Window adjustment parameter prediction method and device, equipment and storage medium
By obtaining the multi-dimensional feature vector of medical images and using prediction neural network to extract coding features, the problem of low subjectivity and accuracy of window adjustment parameter prediction in the prior art is solved, and more efficient and accurate window adjustment parameter prediction is achieved.
Patent Information
- Application Number
- CN202510003074.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-02
- Publication Date
- 2025-05-13
AI Technical Summary
The prior art has problems such as strong subjectivity, cumbersome operation, high time consumption and poor diagnosis consistency in the prediction of window adjustment parameters. The prediction accuracy of the automatic window adjustment method is low, especially when processing different types of medical images.
By obtaining the grayscale value feature vector and sequence category feature vector of the target medical image, the grayscale value encoding feature and image information encoding feature are extracted using the prediction neural network, and combining the image statistics vector and sequence category feature vector to obtain the predicted window adjustment parameters.
It improves the prediction accuracy and prediction speed of window adjustment parameters, can handle different types of medical images more accurately, and enhances the efficiency and consistency of diagnosis.
Smart Images
Figure CN119991577A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of window adjustment parameter prediction, and in particular to a method, device, equipment and storage medium for predicting window adjustment parameters. Background Art
[0002] With the rapid development of medical imaging technology, magnetic resonance imaging (MRI) has been widely used in many clinical fields such as the nervous, musculoskeletal and cardiovascular systems due to its non-invasive, high-resolution and diverse contrast imaging capabilities. However, the complexity and diversity of MRI images pose major challenges to image interpretation and analysis, including slight differences in signal intensity between different tissues, artifacts and noise interference, and multiple imaging sequences that may lead to information redundancy and difficulty in selection. In order to improve the accuracy and efficiency of diagnosis, windowing technology has emerged to optimize the contrast and brightness of the image by adjusting the window width (WW) and window level (WL).
[0003] Currently, the window width and window position can be adjusted manually or predicted by the automatic window adjustment method. However, the manual window adjustment method has problems such as strong subjectivity, cumbersome operation, high time consumption, and poor diagnostic consistency. The automatic window adjustment method only uses a single mapping relationship to predict the window width and window position, and performs poorly when processing different types of medical images, resulting in low prediction accuracy.
[0004] Therefore, how to improve the prediction accuracy and prediction speed of the window adjustment parameters is a problem that technical personnel in this field need to solve. Summary of the invention
[0005] The present application provides a method, device, equipment and storage medium for predicting window adjustment parameters to improve the prediction accuracy and prediction speed of the window adjustment parameters.
[0006] In a first aspect, the present application provides a method for predicting window adjustment parameters, comprising:
[0007] Obtaining gray value feature vectors and sequence category feature vectors of target medical images;
[0008] Determine an image statistics vector using the gray value feature vector;
[0009] The grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector are extracted through a predictive neural network, and the predicted window adjustment parameters are obtained using the grayscale value coding features and the image information coding features; wherein the image information feature vector includes the image statistics vector and the sequence category feature vector.
[0010] Optionally, the step of obtaining the gray value feature vector and the sequence category feature vector of the target medical image includes:
[0011] Obtaining a gray value feature vector and a sequence category feature vector corresponding to a target type sequence in a target medical image; wherein the target type sequence is any one of a T1 sequence, a T2 sequence, and a W / F sequence.
[0012] Optionally, determining the image statistical vector by using the gray value feature vector includes:
[0013] Determine statistical information of the grayscale value feature vector; the statistical information includes at least one of the grayscale maximum value, grayscale minimum value, grayscale median value, grayscale mean value, grayscale variance, and grayscale percentile of the grayscale value feature vector;
[0014] An image statistics vector is generated according to the statistical information.
[0015] Optionally, before obtaining the gray value feature vector and the sequence category feature vector of the target medical image, the method further includes:
[0016] Acquire training data; wherein the training data includes: a sample gray value feature vector and a sample image information feature vector;
[0017] The initial prediction neural network is trained using the training data and the loss function to obtain the trained prediction neural network.
[0018] Optionally, the training of the initial prediction neural network using the training data and the loss function includes:
[0019] Inputting the training data into the initial prediction neural network to obtain prediction window adjustment parameters;
[0020] Calculate the loss value through the loss function, the sample window adjustment parameter and the predicted window adjustment parameter;
[0021] The loss value is used to update the weight parameters of the initial prediction neural network.
[0022] Optionally, extracting the grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector by predicting a neural network includes:
[0023] Extracting a grayscale value encoding feature of the grayscale value feature vector by predicting an image feature extraction neural network of the neural network;
[0024] Extracting a neural network by predicting information features of a neural network to extract image information coding features of the image information feature vector;
[0025] Among them, the image feature extraction neural network is a deep learning feature extraction network; the information feature extraction neural network is a shallow neural network.
[0026] Optionally, obtaining predicted window adjustment parameters using the gray value encoding feature and the image information encoding feature includes:
[0027] Fusing the grayscale value coding feature and the image information coding feature to obtain an image fusion coding feature;
[0028] The image fusion coding features are input into a fully connected layer to obtain predicted window adjustment parameters; the window adjustment parameters include a window width value and a window level value; wherein the fully connected layer is a fully connected layer of the prediction neural network.
[0029] In a second aspect, the present application provides a device for predicting window adjustment parameters, comprising:
[0030] A first acquisition module is used to acquire a gray value feature vector and a sequence category feature vector of a target medical image;
[0031] A determination module, used for determining an image statistical vector using the gray value feature vector;
[0032] A prediction module is used to extract the grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector through a prediction neural network, and obtain predicted window adjustment parameters using the grayscale value coding features and the image information coding features; wherein the image information feature vector includes the image statistics vector and the sequence category feature vector.
[0033] In a third aspect, the present application provides an electronic device, including:
[0034] A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the above-mentioned prediction method of the present application through the computer program.
[0035] In a fourth aspect, the present application further provides a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are used to execute the steps of the above-mentioned prediction method of the present application.
[0036] The above technical solution provided by the embodiment of the present application has the following advantages over the prior art: The embodiment of the present application discloses a prediction scheme for window adjustment parameters. This scheme needs to determine the gray value feature vector and sequence category feature vector of the target medical image, and determine the image statistical vector; extract the gray value encoding features of the gray value feature vector and the image information encoding features of the image information feature vector through the prediction neural network, and obtain the predicted window adjustment parameters using the gray value encoding features and the image information encoding features. It can be seen that when predicting the window adjustment parameters, the present application obtains the multi-dimensional feature vectors of the medical image, such as: gray value feature vector, image statistical vector and sequence category feature vector. The prediction neural network can accurately predict the window adjustment parameters by extracting the features of the multi-dimensional feature vector; and, through the sequence category feature vector, the present application allows the prediction neural network to distinguish the differences in window adjustment parameters of different sequences, thereby more targeted prediction of the window adjustment parameters, and improve the prediction accuracy and prediction speed of the window adjustment parameters. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0039] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.
[0040] Figure 1 A schematic diagram of a method for predicting window adjustment parameters provided in an embodiment of the present application;
[0041] Figure 2 A schematic diagram of another method for predicting window adjustment parameters provided in an embodiment of the present application;
[0042] Figure 3 A schematic diagram of a method for predicting window adjustment parameters provided in an embodiment of the present application;
[0043] Figure 4 A schematic diagram of the structure of a prediction neural network provided in an embodiment of the present application;
[0044] Figure 5A schematic diagram of the structure of a device for predicting window adjustment parameters provided in an embodiment of the present application;
[0045] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0046] At present, the grayscale level of MRI images usually reaches several thousand grayscale levels, far exceeding the grayscale range of 256 levels of ordinary monitors. This difference means that directly displaying the original image may result in loss of details and insufficient contrast. Therefore, in order to ensure that doctors can accurately identify the subtle differences between different tissues and lesions, it is usually necessary to view MRI images through window adjustment technology. Window adjustment technology can dynamically adjust the displayed grayscale range and center value according to the specific situation, thereby highlighting the structural details of interest and effectively improving the accuracy of diagnosis. Among them, the window width WW defines the range of displayed grayscale levels, and the window position WL defines the center grayscale value of the window width range.
[0047] In the related art, the window width and window position adjustment methods are mostly based on the gray value data of the medical image, the histogram distribution of the gray value of the medical image, the maximum and minimum values of the gray value of the medical image, etc. to give an initial window width and window position, and then the original window width and window position are manually adjusted by the reader according to the subjective feeling. However, this method of automatically predicting the window width and window position often adopts a single mapping relationship, for example: predicting the window width and window position by the gray data of the image, or predicting the window width and window position by the maximum and minimum values of the gray data of the image. This method may not perform well when processing medical images of different types and under different conditions. Although the manual window adjustment method can improve the readability of the image to a certain extent, it has the problems of strong subjectivity, cumbersome operation, large time consumption, poor diagnostic consistency, etc., and it is difficult to meet the needs of complex clinical environments for efficient and accurate diagnosis. Especially when facing a large amount of image data, manual window adjustment not only increases the workload of doctors, but also may cause some subtle lesions to be ignored.
[0048] Research has found that due to the diversity and complexity of MRI medical images in different parts and different scanning sequences, the model for predicting window width and window position needs to have stronger generalization ability. The single mapping model has the problem of low prediction accuracy when processing medical images of different types of sequences due to the lack of image data of different sequences. In addition, the automatic window width and window position prediction model should not only simply process the image pixel values, but also have a deep understanding of the pathological changes in the image. However, the relevant technology lacks effective recognition and utilization of these features, resulting in the predicted window width and window position failing to meet the needs of clinical diagnosis.
[0049] Therefore, in an embodiment of the present application, a method, device, equipment and storage medium for predicting window adjustment parameters are disclosed to improve the prediction accuracy and prediction speed of the window adjustment parameters.
[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0051] The disclosure below provides many different embodiments or examples to implement different structures of the present invention. In order to simplify the disclosure of the present invention, the parts and settings of specific examples are described below. Of course, they are only examples, and the purpose is not to limit the present invention. In addition, the present invention can repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.
[0052] See also Figure 1 , Figure 1 A flowchart of a method for predicting window adjustment parameters provided in an embodiment of the present application is provided, and the method comprises the following steps:
[0053] S101, obtaining a gray value feature vector and a sequence category feature vector of a target medical image;
[0054] In the present application, the target medical image is the initial multi-sequence medical image data collected by an imaging system such as magnetic resonance imaging (MRI). Due to the different sites and conditions for collecting and scanning, there may be differences in contrast, brightness, etc. of the image; at the same time, due to the different scanning sequences and parameters used, there will also be differences in the image. Therefore, in order to observe the image more accurately and effectively, the present application extracts the multi-dimensional information of the target medical image to predict the window adjustment parameters, so as to adjust the medical image according to the predicted window adjustment parameters.
[0055] In this application, in order to predict the window adjustment parameters of the target medical image, it is first necessary to read the gray value data and image sequence information of the target medical image, output the gray value data as a gray value feature vector, and output the image sequence information as a sequence category feature vector. The gray value data can reflect the anatomical structure difference of the medical image, and the sequence category feature vector is used to represent the sequence type of the target medical image.
[0056] The target medical image sequence types include: T1 type, T2 type and W / F (fat suppression) type, etc. T1 and T2 are time-related parameters of magnetic resonance signal intensity. When obtaining the gray value feature vector of a certain type of sequence, the image gray value data can be read from the dcm folder of different types of sequence data and saved as a gray value feature vector. The dcm format is a common format in medical images. The dcm files in the dcm folder are usually used to store MRI, CT and other image data. The present application obtains a sequence category feature vector according to the different storage locations of the medical image data. The sequence category feature vector is a one-hot feature vector after label smoothing technology (Label Smoothing). For example, the sequence category feature vector of T1 is [0.9333, 0.3333, 0.3333], and the sequence category feature vector of T2 is [0.3333, 0.9333, 0.3333], and the sequence category feature vector of W / F is [0.3333, 0.3333, 0.9333].
[0057] S102, determining an image statistical vector using a gray value feature vector;
[0058] In this application, after obtaining the gray value feature vector, it is necessary to use the gray value feature vector to obtain the statistical feature information of the medical image, and output the statistical feature information as an image statistical vector. The statistical feature information is the statistical data of the gray value feature vector, and the statistical feature information includes: gray maximum value, gray minimum value, gray median value, etc., which are not specifically limited here and can be set according to actual needs.
[0059] S103, extracting the grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector through a predictive neural network, and obtaining predicted window adjustment parameters using the grayscale value coding features and the image information coding features; wherein the image information feature vector includes an image statistics vector and a sequence category feature vector.
[0060] In the present application, after obtaining the gray value feature vector, sequence category feature vector and image statistical vector, they need to be input into the prediction neural network to extract coding features. Before inputting each vector into the prediction neural network, the present application first needs to merge the image statistical vector and the sequence category feature vector to generate an image information feature vector. When merging the image statistical vector and the sequence category feature vector, there is no need to distinguish the order, but it is necessary to ensure that the size of the image information feature vector is uniform; wherein, the uniform size here means: it is necessary to merge the first predetermined number of elements in the image statistical vector with the second predetermined number of elements in the sequence category feature vector, and the first predetermined number and the second predetermined number are pre-set fixed values, so that the number of elements generated each time the image information feature vector is a fixed value, ensuring the uniform size of the image information feature vector, which is convenient for predictive neural network processing.
[0061] Among them, when training the prediction neural network, the sample medical image needs to be processed accordingly through steps S101 and S102 to obtain the sample gray value feature vector and the sample image information feature vector, and the sample gray value feature vector and the sample image information feature vector are input into the initial prediction neural network for training to obtain the output feature vector, which is the prediction window adjustment parameter. The prediction window adjustment parameter is calculated through the loss function to update the weight parameters of the initial prediction neural network to obtain a prediction neural network that meets the requirements; wherein the sample medical image is an image used to train the prediction neural network.
[0062] After obtaining the grayscale value feature vector and the image information feature vector, the present application needs to input the grayscale value feature vector and the image information feature vector into the trained prediction neural network, which extracts the grayscale value coding features from the grayscale value feature vector, extracts the image information coding features from the image information feature vector, and uses the grayscale value coding features and the image information coding features to obtain predicted window adjustment parameters, which include window width values and window position values.
[0063] From the above, it can be seen that when predicting the window adjustment parameters, the present application obtains multi-dimensional feature vectors of medical images, such as: grayscale value feature vectors, image statistical vectors and sequence category feature vectors. The grayscale value data in the grayscale value feature vector can reflect the anatomical structure differences of the medical images, the image statistical vector can reflect the statistical information of the grayscale value data of the medical images, and the sequence category feature vector is used to represent the sequence type of the target medical image. Therefore, the prediction neural network in the present application realizes accurate prediction of the window adjustment parameters by performing feature extraction in the image dimension, statistical dimension and category difference dimension; and, through the sequence category feature vector, the present application allows the prediction neural network to distinguish the differences in window adjustment parameters of different sequences, thereby predicting the window adjustment parameters more specifically and improving the prediction accuracy and prediction speed of the window adjustment parameters.
[0064] See also Figure 2 , Figure 2 A flowchart of another method for predicting window adjustment parameters provided in an embodiment of the present application is provided, and the method comprises the following steps:
[0065] S201, obtaining a gray value feature vector and a sequence category feature vector corresponding to a target type sequence in a target medical image; the target type sequence is any one of a T1 sequence, a T2 sequence, and a W / F sequence;
[0066] S202, determining statistical information of the grayscale value feature vector; the statistical information includes at least one of the grayscale maximum value, grayscale minimum value, grayscale median value, grayscale mean value, grayscale variance, and grayscale percentile of the grayscale value feature vector;
[0067] S203, generating an image statistical vector according to the statistical information;
[0068] S204, extracting the grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector through a predictive neural network, and obtaining predicted window adjustment parameters using the grayscale value coding features and the image information coding features; wherein the image information feature vector includes an image statistics vector and a sequence category feature vector.
[0069] In the present application, window adjustment prediction can be performed on target medical images of different types of sequences. Therefore, the grayscale value feature vector and sequence category feature vector obtained in the present application are specifically the grayscale value feature vector and sequence category feature vector corresponding to a certain target type sequence, and the target type sequence is any one of the T1 sequence, T2 sequence and W / F sequence.
[0070] Therefore, when the present application obtains the grayscale value feature vector and sequence category feature vector corresponding to the target type sequence, it specifically reads the image grayscale value data and medical image sequence information of different types of sequences from the dcm folders of different types of sequences, and then obtains the grayscale value feature vector and sequence category feature vector of different types of sequences. For example: if the target type sequence is a T1 sequence, it is necessary to obtain the grayscale value feature vector of the T1 sequence, the sequence category feature vector of the T1 sequence, and the image statistics vector of the T1 sequence, and finally the window adjustment parameters of the T1 sequence are predicted by the prediction neural network.
[0071] The statistical information of the grayscale value feature vector determined in the present application specifically includes at least one of the grayscale maximum value, grayscale minimum value, grayscale median value, grayscale mean value, grayscale variance, and grayscale percentile of the grayscale value feature vector, and then generates a corresponding image statistical vector based on the statistical information. The image statistical vector is a one-dimensional vector. For example, the image statistical vector can be: [grayscale maximum value, minimum value, median value, mean value, variance, 95% quantile, 5% quantile].
[0072] After obtaining the image statistics vector ([grey value maximum value, minimum value, median value, mean value, variance, 95% quantile, 5% quantile]) and the sequence category feature vector ([0.9333, 0.3333, 0.3333]), the image statistics vector and the sequence category feature vector can be merged to generate the image information feature vector. After inputting the gray value feature vector and the image information feature vector into the trained prediction neural network, the predicted window adjustment parameters are obtained.
[0073] In summary, the present application not only reads the gray value data of medical images in a certain type of sequence, but also obtains the image sequence type information and the statistical information of the image gray value, so that the present solution can distinguish the differences in window width and window position of different MRI sequences and perform targeted window width and window position adjustment. The present application obtains multi-dimensional information of medical image data, which includes the image dimension corresponding to the gray value feature vector, the statistical dimension corresponding to the image statistical vector, and the sequence category difference dimension corresponding to the sequence category feature vector. After feature extraction of the multi-dimensional information, more accurate window width and window position adjustment results can be obtained.
[0074] Based on any of the above method embodiments, in another embodiment of the present application, before obtaining the gray value feature vector and sequence category feature vector of the target medical image, the present application also includes: obtaining training data; the training data includes: sample gray value feature vector and sample image information feature vector; training the initial prediction neural network through the training data and the loss function to obtain the trained prediction neural network.
[0075] In the present application, the prediction neural network is generated by training sample medical images. First, it is necessary to obtain the sample gray value feature vector and the sample image information feature vector of the sample medical image, and input them into the initial prediction neural network, and obtain the trained prediction neural network based on the output feature vector and loss function. Among them, the present application can collect a large number of sample medical images of different sequences, and train the initial prediction neural network through more sample medical images, thereby improving the prediction accuracy of the prediction neural network.
[0076] In another embodiment of the present application, the process of training the initial prediction neural network through training data and loss function includes: inputting the training data into the initial prediction neural network to obtain prediction window adjustment parameters; calculating the loss value through the loss function, sample window adjustment parameters and prediction window adjustment parameters; and using the loss value to update the weight parameters of the initial prediction neural network.
[0077] When training the initial prediction neural network, the present application inputs the training data into the initial prediction neural network to obtain the prediction window adjustment parameters, and then calculates the loss value through the loss function, the sample window adjustment parameters and the prediction window adjustment parameters, and uses the loss value to update the weight parameters of the initial prediction neural network until the prediction window width value and the prediction window position value output by the initial prediction neural network reach a certain optimal solution, and obtains the trained prediction neural network. Among them, the prediction window adjustment parameters include the prediction window width value and the prediction window position value, and the sample window adjustment parameters are the actual window width value and the actual window position value of the sample medical image.
[0078] The initial prediction neural network in this application uses the loss function for constraint, iteratively optimizes the initial prediction neural network, and continuously adjusts the weights to obtain a prediction neural network. The loss function is used to calculate similarity based on the actual window width value and actual window level value of the sample medical imaging data, and the predicted window width value and predicted window level value of the sample medical imaging data adjusted by the initial prediction neural network, and the similarity result is the loss value. The application can calculate the loss value by any one of the mean square error similarity calculation method, the mean absolute error similarity calculation method, and the structural similarity calculation method.
[0079] In this application, the loss function can be calculated using the mean absolute error (MAE) of the window width and window position. The specific formula is as follows:
[0080]
[0081] Among them, L MAE is the mean absolute error, i.e., the loss value; n is the number of sample medical images, y i Indicates the actual window width of the sample medical image, x i Indicates the actual window level value of the sample medical image, y i ′ represents the prediction window width of the sample medical image output by the initial prediction neural network, x i ′ represents the predicted window value of the sample medical image output by the initial prediction neural network. The integrated neural network model is iteratively optimized according to the loss result, and the weights are continuously adjusted to obtain the trained prediction neural network.
[0082] In summary, it can be seen that the present application can train the initial prediction neural network into the final prediction neural network through the sample grayscale value feature vector, sample image information feature vector and loss function of the sample medical image, and input the grayscale value feature vector and image information feature vector of the target medical image to be adjusted into the prediction neural network, that is, predict the accurate window width and window position values, thereby realizing automatic adjustment of the window width and window position.
[0083] See also Figure 3 , Figure 3 A flowchart of a method for predicting window adjustment parameters provided in an embodiment of the present application is provided, and the method comprises the following steps:
[0084] S301, obtaining a gray value feature vector and a sequence category feature vector of a target medical image;
[0085] S302, determining an image statistical vector using a gray value feature vector;
[0086] S303, extracting a neural network through an image feature prediction neural network to extract grayscale value encoding features of a grayscale value feature vector;
[0087] S304, extracting image information coding features of image information feature vectors by predicting information feature extraction neural networks of neural networks; wherein the image information feature vectors include image statistical vectors and sequence category feature vectors; the image feature extraction neural network is a deep learning feature extraction network; and the information feature extraction neural network is a shallow neural network;
[0088] S305, fusing the gray value coding feature and the image information coding feature to obtain an image fusion coding feature;
[0089] S306, inputting the image fusion coding features into the fully connected layer to obtain predicted window adjustment parameters; the window adjustment parameters include window width value and window level value; wherein the fully connected layer is a fully connected layer of the prediction neural network.
[0090] See also Figure 4 , is a schematic diagram of a prediction neural network structure provided by an embodiment of the present invention, through Figure 4It can be seen that the prediction neural network includes: an image feature extraction neural network, an information feature extraction neural network and a fully connected layer. Among them, the image feature extraction neural network is a deep learning feature extraction network, including but not limited to CNN (Convolutional Neural Networks), VGG (Visual Geometry Group Network, deep convolutional neural network), ResNet (Residual Network, residual network) and other convolutional network models; the gray value feature vector is input into the image feature extraction neural network, and the image feature extraction neural network can extract the gray value encoding features of the gray value feature vector.
[0091] Among them, the information feature extraction neural network is a shallow neural network, including but not limited to fully connected networks (FCN, Fully Convolutional Networks), one-dimensional convolutional networks (1DCNN, 1D ConvolutionalNeuralNetwork,) and other neural network models. The present application inputs the image information feature vector into the information feature extraction neural network, and the information feature extraction neural network can extract the image information coding features of the image information feature vector. Finally, the gray value coding feature is fused with the image information coding feature to obtain the image fusion coding feature. The image fusion coding feature is input into the fully connected layer, and the most suitable window width value and window level value of the target medical image are finally output.
[0092] In summary, it can be seen that the present application has the following beneficial effects in predicting the window adjustment parameters by predicting the neural network:
[0093] 1. This application can process medical images of different sequences:
[0094] After the prediction neural network is trained through the sequence type information of the sample medical images, the prediction neural network can distinguish the differences in window width and window position of different sequences. Therefore, after the present application obtains the sequence type information of the target medical image to be processed and inputs it into the prediction neural network, the prediction neural network can more specifically predict the window width and window position of the medical image according to the difference in sequence type information, so as to adjust the window width and window position. The prediction neural network has strong generalization ability.
[0095] 2. This application integrates multi-dimensional image information:
[0096] The present application needs to obtain multi-dimensional information of medical imaging data, which includes gray value feature vectors, image statistical vectors and sequence category feature vectors, wherein the gray value feature vectors can reflect the anatomical structure differences of medical images, and different anatomical structures have different requirements for window width and window position; the image statistical vectors can reflect the gray value statistical information of medical images, and different gray value statistical information has different corresponding relationships with window width and window position; the sequence category feature vectors can reflect the sequence type of medical images, and different sequence types also have different corresponding relationships with window width and window position. Therefore, the prediction neural network in the present application is based on the gray value feature vectors, image statistical vectors and sequence category feature vectors, and predicts the window adjustment parameters from the above multiple dimensions, thereby outputting more accurate window width and window position adjustment results.
[0097] It should be noted that the present application is a pioneering solution for jointly predicting the window width and window position by combining the gray value feature vector, the image statistics vector and the sequence category feature vector. Here, the potential correlation between the sequence type and the window width and window position is used as an example for explanation:
[0098] 128 cases of lumbar spine imaging data were collected, and each case of imaging data contained different types of data (T1, T2, and fat suppression sequence W / F). For each type of data of each case of imaging data, three experienced professional doctors adjusted the window width and window position to obtain the corresponding window adjustment parameter values. Subsequently, the distribution of window width and window position adjusted by different doctors under each type was statistically analyzed. The experimental results showed that when different doctors adjusted the window width and window position, there were certain fluctuations between the parameter values of each doctor due to differences in subjective evaluation criteria. However, the distribution of different types of window width and window position was generally concentrated and showed significant sequence specificity:
[0099] T1: The window width and window position are mainly concentrated in the range of [200,400] to [1200,2500];
[0100] T2: The window width and window position are mainly concentrated in the range of [300,600] to [1600,3500];
[0101] W / F: Window width and window position are mainly concentrated in the range of [100,200] to [1100,2400].
[0102] It can be seen that although the overall range of different types of window widths and window positions covers [100,200] to [1600,3500], there are significant differences in the distribution intervals and characteristics of each type. This shows that there is a certain correlation between the window width and window position and the sequence type. Different types of sequences need to be specifically optimized according to their sequence characteristics when adjusting the window width and window position, which provides an important basis for the design of the window width and window position prediction model. Therefore, the method proposed in this application that combines sequence types, image dimensions and statistical dimensions can significantly improve the accuracy of window width and window position prediction, and provides a new solution for the accurate processing of medical images.
[0103] See also Figure 5 , Figure 5 A schematic diagram of a device for predicting window adjustment parameters provided in an embodiment of the present application is provided, and the device specifically comprises:
[0104] A first acquisition module 11 is used to acquire a gray value feature vector and a sequence category feature vector of a target medical image;
[0105] A determination module 12, configured to determine an image statistical vector using the gray value feature vector;
[0106] The prediction module 13 is used to extract the grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector through a prediction neural network, and obtain the predicted window adjustment parameters using the grayscale value coding features and the image information coding features; wherein the image information feature vector includes the image statistics vector and the sequence category feature vector.
[0107] As an optional embodiment, the first acquisition module is specifically used to: obtain the gray value feature vector and sequence category feature vector corresponding to the target type sequence in the target medical image; wherein the target type sequence is any one of the T1 sequence, T2 sequence and W / F sequence.
[0108] As an optional embodiment, the determining module includes:
[0109] A determination unit, configured to determine statistical information of the grayscale value feature vector; the statistical information includes at least one of a grayscale maximum value, a grayscale minimum value, a grayscale median value, a grayscale mean value, a grayscale variance, and a grayscale percentile of the grayscale value feature vector;
[0110] A generating unit is used to generate an image statistical vector according to the statistical information.
[0111] As an optional embodiment, the prediction device further includes:
[0112] A second acquisition module is used to acquire training data; wherein the training data includes: a sample gray value feature vector and a sample image information feature vector;
[0113] The training module is used to train the initial prediction neural network through the training data and the loss function to obtain the trained prediction neural network.
[0114] As an optional embodiment, the training module includes:
[0115] An input unit, used for inputting the training data into the initial prediction neural network to obtain prediction window adjustment parameters;
[0116] A calculation unit, used to calculate a loss value through a loss function, a sample window adjustment parameter and the predicted window adjustment parameter;
[0117] An adjustment unit is used to update the weight parameters of the initial prediction neural network using the loss value.
[0118] As an optional embodiment, the prediction module includes:
[0119] A first extraction unit is used to extract a grayscale value encoding feature of the grayscale value feature vector by predicting an image feature extraction neural network of the neural network;
[0120] A second extraction unit is used to extract the image information coding feature of the image information feature vector by predicting the information feature extraction neural network of the neural network;
[0121] Among them, the image feature extraction neural network is a deep learning feature extraction network; the information feature extraction neural network is a shallow neural network.
[0122] As an optional embodiment, the prediction module includes:
[0123] A fusion unit, used for fusing the gray value coding feature and the image information coding feature to obtain an image fusion coding feature;
[0124] An acquisition unit is used to input the image fusion coding features into a fully connected layer to obtain predicted window adjustment parameters; the window adjustment parameters include a window width value and a window level value; wherein the fully connected layer is a fully connected layer of the prediction neural network.
[0125] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0126] See also Figure 6 , Figure 6A schematic diagram of the structure of an electronic device provided in an embodiment of the present application, the electronic device specifically includes:
[0127] The processor 21, the memory 22 and the computer program stored in the memory 22 and executable on the processor 21, the processor 21 executes the steps of the prediction method described in any of the above method embodiments through the computer program.
[0128] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.
[0129] The memory 22 may include one or more computer-readable storage media, which may be non-transitory. The memory 22 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 22 is at least used to store the following computer program 221, wherein, after the computer program is loaded and executed by the processor 21, it can implement the relevant steps in the prediction method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 22 may also include an operating system 222 and data 223, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 222 may include Windows, Unix, Linux, etc.
[0130] In some embodiments, the electronic device may further include a display screen 23 , an input / output interface 24 , a communication interface 25 , a sensor 26 , a power source 27 , and a communication bus 28 .
[0131] certainly, Figure 6 The structure of the electronic device shown does not constitute a limitation on the electronic device in the embodiments of the present application. In actual applications, the electronic device may include Figure 6 More or fewer components than shown, or combinations of certain components.
[0132] In another exemplary embodiment, a computer storage medium is also provided, and when the program instructions are executed by a processor, the steps of the prediction method described in any of the above method embodiments are implemented. The storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0133] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.
[0134] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.
[0135] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A method for predicting window adjustment parameters, characterized in that: include: Obtaining gray value feature vectors and sequence category feature vectors of target medical images; Determine an image statistics vector using the gray value feature vector; The grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector are extracted through a predictive neural network, and the predicted window adjustment parameters are obtained using the grayscale value coding features and the image information coding features; wherein the image information feature vector includes the image statistics vector and the sequence category feature vector.
2. The prediction method according to claim 1, characterized in that: The step of obtaining the gray value feature vector and the sequence category feature vector of the target medical image includes: Obtaining a gray value feature vector and a sequence category feature vector corresponding to a target type sequence in a target medical image; wherein the target type sequence is any one of a T1 sequence, a T2 sequence, and a W / F sequence.
3. The prediction method according to claim 1, characterized in that: The method of determining the image statistical vector by using the gray value feature vector comprises: Determine statistical information of the grayscale value feature vector; the statistical information includes at least one of the grayscale maximum value, grayscale minimum value, grayscale median value, grayscale mean value, grayscale variance, and grayscale percentile of the grayscale value feature vector; An image statistics vector is generated according to the statistical information.
4. The prediction method according to claim 1, characterized in that: Before obtaining the gray value feature vector and the sequence category feature vector of the target medical image, the method further includes: Acquire training data; wherein the training data includes: a sample gray value feature vector and a sample image information feature vector; The initial prediction neural network is trained using the training data and the loss function to obtain the trained prediction neural network.
5. The prediction method according to claim 4, characterized in that: The training of the initial prediction neural network using the training data and the loss function includes: Inputting the training data into the initial prediction neural network to obtain prediction window adjustment parameters; Calculate the loss value through the loss function, the sample window adjustment parameter and the predicted window adjustment parameter; The loss value is used to update the weight parameters of the initial prediction neural network.
6. The prediction method according to any one of claims 1 to 5, characterized in that: The grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector are extracted by predicting the neural network, including: Extracting a grayscale value encoding feature of the grayscale value feature vector by predicting an image feature extraction neural network of the neural network; Extracting a neural network by predicting information features of a neural network to extract image information coding features of the image information feature vector; Among them, the image feature extraction neural network is a deep learning feature extraction network; the information feature extraction neural network is a shallow neural network.
7. The prediction method according to claim 6, characterized in that: Obtaining predicted window adjustment parameters using the grayscale value encoding feature and the image information encoding feature, including: Performing feature fusion on the gray value coding feature and the image information coding feature to obtain an image fusion coding feature; The image fusion coding features are input into a fully connected layer to obtain predicted window adjustment parameters; the window adjustment parameters include a window width value and a window level value; wherein the fully connected layer is a fully connected layer of the prediction neural network.
8. A prediction device for adjusting window parameters, characterized in that: include: A first acquisition module is used to acquire a gray value feature vector and a sequence category feature vector of a target medical image; A determination module, used for determining an image statistical vector using the gray value feature vector; A prediction module is used to extract the grayscale value coding features of the grayscale value feature vector and the image information coding features of the image information feature vector through a prediction neural network, and obtain predicted window adjustment parameters using the grayscale value coding features and the image information coding features; wherein the image information feature vector includes the image statistics vector and the sequence category feature vector.
9. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the steps of the prediction method described in any one of claims 1 to 7 of the present application through the computer program.
10. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, and the computer executable instructions are used to execute the steps of the prediction method described in any one of claims 1 to 7 of the present application.