a feature vector corresponding to the scale channel.

6. The method according to claim 1, characterized in that Based on each significance coefficient, all local calcification features are cascaded to obtain the global calcification features corresponding to the ultrasound image of the thyroid nodule to be identified, including: Obtaining the significance coefficient and local calcification features corresponding to each sub-image block; Standardize all the significant coefficients to obtain the standard significant coefficient of each significant coefficient; Each standard significant coefficient is weighted with the corresponding local calcification feature to obtain the significant features of each sub-image block; All significant features are combined to obtain the global calcification features corresponding to the ultrasound image of the thyroid nodule to be identified.

7. The method according to claim 1, characterized in that The patient's thyroid gland is scanned using an ultrasound device to obtain an ultrasound image of the thyroid nodule to be identified.

8. An automatic recognition system for thyroid nodules ultrasonic images, characterized in that: include: An acquisition module is used to acquire an ultrasonic image of a thyroid nodule to be identified and convert the ultrasonic image of the thyroid nodule to be identified into a grayscale image of the nodule; A processing module, used for dividing the nodule grayscale image into multiple sub-image blocks, selecting a sub-image block as a selected sub-image block, determining multiple calcification regions of the thyroid nodule in the selected sub-image block, and determining local calcification features of the selected sub-image block based on convolution kernels of the selected sub-image block at different scale channels and all calcification regions; The processing module is further used to determine the recognition depth when performing thyroid nodule recognition on the selected sub-image block; The processing module is further used to determine the significance coefficient of the selected sub-image block based on the identification depth and the aspect ratio of the thyroid nodule in the selected sub-image block, and continue to determine the local calcification characteristics and significance coefficients of the remaining sub-image blocks; The execution module is used to perform feature cascade on all local calcification features based on various significance coefficients to obtain the global calcification features corresponding to the ultrasound image of the thyroid nodule to be identified.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores codes, and the processor is configured to acquire the codes and execute the method for automatically identifying thyroid nodules in ultrasound images as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for automatically identifying thyroid nodules in ultrasonic images according to any one of claims 1 to 7 is implemented.

Images