Medical information management system based on big data analysis

Abstract

The invention provides a medical information management system based on big data analysis, and relates to the field of intelligent medical analysis, which is characterized in that detail local feature extraction and detail global feature extraction are carried out on a thyroid nodule region-of-interest image in an original thyroid ultrasound image of a user to obtain thyroid detail features; and judging whether the thyroid nodule is benign or malignant according to the thyroid detail features through a trained classifier. Therefore, the nodule features in the original thyroid ultrasound image are extracted in an automatic mode, and benign and malignant judgment is performed by using the classifier, so that the problems of subjectivity and time consumption in a traditional diagnosis method are effectively solved, and the accuracy and efficiency of thyroid nodule property diagnosis are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent medical analysis, and more specifically, to a medical information management system based on big data analysis. Background Art

[0002] In today's society, with the accelerated pace of life and the great changes in daily diet and living habits, the incidence of thyroid nodules is gradually increasing. Nowadays, people are faced with many factors that may induce thyroid nodules, such as staying up late for a long time, excessive mental stress, high-iodine or low-iodine diet, etc. This has led to an increasing number of patients with thyroid nodules, and has also brought greater challenges to clinical diagnosis and treatment.

[0003] However, the differential diagnosis of thyroid nodules has always been difficult. Due to the complexity and diversity of its symptoms and the lack of specific diagnostic indicators, misdiagnosis of its benign and malignant nature often occurs. This misdiagnosis not only brings unnecessary psychological burden to patients, but also is not conducive to accurate clinical treatment and may delay the best treatment time for patients.

[0004] Traditional diagnosis of thyroid nodules usually relies on the doctor's experience and subjective judgment. Doctors need to make preliminary judgments through palpation, auscultation, and asking about the patient's medical history and symptoms, and then conduct a comprehensive analysis based on the results of imaging examinations such as ultrasound and CT. However, this process may take a long time, especially when there are a large number of patients, and doctors often need to spend a lot of time to make a diagnosis. Moreover, doctors may become fatigued after working for a long time, which may affect the accuracy of the diagnosis. In addition, there are differences in experience and level among different doctors, which may also lead to inconsistent diagnostic results.

[0005] Therefore, a medical information management solution based on big data analysis is needed. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present application provides a medical information management system based on big data analysis.

[0007] According to one aspect of the present application, a medical information management system based on big data analysis is provided, comprising:

[0008] A thyroid ultrasound image acquisition module is used to obtain the original thyroid ultrasound image of the user to be analyzed;

[0009] A thyroid ultrasound image preprocessing module, used for performing image preprocessing on the original thyroid ultrasound image to obtain a preprocessed thyroid ultrasound image;

[0010] A thyroid nodule detection module, used for performing nodule region detection on the preprocessed thyroid ultrasound image to obtain an image of a thyroid nodule region of interest;

[0011] A thyroid nodule region encoding module, used for performing image feature encoding on the thyroid nodule region of interest image to obtain a thyroid nodule detail feature vector;

[0012] The thyroid property diagnosis result generation module is used to obtain the property diagnosis result of the user's thyroid nodule to be analyzed based on the thyroid nodule detail feature vector.

[0013] This application has significant technical effects due to the adoption of the above technical solutions:

[0014] The medical information management system based on big data analysis provided by the present application extracts detailed local features and detailed global features of the image of the region of interest of the thyroid nodule in the original thyroid ultrasound image of the user to obtain the thyroid detail features, and then uses the thyroid detail features through a trained classifier to determine whether the thyroid nodule is benign or malignant. In this way, the nodule features in the original thyroid ultrasound image are extracted in an automated manner, and the classifier is used to make a benign or malignant judgment, which effectively solves the subjective and time-consuming problems in the traditional diagnostic method and improves the accuracy and efficiency of the diagnosis of the nature of the thyroid nodule. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 This is a system block diagram of a medical information management system based on big data analysis according to an embodiment of the present application.

[0017] Figure 2 This is a data flow diagram of a medical information management system based on big data analysis according to an embodiment of the present application.

[0018] Figure 3 It is a block diagram of the thyroid nodule area coding module in the medical information management system based on big data analysis according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0020] As mentioned in the above background technology, with the changes in people's daily diet and living habits, the incidence of thyroid nodules has gradually increased. However, the differential diagnosis of thyroid nodules is more difficult, and misdiagnosis often occurs in the diagnosis of benign and malignant nodules, which is not conducive to accurate clinical treatment. Traditional diagnosis of thyroid nodules usually relies on the doctor's experience and subjective judgment. This process may take a long time, and the doctor may become fatigued after working for a long time, which may affect the accuracy of the diagnosis. Therefore, a medical information management solution based on big data analysis is expected, which can judge the shape of thyroid nodules by analyzing the user's original thyroid ultrasound images, and then assist doctors in making quick and accurate medical diagnoses.

[0021] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also demonstrated a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields. The development of deep learning and neural networks has provided new solutions and solutions for medical information management.

[0022] Figure 1 This is a system block diagram of a medical information management system based on big data analysis according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a medical information management system based on big data analysis according to an embodiment of the present application. Figure 1 and Figure 2 As shown, in the medical information management system 100 based on big data analysis, it includes: a thyroid ultrasound image acquisition module 110, which is used to obtain the original thyroid ultrasound image of the user to be analyzed; a thyroid ultrasound image preprocessing module 120, which is used to perform image preprocessing on the original thyroid ultrasound image to obtain a preprocessed thyroid ultrasound image; a thyroid nodule detection module 130, which is used to perform nodule area detection on the preprocessed thyroid ultrasound image to obtain a thyroid nodule region of interest image; a thyroid nodule region encoding module 140, which is used to perform image feature encoding on the thyroid nodule region of interest image to obtain a thyroid nodule detail feature vector; and a thyroid property diagnosis result generation module 150, which is used to obtain a property diagnosis result of the thyroid nodule of the user to be analyzed based on the thyroid nodule detail feature vector.

[0023] In the embodiment of the present application, the thyroid ultrasound image acquisition module 110 is used to obtain the original thyroid ultrasound image of the user to be analyzed. It should be understood that the original thyroid ultrasound image of the user shows the overall morphology of the thyroid gland, including whether it is enlarged, shrunk or irregular in shape. The presence of thyroid nodules, including their number, location, size, etc., can be clearly observed in the original ultrasound image. The image can also show whether there are calcification foci in the nodules, as well as the type of calcification foci (such as microcalcification, coarse calcification, peripheral calcification, etc.). The presence and type of calcification foci are of great reference value for judging the benign and malignant nature of nodules, especially microcalcification, which has a high risk of malignancy. In addition, the blood flow distribution in the thyroid gland and nodules can also be seen in the original ultrasound image. Malignant nodules often have rich and disordered blood flow, while benign nodules have relatively less and regular blood flow. In general, the rich information contained in the original thyroid ultrasound image provides an important basis for judging the benign and malignant nature of thyroid nodules. By comprehensively analyzing this information, the nature of thyroid nodules can be more accurately evaluated. In particular, in a specific embodiment of the present application, the original thyroid ultrasound image of the user to be analyzed can be obtained from the medical imaging database of the hospital.

[0024] In an embodiment of the present application, the thyroid ultrasound image preprocessing module 120 is used to perform image preprocessing on the original thyroid ultrasound image to obtain a preprocessed thyroid ultrasound image. Considering that the original thyroid ultrasound image is affected by many factors, such as equipment performance, patient position, respiratory movement, etc., resulting in different image quality, in order to improve the image quality and make the thyroid ultrasound image clearer and richer in details, it is necessary to perform image preprocessing on the original thyroid ultrasound image in the technical solution of the present application. The preprocessed thyroid image has higher quality and clearer details, which is more helpful in identifying key information such as the morphology, size, and boundaries of the nodules, thereby improving the accuracy of diagnosis of the nature of thyroid nodules.

[0025] In an embodiment of the present application, the thyroid nodule detection module 130 is used to perform nodule region detection on the preprocessed thyroid ultrasound image to obtain a thyroid nodule region of interest image. Specifically, in an embodiment of the present application, the thyroid nodule detection module is used to: pass the preprocessed thyroid ultrasound image through a nodule target detection network to obtain the thyroid nodule region of interest image. Considering that the thyroid ultrasound image may contain multiple structures, such as normal thyroid tissue, blood vessels, lymph nodes, etc. In order to provide accurate targets for subsequent analysis and diagnosis, it is necessary to accurately locate the area where the thyroid nodules are located. Therefore, in the technical solution of the present application, it is necessary to pass the preprocessed thyroid ultrasound image through a nodule target detection network. Through nodule target detection, the image region of interest containing nodules can be quickly screened out, avoiding a comprehensive analysis of the entire thyroid ultrasound image, thereby improving diagnostic efficiency. In particular, the nodule target detection network in the present application is a target detection network based on an anchor window. The target detection network uses an anchor window mechanism to preset multiple candidate boxes at different positions and scales of thyroid ultrasound images, thereby achieving comprehensive coverage of possible nodules in the image. This mechanism helps capture nodules of different sizes and shapes and improves the recall rate of nodule detection.

[0026] In the embodiment of the present application, the thyroid nodule region encoding module 140 is used to perform image feature encoding on the thyroid nodule region of interest image to obtain a thyroid nodule detail feature vector. Specifically, Figure 3 FIG. 1 is a block diagram of a thyroid nodule region coding module in a medical information management system based on big data analysis according to an embodiment of the present application. Figure 3As shown, the thyroid nodule region encoding module 140 includes: a thyroid detail local feature generating unit 141, used to pass the thyroid nodule region of interest image through a multi-scale thyroid nodule local detail feature encoder to obtain a thyroid nodule detail local feature map; a thyroid nodule detail local feature map pooling unit 142, used to perform maximum pooling on each feature matrix of the thyroid nodule detail local feature map along the channel dimension to obtain a thyroid nodule detail local feature vector; a thyroid nodule detail global feature generating unit 143, used to pass the thyroid nodule detail local feature map through a non-local neural network to obtain a thyroid nodule detail global feature map; The global feature map pooling unit 144 is used to perform maximum pooling on each feature matrix of the thyroid nodule detail global feature map along the channel dimension to obtain a thyroid nodule detail global feature vector; the thyroid nodule detail local feature optimization unit 145 is used to perform feature hierarchical precipitation expression optimization based on local fluctuation suppression on the thyroid nodule detail local feature vector based on the thyroid nodule detail global feature vector to obtain an optimized thyroid nodule detail local feature vector; the thyroid nodule detail feature generation unit 146 is used to fuse the optimized thyroid nodule detail local feature vector and the thyroid nodule detail global feature vector to obtain the thyroid nodule detail feature vector.

[0027] In order to more comprehensively extract the detail features of thyroid nodules, including but not limited to texture features that can reflect the complexity and uniformity of the internal structure of the nodules, and shape features that can reflect the boundaries and contours of the nodules, the thyroid nodule region of interest image can be passed through a multi-scale thyroid nodule local detail feature encoder to obtain a thyroid nodule detail local feature map. In the present application, the multi-scale thyroid nodule local detail feature encoder is a convolutional neural network model including a hybrid convolutional layer, wherein the hybrid convolutional layer includes a parallel first convolution branch structure, a second convolution branch structure, a third convolution branch structure, and a fourth convolution branch structure, and a multi-scale fusion structure connected to the first to fourth convolution branch structures, the first convolution branch uses a first convolution kernel having a first size, the second convolution branch uses a second convolution kernel having a first size and a first void rate, the third convolution branch uses a third convolution kernel having a first size and a second void rate, and the fourth convolution branch uses a fourth convolution kernel having a first size and a third void rate. It should be understandable that by setting convolution branch structures with different voiding rates in parallel, the features of nodules of different scales can be captured, which is conducive to processing nodules of different sizes and shapes that may exist in ultrasound images, and the use of convolution kernels with different voiding rates can increase the receptive field without increasing the amount of calculation, thereby capturing richer contextual information, which can better extract the detailed feature information of the nodules, and thus provide strong support for accurately judging the nature of the nodules.

[0028] Specifically, in an embodiment of the present application, a possible implementation method of passing the thyroid nodule region of interest image through a multi-scale thyroid nodule local detail feature encoder to obtain a thyroid nodule detail local feature map can be: using each mixed convolution layer of the multi-scale thyroid nodule local detail feature encoder to respectively perform on the input data in the forward pass of the layer: using a first convolution kernel with a first size to convolutionally encode the thyroid nodule region of interest image to obtain a first-scale thyroid nodule detail local feature map; using a second convolution kernel with a first dilation rate to convolutionally encode the thyroid nodule region of interest image to obtain a second-scale thyroid nodule detail local feature map; using a third convolution kernel with a second dilation rate to convolutionally encode the thyroid nodule region of interest image to obtain a third-scale thyroid nodule detail local feature map. The invention discloses a method for obtaining a local feature map of thyroid nodule details by using a fourth convolution kernel with a third dilation rate to convolutionally encode the thyroid nodule region of interest image to obtain a fourth-scale thyroid nodule detail local feature map; aggregating the first-scale thyroid nodule detail local feature map, the second-scale thyroid nodule detail local feature map, the third-scale thyroid nodule detail local feature map and the fourth-scale thyroid nodule detail local feature map along the channel dimension to obtain a multi-scale thyroid nodule detail local aggregated feature map; performing pooling processing on the multi-scale thyroid nodule detail local aggregated feature map to generate a pooled feature map; and performing activation processing on the pooled feature map to generate an activated feature map; wherein the output of the last layer of the multi-scale thyroid nodule local detail feature encoder is the thyroid nodule detail local feature map.

[0029] The original local feature map of thyroid nodule details usually contains a large amount of feature information, and the dimension of the feature map is high. High-dimensional data often easily leads to model overfitting. In order to reduce the dimension of the feature and reduce the risk of overfitting, while reducing the amount of calculation and increasing the speed of data processing, it is necessary to perform maximum pooling on each feature matrix of the local feature map of thyroid nodule details along the channel dimension to obtain the local feature vector of thyroid nodule details. Through maximum pooling, the most significant feature information in each feature matrix on the channel dimension of the feature map is retained, which helps to highlight the key features of the nodule, thereby improving the accuracy of the diagnosis of the nature of the nodule.

[0030] Considering that the multi-scale thyroid nodule local detail feature encoder mainly extracts features through local convolution operations, its receptive field is limited by the size and number of layers of the convolution kernel. In order to increase the expression of features in the feature map obtained by encoding, the thyroid nodule detail local feature map is passed through a non-local neural network in the technical solution of the present application to obtain a thyroid nodule detail global feature map. It should be known to those of ordinary skill in the art that non-local neural networks expand the capabilities of traditional neural networks by introducing non-local operations. Non-local operations calculate the response at one position as the weighted sum of features at all positions, thereby directly capturing long-range dependencies. Through non-local operations, non-local neural networks can provide a global information for the pixels at each position in the local feature map of the thyroid nodule details, and establish spatial connections between pixels. This global information helps to extract more discriminative features in the local feature map of the thyroid nodule details, and improve the model's ability to judge the nature of thyroid nodules.

[0031] Specifically, in an embodiment of the present application, a feasible implementation method of passing the thyroid nodule detail local feature map through a non-local neural network to obtain a thyroid nodule detail global feature map may be: further using the non-local neural network to encode the thyroid nodule detail local feature map in the following manner to obtain the global correlation feature map; wherein the encoding process includes: performing a first point convolution process, a second point convolution process, and a third point convolution process on the thyroid nodule detail local feature map respectively to obtain a first thyroid nodule detail feature map, a second thyroid nodule detail feature map, and a third thyroid nodule detail feature map; calculating the position-weighted sum between the first thyroid nodule detail feature map and the second thyroid nodule detail feature map to obtain a thyroid nodule detail fusion feature map; and converting the thyroid nodule detail fusion feature map into a global correlation feature map. The image is input into the Softmax function to map the feature values ​​of each position in the fusion feature map into the probability space to obtain a normalized thyroid nodule detail fusion feature map; the position-based point multiplication between the normalized thyroid nodule detail fusion feature map and the third thyroid nodule detail feature map is calculated to obtain a thyroid nodule detail re-fusion feature map; the thyroid nodule detail re-fusion feature map is embedded in the Gaussian similarity function to obtain a thyroid nodule detail global similarity feature map; the thyroid nodule detail global similarity feature map is subjected to a fourth point convolution process to adjust the number of channels of the global similarity feature map to obtain a thyroid nodule detail channel-adjusted global similarity feature map; and the position-based weighted sum of the thyroid nodule detail channel-adjusted global similarity feature map and the thyroid nodule detail local feature map is calculated to obtain the thyroid nodule detail global feature map.

[0032] Considering that the global feature map of thyroid nodule details contains rich global information, its dimension is usually high, which may lead to excessive calculation in the subsequent processing process and may introduce redundant information. Therefore, it is necessary to perform maximum pooling on each feature matrix of the global feature map of thyroid nodule details along the channel dimension to obtain the global feature vector of thyroid nodule details. Through the maximum pooling operation, the feature dimension can be effectively reduced, the amount of data can be reduced, and a more compact feature representation can be provided for subsequent feature data fusion. It is also worth mentioning that since the pooling operation has certain translation invariance and scale invariance, the global feature vector of thyroid nodule details obtained by pooling is more robust to small changes in input data (such as translation, rotation, etc.), which helps to enhance the generalization ability of the model and enable it to better adapt to data input in different situations.

[0033] In particular, considering that the multi-scale thyroid nodule local detail feature encoder and the non-local neural network use different strategies and weights in feature extraction, this leads to significant differences in the feature distribution of the generated thyroid nodule detail local feature vector and the thyroid nodule detail global feature vector. Due to the inconsistency in the fine-grained feature distribution between the thyroid nodule detail local feature vector and the thyroid nodule detail global feature vector, this will cause the fused feature vector to have abnormal stacking on the boundary of the feature manifold. Based on this, in the technical solution of the present application, before fusing the thyroid nodule detail local feature vector and the thyroid nodule detail global feature vector to obtain the thyroid nodule detail feature vector, the thyroid nodule detail local feature vector is optimized based on the feature hierarchical precipitation expression based on local fluctuation suppression based on the thyroid nodule detail global feature vector.

[0034] Specifically, in an embodiment of the present application, based on the global feature vector of the thyroid nodule details, the local feature vector of the thyroid nodule details is subjected to feature hierarchical precipitation expression optimization based on local fluctuation suppression to obtain an optimized local feature vector of the thyroid nodule details, including: calculating the positional difference between the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details to obtain a thyroid nodule detail shifted positional difference representation vector; after cascading the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details into a thyroid nodule detail joint feature vector, the thyroid nodule detail joint feature vector is input into a one-dimensional convolutional layer to obtain a thyroid nodule detail shifted convolutional coding representation vector; after linearly embedding and modulating the local feature vector of the thyroid nodule details using a first embedding matrix to obtain a thyroid nodule detail local feature linear modulation vector, the thyroid nodule detail is calculated. The position-based difference between the local feature linear modulation vector of the nodule detail and the global feature vector of the thyroid nodule detail is obtained to obtain a thyroid nodule detail shifted modulation differential representation vector; the thyroid nodule detail shifted position differential representation vector is used as the value vector, the thyroid nodule detail shifted convolutional coding representation vector and the thyroid nodule detail shifted modulation differential representation vector are used as the query vector and the key vector, and the thyroid nodule detail shifted position differential representation vector, the thyroid nodule detail shifted convolutional coding representation vector and the thyroid nodule detail shifted modulation differential representation vector are input into a shift corrector based on an emulated converter structure to obtain a thyroid nodule detail multi-scale hierarchical fusion shifted modulation weight vector; the position-based point multiplication between the thyroid nodule detail multi-scale hierarchical fusion shifted modulation weight vector and the thyroid nodule detail local feature vector is calculated to obtain the optimized thyroid nodule detail local feature vector.

[0035] In an embodiment of the present application, another expression method of performing feature hierarchical precipitation expression optimization based on local fluctuation suppression on the local feature vector of the thyroid nodule details based on the global feature vector of the thyroid nodule details to obtain an optimized local feature vector of the thyroid nodule details can be: processing the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details by the following formula to obtain an optimized local feature vector of the thyroid nodule details; wherein the formula is:

[0036]

[0037] V 2d =Conv 1D ([V1; V2])

[0038]

[0039] V1'=V1⊙V d

[0040] Wherein, V1 represents the local feature vector of the thyroid nodule details, V2 represents the global feature vector of the thyroid nodule details, Indicates point-by-point subtraction, V 1d represents the thyroid nodule detail shift according to the position difference representation vector, [;] represents the cascade operation, Conv1D(·) represents the one-dimensional convolutional coding, V 2d represents the thyroid nodule detail shift convolutional coding representation vector, W1 represents the first embedding matrix, b1 represents the first modulation bias vector, V 3d The thyroid nodule detail shift modulation differential representation vector, represents vector multiplication, Softmax represents the normalized exponential function, T represents the transposed vector of the feature vector, d represents the length of the feature vector, Relu represents the activation function, V d represents the multi-scale hierarchical fusion shift modulation weight vector of the thyroid nodule details, ⊙ represents the point multiplication by position, and V1' represents the optimized local feature vector of the thyroid nodule details.

[0041] It should be understood that in the technical solution of the present application, in the process of fusing the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details, a feature hierarchical precipitation expression optimization method based on local fluctuation suppression is proposed to address the problem of fine-grained inconsistency in feature distribution that arises in the process of fusing the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details, aiming to reduce the abnormal stacking phenomenon of the fused feature vector on the boundary of the feature manifold, so as to improve the robustness and classification performance of the model.

[0042] Specifically, the method first obtains a thyroid nodule detail shift position differential representation vector by calculating the element-by-element difference between the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details, thereby capturing the difference between the two feature vectors. Then, the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details are cascaded into a joint feature vector and input into a one-dimensional convolution layer to generate a thyroid nodule detail shift convolution coding representation vector. That is, the deep-level shift implicit information between the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details is captured by one-dimensional convolution coding. At the same time, the local feature vector of the thyroid nodule details is linearly embedded and modulated by using the first embedding matrix to be a thyroid nodule detail local feature linear modulation vector, and the element-by-element difference between the local feature linear modulation vector of the thyroid nodule details and the global feature vector of the thyroid nodule details is calculated to obtain a thyroid nodule detail shift modulation differential representation vector. The thyroid nodule detail shift modulation differential representation vector is used to represent the position expression of the shift information of the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details in the linear space. That is, in the technical solution of the present application, the thyroid nodule detail shift position differential representation vector, the thyroid nodule detail shift convolutional coding representation vector and the thyroid nodule detail shift modulated differential representation vector respectively represent the explicit expression of the fine-grained distribution difference shift information between the thyroid nodule detail local feature vector and the thyroid nodule detail global feature vector in different feature spaces.

[0043] Next, the thyroid nodule detail shifted position differential representation vector is used as the value vector, the thyroid nodule detail shifted convolutional coding representation vector and the thyroid nodule detail shifted modulation differential representation vector are used as the query vector and the key vector respectively, and these vectors are input into the shift corrector based on the simulated transformer structure to generate the thyroid nodule detail multi-scale hierarchical fusion shift modulation weight vector. In this way, the attention mechanism is used to evaluate the importance of features at different scales, that is, the attention mechanism is used to simulate the hierarchical precipitation expression of the fine-grained difference features between the thyroid nodule detail local feature vector and the thyroid nodule detail global feature vector in different feature spaces. Finally, the element-by-element product of the thyroid nodule detail multi-scale hierarchical fusion shift modulation weight vector and the thyroid nodule detail local feature vector is calculated to obtain the optimized thyroid nodule detail local feature vector, that is, the weighted operation is used to assign weights to each element in the thyroid nodule detail local feature vector, which reduces the influence of local disturbances and improves the quality of the feature vector.

[0044] In summary, in the method, the position differential representation vector of the thyroid nodule detail shift, the convolutional coding representation vector of the thyroid nodule detail shift, and the modulation differential representation vector of the thyroid nodule detail shift respectively represent the explicit expression of the fine-grained distribution difference shift information between the local feature vector of the thyroid nodule detail and the global feature vector of the thyroid nodule detail in different feature spaces. The generation of the multi-scale hierarchical fusion shift modulation weight vector of the thyroid nodule detail uses the attention mechanism to evaluate the importance of features at different scales and provide weight information for feature shift correction. Ultimately, the generation of the optimized local feature vector of thyroid nodule details reduces the impact of disturbances, improves the robustness of the model to local disturbances in the input data, enhances the model's adaptability to noisy or variant data sets, improves the fusion quality of the feature vector, and improves the overall classification performance and training efficiency of the model.

[0045] The optimized local feature vector of thyroid nodule details focuses on the local details of the thyroid nodule, such as edge contours, texture patterns, local morphological changes, etc. The global feature vector of thyroid nodule details provides the relationship between the data of different positions in the nodule area. They have different focuses and advantages in describing thyroid nodules. In order to integrate the information expressed in these two feature vectors, the optimized local feature vector of thyroid nodule details and the global feature vector of thyroid nodule details are fused in the technical solution of the present application to obtain the thyroid nodule detail feature vector. By fusing the optimized local feature vector of thyroid nodules and the global feature vector of thyroid details, a feature vector containing richer information can be generated. This comprehensive feature description helps to improve the model's ability to discriminate the nature of thyroid nodules, enabling the model to more accurately distinguish between benign and malignant nodules.

[0046] In an embodiment of the present application, the thyroid property diagnosis result generation module 150 is used to obtain the property diagnosis result of the thyroid nodule of the user to be analyzed based on the thyroid nodule detail feature vector. Specifically, in an embodiment of the present application, the thyroid property diagnosis result generation module is used to: pass the thyroid nodule detail feature vector through a thyroid nodule property diagnostic device based on a classifier to obtain a classification result, and the classification result is used to indicate whether the thyroid nodule of the user to be analyzed is benign or malignant. It should be understood that the classifier, as a machine learning model, can analyze and judge according to the input data and map it to different categories. The classifier is usually trained using a large amount of labeled data, and can predict the benign and malignant nature of the thyroid nodule based on the detailed features of the thyroid nodule learned during the training process. Doctors can make a more accurate assessment of the nature of the nodule based on the judgment results of the classifier, combined with their own clinical experience and professional knowledge. This diagnostic method is conducive to improving the efficiency and accuracy of the diagnosis of the nature of the thyroid nodule.

[0047] Specifically, in an embodiment of the present application, the thyroid nodule detail feature vector is passed through a thyroid nodule property diagnostic device based on a classifier to obtain a classification result, and the classification result is used to indicate whether the thyroid nodule of the user to be analyzed is benign or malignant. One possible implementation method may be: using the fully connected layer of the classifier to perform fully connected encoding on the thyroid nodule detail feature vector to obtain a thyroid nodule detail fully connected encoded feature vector; inputting the thyroid nodule detail fully connected encoded feature vector into the Softmax classification function of the classifier to obtain probability values ​​belonging to each classification label in the thyroid nodule detail feature vector, the classification labels including those used to indicate that the thyroid nodule of the user to be analyzed is benign and those used to indicate that the thyroid nodule of the user to be analyzed is malignant; and determining the classification label corresponding to the largest of the probability values ​​as the classification result.

[0048] In summary, the medical information management system 100 based on big data analysis based on the embodiment of the present application is explained, which extracts detailed local features and detailed global features from the image of the region of interest of the thyroid nodule in the original thyroid ultrasound image of the user to obtain thyroid detail features, and then uses the thyroid detail features through a trained classifier to determine whether the thyroid nodule is benign or malignant. In this way, the nodule features in the original thyroid ultrasound image are extracted in an automated manner, and the classifier is used to make a benign or malignant judgment, which effectively solves the subjective and time-consuming problems in the traditional diagnostic method, and improves the accuracy and efficiency of the diagnosis of the nature of the thyroid nodule.

[0049] As described above, the medical information management system 100 based on big data analysis according to the embodiment of the present application can be implemented in various terminal devices, such as a server for medical information management based on big data analysis. In one example, the medical information management system 100 based on big data analysis according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the medical information management system 100 based on big data analysis can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the medical information management system 100 based on big data analysis can also be one of the many hardware modules of the terminal device.

[0050] Alternatively, in another example, the medical information management system 100 based on big data analysis and the terminal device may also be separate devices, and the medical information management system 100 based on big data analysis may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

Claims

1. A medical information management system based on big data analysis, characterized in that: include: A thyroid ultrasound image acquisition module is used to obtain the original thyroid ultrasound image of the user to be analyzed; A thyroid ultrasound image preprocessing module, used for performing image preprocessing on the original thyroid ultrasound image to obtain a preprocessed thyroid ultrasound image; A thyroid nodule detection module, used for performing nodule region detection on the preprocessed thyroid ultrasound image to obtain an image of a thyroid nodule region of interest; A thyroid nodule region encoding module, used for performing image feature encoding on the thyroid nodule region of interest image to obtain a thyroid nodule detail feature vector; The thyroid property diagnosis result generation module is used to obtain the property diagnosis result of the user's thyroid nodule to be analyzed based on the thyroid nodule detail feature vector.

2. The medical information management system based on big data analysis according to claim 1, characterized in that: The thyroid nodule detection module is used to: pass the preprocessed thyroid ultrasound image through a nodule target detection network to obtain the thyroid nodule region of interest image.

3. The medical information management system based on big data analysis according to claim 2 is characterized in that: The thyroid nodule region coding module comprises: A thyroid detail local feature generating unit, used for passing the thyroid nodule region of interest image through a multi-scale thyroid nodule local detail feature encoder to obtain a thyroid nodule detail local feature map; A thyroid nodule detail local feature map pooling unit, used for performing maximum pooling on each feature matrix of the thyroid nodule detail local feature map along the channel dimension to obtain a thyroid nodule detail local feature vector; A thyroid nodule detail global feature generation unit, used for obtaining a thyroid nodule detail global feature map by passing the thyroid nodule detail local feature map through a non-local neural network; A thyroid nodule detail global feature map pooling unit, used for performing maximum pooling on each feature matrix of the thyroid nodule detail global feature map along the channel dimension to obtain a thyroid nodule detail global feature vector; A thyroid nodule detail local feature optimization unit, used for performing feature hierarchical precipitation expression optimization based on local fluctuation suppression on the thyroid nodule detail local feature vector based on the thyroid nodule detail global feature vector to obtain an optimized thyroid nodule detail local feature vector; The thyroid nodule detail feature generating unit is used to fuse the optimized thyroid nodule detail local feature vector and the thyroid nodule detail global feature vector to obtain the thyroid nodule detail feature vector.

4. The medical information management system based on big data analysis according to claim 3 is characterized in that: The nodule target detection network is an anchor window-based target detection network, and the multi-scale thyroid nodule local detail feature encoder is a convolutional neural network model including a hybrid convolutional layer, wherein the hybrid convolutional layer includes a parallel first convolution branch structure, a second convolution branch structure, a third convolution branch structure and a fourth convolution branch structure, and a multi-scale fusion structure connected to the first to fourth convolution branch structures, the first convolution branch uses a first convolution kernel with a first size, the second convolution branch uses a second convolution kernel with a first size and a first dilation rate, the third convolution branch uses a third convolution kernel with a first size and a second dilation rate, and the fourth convolution branch uses a fourth convolution kernel with a first size and a third dilation rate.

5. The medical information management system based on big data analysis according to claim 4 is characterized in that: The thyroid nodule detail local feature optimization unit is used for: Calculating the position difference between the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details to obtain a position difference representation vector of the thyroid nodule detail displacement; After cascading the local feature vector of the thyroid nodule details and the global feature vector of the thyroid nodule details into a joint feature vector of the thyroid nodule details, the joint feature vector of the thyroid nodule details is input into a one-dimensional convolutional layer to obtain a shifted convolutional coding representation vector of the thyroid nodule details; After performing linear embedding modulation encoding on the local feature vector of the thyroid nodule details into a local feature linear modulation vector of the thyroid nodule details using a first embedding matrix, a positional difference between the local feature linear modulation vector of the thyroid nodule details and the global feature vector of the thyroid nodule details is calculated to obtain a thyroid nodule detail shift modulation difference representation vector; Taking the thyroid nodule detail shifted position differential representation vector as the value vector, the thyroid nodule detail shifted convolutional coding representation vector and the thyroid nodule detail shifted modulation differential representation vector as the query vector and key vector, the thyroid nodule detail shifted position differential representation vector, the thyroid nodule detail shifted convolutional coding representation vector and the thyroid nodule detail shifted modulation differential representation vector are input into a shift corrector based on a pseudo-converter structure to obtain a thyroid nodule detail multi-scale hierarchical fusion shift modulation weight vector; The optimized local feature vector of thyroid nodule details is obtained by calculating the point-by-point multiplication of the multi-scale hierarchical fusion shift modulation weight vector of the thyroid nodule details and the local feature vector of the thyroid nodule details.

6. The medical information management system based on big data analysis according to claim 5, characterized in that: The thyroid property diagnosis result generation module is used to: pass the thyroid nodule detail feature vector through a classifier-based thyroid nodule property diagnostic device to obtain a classification result, and the classification result is used to indicate whether the thyroid nodule of the user to be analyzed is benign or malignant.

7. The medical information management system based on big data analysis according to claim 6, characterized in that: The thyroid property diagnosis result generating module is used for: Using the fully connected layer of the classifier to perform fully connected encoding on the thyroid nodule detail feature vector to obtain a thyroid nodule detail fully connected encoded feature vector; Inputting the thyroid nodule detail fully connected encoded feature vector into the Softmax classification function of the classifier to obtain probability values ​​belonging to various classification labels in the thyroid nodule detail feature vector, wherein the classification labels include a value for indicating that the thyroid nodule of the user to be analyzed is benign and a value for indicating that the thyroid nodule of the user to be analyzed is malignant; The classification label corresponding to the largest probability value among the probability values ​​is determined as the classification result.

Images