An artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors

Abstract

The present invention relates to the field of artificial intelligence technology, and in particular, to an artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors, which includes: a data preparation unit for acquiring images related to a patient's thyroid nodules, and annotating, classifying and saving the acquired images in a standard format; a feature extraction unit for extracting nodule structural features and the echo contrast of nodules relative to surrounding normal thyroid tissues in grayscale ultrasonic images, and acquiring blood flow conditions in nodules in color Doppler ultrasonic images; a data analysis unit for inputting the extracted features into a statistical model to determine the malignancy and benignity of nodules; and a clinical combination unit for combining the analysis results with the patient's clinical information for comprehensive judgment. In this system, the accuracy of predictive diagnosis of thyroid tumors can be improved by considering the structural features, echoes and blood flow conditions of nodules together for judgment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors. Background Art

[0002] The thyroid gland is an important endocrine gland in the human body, located in the front of the neck. It is mainly responsible for the synthesis and secretion of thyroid hormones and the regulation of the body's metabolism. Thyroid diseases include many types, such as hyperthyroidism, hypothyroidism, thyroiditis, thyroid nodules, and thyroid cancer. Among them, thyroid nodules are one of the most common thyroid diseases. Most thyroid nodules are benign, but about 5% to 15% of nodules may be malignant, that is, thyroid cancer.

[0003] Despite the high incidence of thyroid nodules, the accuracy of differentiating benign from malignant nodules has always been a major clinical challenge. Traditional diagnostic methods, such as clinical examination, blood tests, ultrasound, and fine needle aspiration biopsy (FNAB), although they can provide useful information to a certain extent, still have some limitations. For example, the results of ultrasound examination are highly dependent on the skills and experience of the operator, while fine needle aspiration biopsy (FNAB) also has a certain false negative and false positive rate, and some nodules are difficult to be clearly diagnosed by FNAB;

[0004] Traditional echo contrast calculation methods may not be able to fully capture the subtle differences between nodules and surrounding normal tissues, resulting in insufficient diagnostic accuracy. Therefore, an artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors was designed. Summary of the invention

[0005] The purpose of the present invention is to provide an artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors to solve the problem raised in the above-mentioned background technology that the subtle differences between nodules and surrounding normal tissues cannot be fully captured, resulting in insufficient diagnostic accuracy.

[0006] To achieve the above objectives, the present invention provides an artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors, comprising:

[0007] A data preparation unit, the data preparation unit is used to obtain images related to the patient's thyroid nodules, and annotate, classify and save the obtained images in a standard format; wherein the images related to the patient's thyroid nodules include grayscale ultrasound images and color Doppler ultrasound images;

[0008] A feature extraction unit, the feature extraction unit is used to extract nodule structural features and the echo contrast of the nodule relative to the surrounding normal thyroid tissue in the grayscale ultrasonic image, introduce weighted average and grayscale co-occurrence matrix methods to optimize the echo contrast, and track the change trend of the nodule according to the extracted nodule structural features, and obtain the blood flow situation in the nodule in the color Doppler ultrasonic image;

[0009] A data analysis unit, the data analysis unit is used to input the extracted features into a statistical model, and train the model using a machine learning algorithm so that it can predict the properties of new nodules based on historical data, thereby determining the malignancy or benignity of the nodules;

[0010] The clinical combination unit is used to combine the analysis results with the patient's clinical information for comprehensive judgment. If the possibility of malignancy is high, further examination is performed.

[0011] As a further improvement of the technical solution, the feature extraction unit extracts nodule structural features from the grayscale ultrasound image, specifically as follows:

[0012] S1, preprocessing the grayscale ultrasonic image and removing noise interference in the image, the preprocessing at least includes grayscale normalization and size adjustment;

[0013] S2, segmenting the grayscale ultrasound image to identify the nodule area, and at the same time, performing manual correction based on the segmentation according to the actual situation;

[0014] S3. Evaluate the size of the nodule based on the segmented nodule area, identify the boundary of the nodule using edge detection, and evaluate the clarity of the boundary.

[0015] As a further improvement of the technical solution, the grayscale ultrasound image is segmented in S2 to identify the area of ​​the nodule, as follows:

[0016] use and The combination generates the loss function:

[0017] ;

[0018] in, is the loss function of the combination; is the weight parameter; is the total number of pixels in the image; For the Whether the pixel belongs to the nodule area; The probability value of the network output after the activation function; ;

[0019] The loss function is minimized to train the model, and the trained model is then used to segment the grayscale ultrasound image to obtain the nodule area.

[0020] As a further improvement of the technical solution, S3 is specifically as follows:

[0021] Extracting features from grayscale ultrasound images ;

[0022] In the feature map The Sobel operator is applied to perform edge detection. The Sobel operator calculates the gradient in the horizontal and vertical directions as:

[0023] ;

[0024] ;

[0025] in, is the gradient in the horizontal direction; is the gradient in the vertical direction;

[0026] ;

[0027] ;

[0028] in, Represents the convolution operation; is the convolution result of the gradient in the horizontal direction; is the convolution result of the gradient in the vertical direction; is a grayscale ultrasonic image;

[0029] Combine horizontal and vertical gradients to get edge strength:

[0030] ;

[0031] in, The feature map The edge strength of each pixel;

[0032] The results of traditional edge detection are combined with the features extracted by the deep learning model to generate the final fused edge map:

[0033] ;

[0034] in, is the fused edge map; is the weight parameter; For the original grayscale ultrasound image The edge map obtained by applying traditional edge detection methods.

[0035] As a further improvement of the present technical solution, the feature extraction unit extracts the echo of the nodule relative to the surrounding normal thyroid tissue in the grayscale ultrasound image, including classifying the echo type and calculating the grayscale difference between the nodule and the surrounding normal thyroid tissue, and evaluating the echo contrast.

[0036] As a further improvement of the technical solution, the echo type classification is specifically as follows:

[0037] If the pixel intensity of the nodule area is higher than that of the surrounding normal thyroid tissue, it is marked as hyperechoic; if the pixel intensity of the nodule area is lower than that of the surrounding normal thyroid tissue, it is marked as hypoechoic; if the pixel intensity of the nodule area is similar to that of the surrounding normal thyroid tissue, it is marked as isoechoic.

[0038] As a further improvement of the technical solution, the grayscale difference between the nodule and the surrounding normal thyroid tissue is calculated and the echo contrast is evaluated to achieve the following:

[0039] ;

[0040] in, is echo contrast; is the average gray value of all pixels in the nodule area; It is the average gray value of all pixels in the normal thyroid tissue around the nodule; is the standard deviation of the grayscale values ​​of all pixels in the nodule area; is the standard deviation of the grayscale values ​​of all pixels in the normal thyroid tissue surrounding the nodule.

[0041] As a further improvement of the technical solution, the weighted average and gray level co-occurrence matrix method is introduced into the echo contrast for optimization, and the optimization is as follows:

[0042] Calculate the weighted average gray value:

[0043] ;

[0044] in, is the weighted average gray value of the optimized nodule area; is a pixel; for the nodular area; Pixel The weight of Pixel Gray value of

[0045] ;

[0046] in, is the weighted average gray value of the optimized background area; is the background area;

[0047] Calculate the gray-level co-occurrence matrix:

[0048] ;

[0049] in, For the nodular region matrix; and is the gray value; for distance; For direction; is the pixel set of the nodule area; is the Kronecker delta function;

[0050] ;

[0051] in, For the background area matrix; is the pixel set of the background area;

[0052] according to The contrast, homogeneity, energy and correlation of Similarly, according to The contrast, homogeneity, energy and correlation of ;

[0053] Evaluate echo contrast:

[0054] ;

[0055] in, is the optimized echo contrast; is the weight coefficient; is the feature vector of the nodule area; is the feature vector of the background area.

[0056] As a further improvement of the technical solution, the blood flow condition in the nodule is obtained in the color Doppler ultrasound image, specifically as follows:

[0057] The blood flow density, blood flow velocity and blood flow pattern in the nodule are evaluated, and the blood flow density, blood flow velocity and blood flow pattern are fused into a comprehensive feature vector. The blood flow characteristics are determined according to the blood flow pattern. The properties of the nodule are obtained according to the blood flow characteristics and the comprehensive feature vector.

[0058] As a further improvement of the present technical solution, the blood flow patterns include central blood flow, marginal blood flow, mixed blood flow and no blood flow.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] In this AI-based early diagnosis and risk assessment system for thyroid tumors, weighted average and gray-level co-occurrence matrix methods are introduced to optimize the echo contrast. The gray-level co-occurrence matrix can capture the texture features in the image and calculate the nodule area and background area respectively. , and extracting useful texture features from them can further evaluate the echo contrast between thyroid nodules and surrounding normal tissues, thereby improving the accuracy and reliability of diagnosis; at the same time, the structural characteristics, echoes, and blood flow conditions of the nodules are taken into consideration together to improve the accuracy of thyroid tumor prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is the overall flow chart of the present invention;

[0062] The meaning of each number in the figure is:

[0063] 1. Data preparation unit; 2. Feature extraction unit; 3. Data analysis unit; 4. Clinical integration unit. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] See also Figure 1 As shown, a thyroid tumor early diagnosis and risk assessment system based on artificial intelligence is provided, including a data preparation unit 1, a feature extraction unit 2, a data analysis unit 3 and a clinical combination unit 4; wherein,

[0066] The data preparation unit 1 is used to obtain images related to the patient's thyroid nodules, and annotate, classify and save the obtained images in a standard format; wherein the images related to the patient's thyroid nodules include grayscale ultrasound images and color Doppler ultrasound images;

[0067] The feature extraction unit 2 is used to extract the nodule structural features and the echo of the nodule relative to the surrounding normal thyroid tissue in the grayscale ultrasonic image, and track the change trend of the nodule according to the extracted nodule structural features, and obtain the blood flow situation in the nodule in the color Doppler ultrasonic image;

[0068] The feature extraction unit 2 extracts the nodule structure features from the grayscale ultrasound image, as follows:

[0069] S1, preprocessing the grayscale ultrasonic image and removing noise interference in the image, the preprocessing at least includes grayscale normalization and size adjustment;

[0070] S2, segmenting the grayscale ultrasound image to identify the nodule area, and at the same time, performing manual correction based on the segmentation according to the actual situation;

[0071] S3. Evaluate the size of the nodule based on the segmented nodule area, identify the boundary of the nodule using edge detection, and evaluate the clarity of the boundary.

[0072] In S2, the grayscale ultrasound image is segmented to identify the nodule area, as follows:

[0073] use and The combination generates the loss function:

[0074] (Also known as The coefficient is a metric used to evaluate the similarity between two sets, especially useful when dealing with unbalanced datasets. In image segmentation tasks, it is used to measure the overlap between the predicted segmentation results and the true labels. The goal is to minimize the difference between the two sets so that the predicted results are as close to the true labels as possible;

[0075] (Cross Entropy Loss) is a method to measure the difference between two probability distributions. In binary classification problems, it is used to quantify the distance between the predicted probability distribution and the true probability distribution. The cross entropy loss function attempts to minimize the difference between the predicted value and the true label;

[0076] ;

[0077] in, is the loss function of the combination; is a weight parameter, ranging from 0 to 1, which determines The proportion of total losses is When work; when When kick in; is the total number of pixels in the image; For the Whether the pixel belongs to the nodule area, if it is a nodule area, then ; if not, then ; The probability value of the network output after the activation function; ;

[0078] At the same time, the accuracy and boundary clarity of the predicted area are optimized by adjusting The value of is used to balance the impact of the two loss functions;

[0079] The loss function is minimized to train the model, and the trained model is then used to segment the grayscale ultrasound image to obtain the nodule area.

[0080] S3 is as follows:

[0081] Extracting features from grayscale ultrasound images ;

[0082] In the feature map The Sobel operator is applied to perform edge detection. The Sobel operator calculates the gradient in the horizontal and vertical directions as:

[0083] ;

[0084] ;

[0085] in, is the gradient in the horizontal direction; is the gradient in the vertical direction;

[0086] ;

[0087] ;

[0088] in, Represents the convolution operation; is the convolution result of the gradient in the horizontal direction; is the convolution result of the gradient in the vertical direction; is a grayscale ultrasonic image;

[0089] Combine horizontal and vertical gradients to get edge strength:

[0090] ;

[0091] in, The feature map The edge strength of each pixel;

[0092] The results of traditional edge detection are combined with the features extracted by the deep learning model to generate the final fused edge map:

[0093] ;

[0094] in, It is a fused edge map, which is obtained by combining the results of traditional edge detection with the edge detection results on the feature map; is the weight parameter, which is a scalar between 0 and 1 and is used to control the relative importance of the traditional edge detection results and the edge detection results based on the feature map. , the optimal edge detection effect can be found to balance the contribution of traditional edge detection results and feature map-based edge detection results; For the original grayscale ultrasound image The edge map obtained by applying the traditional edge detection method, where the traditional edge detection method is the Sobel operator, is a binary image, where edge points are marked in white and non-edge points are marked in black;

[0095] Suppose we have a grayscale ultrasound image , we first extract its features through the pre-trained deep learning model Next, we Apply the Sobel operator to calculate the gradients in the horizontal and vertical directions respectively. and , and then merge the gradient values ​​in these two directions to get the edge strength on the feature map ;

[0096] Finally, we will Directly on the original image The traditional edge detection result obtained by applying the Sobel operator Combined to generate the final fused edge map , this fusion result is adjusted by To balance the contribution of the two methods, thus achieving better edge detection effect.

[0097] By accurately extracting the size, shape, boundary and internal structure of nodules, doctors can more accurately determine the nature of nodules and make correct diagnoses. Detailed nodule characteristics can help doctors develop personalized treatment plans based on the specific conditions of nodules. Regular extraction of nodule characteristics can track the changing trends of nodules and monitor the progression of the disease.

[0098] The feature extraction unit 2 extracts the echo of the nodule relative to the surrounding normal thyroid tissue in the grayscale ultrasonic image, including classifying the echo type and calculating the grayscale difference between the nodule and the surrounding normal thyroid tissue, and evaluating the echo contrast;

[0099] The echo type classification is as follows: if the pixel intensity of the nodule area is higher than that of the surrounding normal thyroid tissue, it is marked as hyperechoic; if the pixel intensity of the nodule area is lower than that of the surrounding normal thyroid tissue, it is marked as hypoechoic; if the pixel intensity of the nodule area is similar to that of the surrounding normal thyroid tissue, it is marked as isoechoic.

[0100] The gray value of a pixel is the pixel intensity, that is, the brightness or signal intensity of the point in the image. Specifically, each pixel in the nodule area is traversed and their gray values ​​are recorded. At the same time, the average pixel intensity, maximum pixel value, minimum pixel value and standard deviation of the pixel in the entire nodule area are calculated; the average pixel intensity can be obtained by summing the gray values ​​of all pixels and then dividing by the number of pixels. The maximum pixel intensity and the minimum pixel intensity are the maximum and minimum values ​​of the pixel gray values ​​in the nodule area. The standard deviation can be used to measure the consistency or degree of change of the gray value inside the nodule.

[0101] This is achieved by calculating the grayscale difference between the nodule and the surrounding normal thyroid tissue and evaluating the echo contrast as follows:

[0102] ;

[0103] in, is echo contrast; is the average gray value of all pixels in the nodule area; It is the average gray value of all pixels in the normal thyroid tissue around the nodule; is the standard deviation of the grayscale values ​​of all pixels in the nodule area; is the standard deviation of the grayscale values ​​of all pixels in the normal thyroid tissue surrounding the nodule;

[0104] This formula not only takes into account the average grayscale difference between the nodule and the background, but also takes into account the grayscale changes within the two regions, thus providing a more accurate echo contrast assessment. Such a method can effectively reflect the echo characteristics of the nodule relative to the surrounding tissue, which is of great significance for distinguishing benign and malignant nodules.

[0105] The weighted average and gray-level co-occurrence matrix methods are introduced into the echo contrast for optimization. The optimization results are as follows:

[0106] Calculate the weighted average gray value:

[0107] ;

[0108] in, is the weighted average gray value of the optimized nodule area; is a pixel; for the nodular area; Pixel The weight of Pixel Gray value of

[0109] ;

[0110] in, is the weighted average gray value of the optimized background area; is the background area;

[0111] Calculate the gray-level co-occurrence matrix:

[0112] ;

[0113] in, For the nodular region matrix; and is the gray value; for distance; For direction; is the pixel set of the nodule area; is the Kronecker delta function, which returns 1 if the condition is true, otherwise it returns 0;

[0114] The specific steps are: traverse the nodule area All pixels in , for each pixel , check its direction and distance Neighboring pixels at ;if and ,but Returns 1 if yes, otherwise returns 0.

[0115] All pixel pairs that meet the conditions The count is accumulated to middle;

[0116] ;

[0117] in, For the background area matrix; is the pixel set of the background area;

[0118] The specific steps are: traverse the nodule area All pixels in , for each pixel , check its direction and distance Neighboring pixels at ;if and ,but Returns 1 if yes, otherwise returns 0.

[0119] All pixel pairs that meet the conditions The count is accumulated to middle;

[0120] The gray-level co-occurrence matrix can capture the texture features in the image and calculate the gray-level co-occurrence matrix of the nodule area and the background area respectively. , and extract useful texture features from them, which can further evaluate the echo contrast between thyroid nodules and surrounding normal tissues and improve the accuracy and reliability of diagnosis;

[0121] according to The contrast, homogeneity, energy and correlation of Similarly, according to The contrast, homogeneity, energy and correlation of ;

[0122] Evaluate echo contrast:

[0123] ;

[0124] in, is the optimized echo contrast; is the weight coefficient, which is used to balance the influence of grayscale difference and texture features; is the feature vector of the nodule area; is the feature vector of the background area;

[0125] The evaluation of echogenicity helps to differentiate benign from malignant nodules. For example, hyperechoic nodules are generally considered benign, while hypoechoic or very hypoechoic nodules are more likely to be malignant.

[0126] The blood flow in the nodule is obtained in the color Doppler ultrasound image. Checking the blood flow in the nodule can reflect the blood supply status of the nodule. Active blood flow signals may indicate that the nodule has a higher metabolic activity, which is related to the benign or malignant nature of the nodule to a certain extent. Nodules with abundant blood flow may require more active treatment measures, while nodules with less blood flow may be more suitable for conservative observation. The principle is to detect the speed and direction of blood flow through the Doppler effect. Different colors represent different blood flow directions (such as red represents blood flow toward the probe, and blue represents blood flow away from the probe), as follows:

[0127] Evaluate the blood flow density, blood flow velocity and blood flow pattern in the nodule, and fuse the blood flow density, blood flow velocity and blood flow pattern into a comprehensive feature vector, determine the blood flow characteristics according to the blood flow pattern, and obtain the properties of the nodule according to the blood flow characteristics and the comprehensive feature vector;

[0128] By calculating the blood flow density, the richness of blood vessels in the nodule can be evaluated, thus helping to distinguish between benign and malignant tumors. By calculating the blood flow velocity, the speed of blood flow can be evaluated to further assist in diagnosis. By analyzing the blood flow pattern (central type, marginal type, mixed type, no blood flow), the distribution of blood vessels in the nodule can be understood in more detail, which helps to distinguish different types of tumors. Malignant tumors usually have higher blood flow density and higher blood flow velocity.

[0129] Blood flow patterns include central blood flow, marginal blood flow, mixed blood flow and no blood flow; central blood flow means that the blood flow is mainly concentrated in the center of the nodule; marginal blood flow means that the blood flow is mainly distributed at the edge of the nodule; mixed blood flow means that the blood flow appears both in the center and at the edge of the nodule; no blood flow means that there is no obvious blood flow signal in the nodule. According to the distribution pattern of blood flow signals (such as central type, marginal type, etc.), the blood supply characteristics of the nodule can be identified.

[0130] The data analysis unit 3 is used to input the extracted features into the statistical model, and train the model using a machine learning algorithm so that it can predict the nature of new nodules based on historical data, thereby judging the malignancy and benignity of the nodules; the structural features, echoes, and blood flow conditions of the nodules are considered together for judgment, which can improve the accuracy of thyroid tumor prediction;

[0131] The clinical combination unit 4 is used to combine the analysis results with the patient's clinical information for comprehensive judgment. If the possibility of malignancy is high, further examination is performed.

[0132] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. An artificial intelligence-based thyroid tumor early diagnosis and risk assessment system, characterized in that: include: A data preparation unit (1), the data preparation unit (1) is used to obtain images related to a patient's thyroid nodules, and annotate, classify and save the obtained images in a standard format; wherein the images related to the patient's thyroid nodules include grayscale ultrasound images and color Doppler ultrasound images; A feature extraction unit (2), the feature extraction unit (2) is used to extract nodule structural features and the echo contrast of the nodule relative to the surrounding normal thyroid tissue in the grayscale ultrasonic image, introduce weighted average and grayscale co-occurrence matrix methods into the echo contrast for optimization, and track the change trend of the nodule based on the extracted nodule structural features, and simultaneously obtain the blood flow situation in the nodule in the color Doppler ultrasonic image; A data analysis unit (3), the data analysis unit (3) is used to input the extracted features into a statistical model, and train the model using a machine learning algorithm so that the model can predict the properties of new nodules based on historical data, thereby determining the malignancy or benignity of the nodules; A clinical combination unit (4), the clinical combination unit (4) is used to combine the analysis results with the patient's clinical information to make a comprehensive judgment, and if the possibility of malignancy is high, further examination is performed; The feature extraction unit (2) extracts nodule structural features from the grayscale ultrasonic image as follows: S1, preprocessing the grayscale ultrasonic image and removing noise interference in the image, the preprocessing at least includes grayscale normalization and size adjustment; S2, segmenting the grayscale ultrasound image to identify the nodule area, and at the same time, performing manual correction based on the segmentation according to the actual situation; S3, evaluating the size of the nodule based on the segmented nodule area, identifying the boundary of the nodule using edge detection, and evaluating the clarity of the boundary; In S2, the grayscale ultrasound image is segmented to identify the region of the nodule, as follows: use and The combination generates the loss function: ； in, is the loss function of the combination; is the weight parameter; is the total number of pixels in the image; For the Whether the pixel belongs to the nodule area; The probability value of the network output after the activation function; ; The loss function is minimized to train the model, and then the trained model is used to segment the grayscale ultrasound image to obtain the nodule area; The S3 is as follows: Extracting features from grayscale ultrasound images ; In the feature map The Sobel operator is applied to perform edge detection. The Sobel operator calculates the gradient in the horizontal and vertical directions as: ； ； in, is the gradient in the horizontal direction; is the gradient in the vertical direction; ； ； in, Represents the convolution operation; is the convolution result of the gradient in the horizontal direction; is the convolution result of the gradient in the vertical direction; is a grayscale ultrasonic image; Combine horizontal and vertical gradients to get edge strength: ； in, The feature map The edge strength of each pixel; The results of traditional edge detection are combined with the features extracted by the deep learning model to generate the final fused edge map: ； in, is the fused edge map; is the weight parameter; For the original grayscale ultrasound image The edge map obtained by applying the traditional edge detection method; The grayscale difference between the nodule and the surrounding normal thyroid tissue is calculated and the echo contrast is evaluated as follows: ； in, is echo contrast; is the average gray value of all pixels in the nodule area; It is the average gray value of all pixels in the normal thyroid tissue around the nodule; is the standard deviation of the grayscale values ​​of all pixels in the nodule area; is the standard deviation of the grayscale values ​​of all pixels in the normal thyroid tissue surrounding the nodule; The weighted average and gray level co-occurrence matrix method is introduced into the echo contrast for optimization, and the optimization is as follows: Calculate the weighted average gray value: ； in, is the weighted average gray value of the optimized nodule area; is a pixel; for the nodular area; Pixel The weight of Pixel Gray value of ； in, is the weighted average gray value of the optimized background area; is the background area; Calculate the gray-level co-occurrence matrix: ； in, For the nodular area matrix; and is the gray value; for distance; For direction; is the pixel set of the nodule area; is the Kronecker delta function; ； in, For the background area matrix; is the pixel set of the background area; according to The contrast, homogeneity, energy and correlation of Similarly, according to The contrast, homogeneity, energy and correlation of ; Evaluate echo contrast: ； in, is the optimized echo contrast; is the weight coefficient; is the feature vector of the nodule area; is the feature vector of the background area.

2. The artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors according to claim 1, characterized in that: The feature extraction unit (2) extracts the echo of the nodule relative to the surrounding normal thyroid tissue in the grayscale ultrasonic image, including classifying the echo type and calculating the grayscale difference between the nodule and the surrounding normal thyroid tissue, and evaluating the echo contrast.

3. The artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors according to claim 2 is characterized in that: The echo type classification is specifically as follows: If the pixel intensity of the nodule area is higher than that of the surrounding normal thyroid tissue, it is marked as hyperechoic; if the pixel intensity of the nodule area is lower than that of the surrounding normal thyroid tissue, it is marked as hypoechoic; if the pixel intensity of the nodule area is similar to that of the surrounding normal thyroid tissue, it is marked as isoechoic.

4. The artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors according to claim 3 is characterized by: The blood flow condition in the nodule is obtained in the color Doppler ultrasound image, specifically as follows: The blood flow density, blood flow velocity and blood flow pattern in the nodule are evaluated, and the blood flow density, blood flow velocity and blood flow pattern are fused into a comprehensive feature vector. The blood flow characteristics are determined according to the blood flow pattern. The properties of the nodule are obtained according to the blood flow characteristics and the comprehensive feature vector.

5. The artificial intelligence-based early diagnosis and risk assessment system for thyroid tumors according to claim 4 is characterized in that: The blood flow patterns include central blood flow, marginal blood flow, mixed blood flow and no blood flow.

Images