Nuclear magnetic resonance image prediction method and system

Abstract

The invention relates to the technical field of image analysis, in particular to a nuclear magnetic resonance image prediction method and system. The method comprises the following steps: acquiring basic information, historical image data and current image data of a patient, performing data multi-modal feature comparison, and outputting a comparison result; respectively positioning focus areas of the current image and the historical image, and filling buffer areas for the focus areas; calculating similarity weights of the spatial features and the time features, optimizing a fusion proportion of historical data and current data according to the individual features of the patient, obtaining prediction data, and outputting the prediction data; the system comprises a multi-modal feature comparison module, a focus area positioning module and a fusion proportion optimization prediction module. Data multi-modal feature comparison is performed on basic information, historical image data and current image data of a patient, whether a historical nuclear magnetic resonance image and a current magnetic resonance image belong to the same patient or not is judged, and then prediction operation is performed, so that diagnosis accuracy is ensured, and medical errors are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to a method and system for predicting nuclear magnetic resonance images. Background Art

[0002] In current medical diagnosis and treatment, accurately obtaining patient information and making reasonable predictions are crucial for improving treatment outcomes and prognosis. However, traditional medical diagnosis relies primarily on the doctor's intuitive judgment based on current images and basic patient information. Sometimes, patients provide historical MRI images, which can be used to predict the patient's condition.

[0003] However, doctors cannot determine whether the historical MRI images provided by patients themselves are the patient's data, and cannot ensure the accuracy of the diagnosis, which leads to medical errors. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for predicting nuclear magnetic resonance images, aiming to solve the technical problem in the prior art that it is impossible to determine whether the historical nuclear magnetic resonance images provided by the patient themselves are the patient's data, and it is impossible to ensure the accuracy of the diagnosis, which leads to medical errors.

[0005] To achieve the above object, the present invention adopts a method for predicting nuclear magnetic resonance images, comprising the following steps:

[0006] Obtain patient basic information, historical imaging data, and current imaging data, perform multimodal feature comparison of the data, and output the comparison results;

[0007] Locate the lesion areas of the current image and the historical image respectively, and fill the buffer area for the lesion area;

[0008] Calculate the similarity weights of spatial features and temporal features, optimize the fusion ratio of historical data and current data according to the individual characteristics of the patient, obtain predicted data, and output it.

[0009] Among them, in the step of obtaining the patient's basic information, historical imaging data and current imaging data, performing multimodal feature comparison of the data, and outputting the comparison results:

[0010] Obtain basic patient information and current images, and obtain uploaded data of historical images;

[0011] Characterize the patient's basic information and extract basic information features;

[0012] Perform feature extraction on image data to obtain historical image features and current image features respectively;

[0013] Compare multimodal features and output comparison results.

[0014] Among them, in the step of characterizing the patient's basic information and extracting the basic information features:

[0015] Extract the numerical data from the basic information and obtain the numerical feature vector after normalization;

[0016] Extract categorical data from basic information and obtain categorical feature vectors after dimension division;

[0017] Extract text data from basic information and obtain text feature vectors by weighted averaging the vectors of text words.

[0018] Among them, in the step of extracting features from image data and obtaining historical image features and current image features respectively:

[0019] Calculate the image texture features and perform feature extraction to obtain the historical image feature vector and the current image feature vector respectively.

[0020] Among them, in the step of comparing multimodal features and outputting the comparison results:

[0021] Each vector is spliced ​​together to obtain a comprehensive feature vector, and a similarity measure is calculated. Based on the result of the similarity measure, a comparison report is generated.

[0022] Among them, in the steps of locating the lesion areas of the current image and the historical image respectively, and filling the buffer area for the lesion area:

[0023] Perform pixel-level segmentation on the current image and historical images respectively, set a threshold, divide the pixels into lesions and non-lesions based on the threshold, and locate the lesion area.

[0024] Among them, after performing pixel-level segmentation on the current image and the historical image, setting a threshold, dividing the pixels into lesions and non-lesions based on the threshold, and locating the lesion area:

[0025] The lesion area is expanded and filled with a buffer area.

[0026] Among them, in the steps of calculating the similarity weights of spatial features and temporal features, optimizing the fusion ratio of historical data and current data according to individual patient characteristics, obtaining predicted data, and outputting:

[0027] Extract spatial features and temporal features separately;

[0028] The similarity weights of spatial features and temporal features are calculated separately, and the historical data and current data are fused to obtain the predicted data.

[0029] Among them, in the steps of extracting spatial features and temporal features respectively:

[0030] Extract spatial features of the lesion area and its buffer area of ​​the historical image and the current image respectively;

[0031] Compare the changes of lesions in historical images and current images to extract temporal features.

[0032] The present invention also provides a nuclear magnetic resonance image prediction system, comprising a multimodal feature comparison module, a lesion area positioning module, and a fusion ratio optimization prediction module; wherein:

[0033] The multimodal feature comparison module is used to obtain basic patient information, historical image data and current image data, perform multimodal feature comparison of the data, and output the comparison results;

[0034] The lesion area positioning module is used to respectively locate the lesion area of ​​the current image and the historical image, and fill the buffer area for the lesion area;

[0035] The fusion ratio optimization prediction module is used to calculate the similarity weights of spatial features and temporal features, optimize the fusion ratio of historical data and current data according to individual patient characteristics, obtain predicted data, and output it.

[0036] A nuclear magnetic resonance image prediction method and system of the present invention respectively adopt the multimodal feature comparison module, the lesion area positioning module, and the fusion ratio optimization prediction module to perform the following steps: obtaining patient basic information, historical image data and current image data, performing data multimodal feature comparison, and outputting the comparison result; locating the lesion area of ​​the current image and the historical image respectively, and filling the buffer area for the lesion area; calculating the similarity weight of spatial features and temporal features, optimizing the fusion ratio of historical data and current data according to the individual characteristics of the patient, obtaining predicted data, and outputting it; by performing data multimodal feature comparison on the patient basic information, historical image data and current image data, determining whether the historical nuclear magnetic resonance image and the current magnetic resonance image belong to the same patient, and then performing a prediction operation, thereby ensuring the accuracy of the diagnosis and avoiding medical errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 It is a flowchart of the steps of the nuclear magnetic resonance image prediction method of the present invention.

[0039] Figure 2 It is a step flow chart of S100 of the present invention.

[0040] Figure 3 It is a step flow chart of S200 of the present invention.

[0041] Figure 4 It is a step flow chart of S300 of the present invention.

[0042] Figure 5 It is a structural principle diagram of the nuclear magnetic resonance image prediction system of the present invention.

[0043] Figure 6 It is a structural principle diagram of the electronic device of the present invention.

[0044] 401-Multimodal feature comparison module, 402-Lesion area positioning module, 403-Fusion ratio optimization prediction module. DETAILED DESCRIPTION

[0045] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with this application.

[0046] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. As used in this application and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0047] It should be understood that although the terms first, second, third, etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0048] See also Figures 1 to 4 The present invention provides a method for predicting nuclear magnetic resonance images, comprising the following steps:

[0049] S100: Obtain patient basic information, historical imaging data, and current imaging data, perform multimodal feature comparison of the data, and output the comparison results.

[0050] In this embodiment, basic patient information, historical image data, and current image data are obtained, and multimodal feature comparison of the data is performed to output the comparison results. The specific process is as follows:

[0051] S101: Obtaining basic patient information and current images, and obtaining uploaded data of historical images;

[0052] S102: Characterizing the patient's basic information and extracting basic information features;

[0053] S103: Extract features from the image data to obtain historical image features and current image features;

[0054] S104: Compare the multimodal features and output the comparison results.

[0055] Furthermore, in the step of characterizing the patient's basic information and extracting basic information features:

[0056] Extract the numerical data from the basic information and obtain the numerical feature vector after normalization;

[0057] Extract categorical data from basic information and obtain categorical feature vectors after dimension division;

[0058] Extract text data from basic information and obtain text feature vectors by weighted averaging the vectors of text words.

[0059] Furthermore, in the step of extracting features from the image data and obtaining historical image features and current image features respectively:

[0060] Calculate the image texture features and perform feature extraction to obtain the historical image feature vector and the current image feature vector respectively.

[0061] Furthermore, in the step of comparing the multimodal features and outputting the comparison results:

[0062] Each vector is spliced ​​together to obtain a comprehensive feature vector, and a similarity measure is calculated. Based on the result of the similarity measure, a comparison report is generated.

[0063] In this process, a data interface is established between the hospital information system (HIS) and the electronic medical record module. Using SQL queries, basic patient information, including age, gender, medical history, and allergy history, is extracted from the electronic medical record database based on the patient's unique identifier (such as the medical record number). For example, for the age field, the numerical value is directly read from the database; for the medical history field, a detailed text description is obtained.

[0064] At the same time, the relevant information of the current visit of the patient is obtained from the registration system, such as the department of visit, visit time, etc., so as to more comprehensively understand the current diagnosis and treatment situation of the patient.

[0065] With the query interface of the medical image archive and communication system (PACS). Input the patient medical record number and other key information, and the PACS system returns the image data list of the current examination of the patient. According to the predetermined image format (such as DICOM format) and transmission protocol (such as HTTP or FTP), the image data is downloaded from the PACS server. During the downloading process, the integrity of the image data is checked, such as whether the pixel matrix of the examination image is complete, whether there are missing frames, etc.

[0066] If the patient provides historical magnetic resonance images by himself, the historical magnetic resonance images are scanned to obtain digital information of the historical magnetic resonance images.

[0067] For numerical data such as age, directly as a feature value. In order to adapt to subsequent model analysis, the age is normalized. Assuming that the age range is between 0-100 years old, the normalization formula is:

[0068]

[0069] Where x is the original age value, x min is the minimum age (0 years old), x max is the maximum age (100 years old), x norm is the normalized value. The normalized age value forms a numerical feature vector

[0070] For gender, occupation and other categorical data, One-Hot Encoding is used. For example, gender has two categories of "male" and "female", and the encoding is "male": [1, 0], "female": [0, 1]. If there are N types of occupation categories, an N-dimensional vector is used, and each occupation corresponds to a dimension. The occupation corresponding to the dimension is 1, and the rest is 0. The One-Hot Encoding vectors of all categorical data are combined into a categorical feature vector

[0071] For text data such as medical history description and allergy history, natural language processing technology is used for feature extraction. First, the text is split into a sequence of words through word segmentation processing. Then, a word embedding model such as Word2Vec is used to map each word to a fixed-dimensional vector. Assuming that the vector dimension of the Word2Vec model is d, for a text containing m words, its feature vector can be the weighted average of all word vectors, that is:

[0072]

[0073] in, is the word vector of the i-th word, w i is the weight of the i-th word.

[0074] The numerical feature vector Categorical feature vector and text feature vector Perform splicing to obtain the basic information feature vector:

[0075]

[0076] Normalize the historical and current images and adjust the grayscale value range of the images to the [0,1] interval. The formula is:

[0077]

[0078] Among them, I is the original image pixel value, I min is the minimum pixel value in the image, I max is the maximum pixel value in the image, I norm is the normalized pixel value.

[0079] Median filtering is used for denoising. For each pixel in the image, the median of the pixel values ​​in its neighborhood is taken as the new value of the pixel.

[0080] Use histogram equalization for image enhancement to redistribute pixel values ​​to make the histogram of the image more uniform, thereby enhancing the contrast of the image.

[0081] Calculate the texture features of the image, such as using the gray-level co-occurrence matrix (GLCM). The gray-level co-occurrence matrix P(i,j) represents the probability that pixels with gray levels i and j appear at the same time in a specific direction and distance in the image. Various texture features can be extracted from the GLCM, such as contrast:

[0082]

[0083] Where L is the number of gray levels;

[0084] Dependencies:

[0085]

[0086] Among them, μ i and μ j are the means of i and j, σ i and σ j are the standard deviations of i and j respectively.

[0087] The extracted texture features are grouped into a feature vector. For the historical image and the current image, the above texture feature extraction process is performed respectively to obtain a historical image feature vector and a current image feature vector

[0088] The Euclidean distance is used for similarity measurement. For two feature vectors and Euclidean distance:

[0089]

[0090] where n is the dimension of the feature vector, F 1i and F 2i are the i-th elements of the two feature vectors respectively. The Euclidean distance d basic between the basic information feature vectors, the Euclidean distance d image between the historical image and the current image feature vectors can be calculated respectively. If the distance is less than a threshold, such as a threshold of 5, it is determined that it is the same patient; otherwise, it is prompted that the data does not match.

[0091] According to the result of similarity measurement, a comparison report is generated. The report includes the similarity scores of the basic information features, the historical image features and the current image features in each dimension, as well as the overall multi-modal feature similarity comprehensive evaluation. At the same time, charts (such as bar charts, line charts) can be used to visually display the comparison results.

[0092] S200: The lesion area of the current image and the historical image is located respectively, and a buffer area is filled for the lesion area.

[0093] In this embodiment, the lesion area of the current image and the historical image is located respectively, and a buffer area is filled for the lesion area. The specific process is as follows:

[0094] S201: The current image and the historical image are segmented at the pixel level respectively, a threshold is set, the pixels are divided into lesions and non-lesions according to the threshold, and the lesion area is located;

[0095] S202: The lesion area is expanded to fill the buffer area.

[0096] In the above process, the threshold T is set according to the statistical characteristics of the image. Based on the histogram distribution of the image, the threshold is determined by calculating the peak and valley of the histogram, for example, the threshold is automatically calculated using Otsu's method. Otsu's method finds the best threshold by maximizing the inter-class variance. Let the gray level of the image be 0 to L-1, the number of pixels of gray level i be n i , and the total number of pixels be The probability of gray level i is:

[0097]

[0098] The image is divided into two categories: C0 (gray level between 0 and T) and C1 (gray level between T+1 and L-1), and the between-class variance is:

[0099]

[0100] in:

[0101]

[0102]

[0103] Traverse all possible thresholds T and find The largest T is taken as the optimal threshold.

[0104] Pixel classification, according to the set threshold T, divides the pixels in the image into lesions and non-lesions. For the normalized image, if the pixel value I norm If (x,y)>T, the pixel is considered to belong to the lesion area; otherwise, it belongs to the non-lesion area.

[0105] The binary image B is obtained through the binarization operation, where:

[0106]

[0107] To locate the lesion area, perform connected region analysis on the binary image B and mark all connected regions using a connected region marking algorithm (such as a two-pass scanning algorithm). A connected region is an area consisting of adjacent pixels with the same label. For the marked connected regions, the area threshold A is set according to the preset area threshold. min The lesion area is screened out by using the shape characteristics (such as aspect ratio, circularity, etc.).

[0108] For example, circularity:

[0109]

[0110] Where S is the area of ​​the connected region, L is the perimeter of the connected region, and the closer the circularity is to 1, the closer the region is to a circle, which may be a lesion area.

[0111] Buffer area expansion, using morphological dilation operation to expand the lesion area. The dilation operation is to add the structural element S to the binary image B of the lesion area. lesion Slide on the image and set the pixel value covered by the structural element to 1. Assume that the center of the structural element S is located at (x0, y0), and the expanded image B dilated The calculation formula is:

[0112] B dilated(x,y)=max{B lesion (x+x s ,y+y s )∣(x s ,y s )∈S}

[0113] The size and shape of the structuring element can be adjusted according to actual needs. Common structuring elements include circles and rectangles. For example, using a circular structuring element with a radius of r to perform an expansion operation can increase the area of ​​the lesion region and form a buffer region.

[0114] Buffer area size control: To control the size of the buffer area, you can set the number of dilations n or the dilation pixel distance d. If the number of dilations n is set, the dilation operation is performed n times; if the dilation pixel distance d is set, the dilation step size is adjusted based on the size and shape of the structuring element, so that the lesion area expands outward by d pixels.

[0115] There may be some holes or discontinuous areas in the dilated buffer area. Use the morphological fill operation to fill these areas.

[0116] S300: Calculate the similarity weights of spatial features and temporal features, optimize the fusion ratio of historical data and current data according to the individual characteristics of the patient, obtain predicted data, and output it.

[0117] In this embodiment, the similarity weights of spatial features and temporal features are calculated, the fusion ratio of historical data and current data is optimized according to the individual characteristics of the patient, and the predicted data is obtained and output. The specific process is:

[0118] S301: Extracting spatial features of the lesion area and its buffer area in the historical image and the current image respectively;

[0119] S302: Compare the changes of the lesions in the historical image and the current image to extract the time features;

[0120] S303: Calculate the similarity weights of spatial features and temporal features respectively, fuse historical data with current data, and obtain predicted data.

[0121] In the above process, shape features are extracted and the area S of the lesion region and its buffer region is calculated. For the binary lesion region image B, the area can be obtained by counting the number of pixels with a pixel value of 1, that is:

[0122]

[0123] Where M and N are the number of rows and columns of the image, respectively.

[0124] The chain code method is used to calculate the perimeter L of the lesion area. Chain code is a coding method used to represent boundaries. By traversing the boundary pixels of the lesion area, recording the direction changes between adjacent pixels, and then calculating the perimeter based on the length of the chain code.

[0125] The circularity C is calculated based on the area and perimeter, and the formula is:

[0126]

[0127] The closer the circularity is to 1, the closer the lesion area is to a circle.

[0128] Texture feature extraction and gray level co-occurrence matrix (GLCM), as shown in step S100, respectively obtain contrast and correlation.

[0129] Combine the extracted shape features and texture features into a spatial feature vector:

[0130]

[0131] Image registration techniques are used to align the historical image with the current image, ensuring spatial alignment of the same anatomical structures in the two images. Common image registration methods include feature point-based registration, such as using the SIFT (Scale-Invariant Feature Transform) algorithm to extract feature points from the two images. The transformation parameters are then found by matching the feature points to transform the historical image to the coordinate system of the current image.

[0132] The center coordinate differences Δx and Δy of the lesion area in the historical image and the current image after registration, as well as the offset angle θ of the lesion area, are calculated. These parameters can reflect the changes in the spatial position of the lesion.

[0133] Changes in lesion size, calculating the area S of the lesion region in the historical image and the current image respectively history and S current , calculate the area change rate:

[0134]

[0135] Changes in lesion morphology, calculation of the circularity C of the lesion area in the historical image and the current image history and C current , and the rate of change of circumference:

[0136]

[0137] These parameters can reflect the morphological changes of the lesions.

[0138] Temporal feature vector construction: Combine the above extracted features such as position change, size change, and shape change into a temporal feature vector:

[0139]

[0140] Cosine similarity calculation: calculate spatial feature vectors separately and time feature vector With the predefined reference feature vector and The cosine similarity of . The spatial feature reference vector is generated by extracting typical lesion features from images of healthy people; the temporal feature reference vector is generated by extracting temporal features from images of the same patient during the stable period. The cosine similarity formula is:

[0141]

[0142] in, is the dot product of two eigenvectors, and are the moduli of the two feature vectors. Let the spatial feature cosine similarity be similarity spatial , the cosine similarity of time features is similarity temporal .

[0143] Weight normalization: normalize the similarity of spatial features and temporal features to obtain the similarity weight w spatial and w temporal The normalization formula is:

[0144]

[0145] Consider the individual characteristics of the patient, such as age. For example, for older patients, due to the large changes in physical function, they may be more dependent on the current imaging data, so the weight of the time feature can be appropriately increased. Assume that the weight adjusted according to the individual characteristics of the patient is:

[0146]

[0147] Where I and β are adjustment coefficients determined according to individual patient characteristics, and α + β = 1.

[0148] Assume that the historical image data is D history , the current image data is D current , then the fused prediction data D predict It can be calculated by the following formula:

[0149]

[0150] The fused prediction data Dpredict It is output in the form of an image and can be accompanied by relevant feature parameters and weight information to facilitate further analysis and diagnosis by doctors.

[0151] Corresponding to the aforementioned embodiment of the nuclear magnetic resonance image prediction method, the present application also provides an embodiment of a nuclear magnetic resonance image prediction system.

[0152] Figure 5 FIG. 1 is a block diagram of a magnetic resonance imaging prediction system according to an exemplary embodiment. Figure 5 The system may include: a multimodal feature comparison module 401, a lesion area positioning module 402, and a fusion ratio optimization prediction module 403; wherein:

[0153] The multimodal feature comparison module 401 is used to obtain basic patient information, historical image data, and current image data, perform multimodal feature comparison on the data, and output the comparison results;

[0154] The lesion region positioning module 402 is used to respectively locate the lesion region of the current image and the historical image, and fill the buffer region for the lesion region;

[0155] The fusion ratio optimization prediction module 403 is used to calculate the similarity weights of spatial features and temporal features, optimize the fusion ratio of historical data and current data according to individual patient characteristics, obtain predicted data, and output it.

[0156] In this embodiment, the multimodal feature comparison module 401 obtains the patient's basic information, historical image data and current image data, performs data multimodal feature comparison, and outputs the comparison result; the lesion area positioning module 402 locates the lesion area of ​​the current image and the historical image respectively, and fills the buffer area for the lesion area; the fusion ratio optimization prediction module 403 calculates the similarity weights of spatial features and temporal features, optimizes the fusion ratio of historical data and current data according to the individual characteristics of the patient, obtains predicted data, and outputs it; by performing data multimodal feature comparison on the patient's basic information, historical image data and current image data, it is determined whether the historical magnetic resonance image and the current magnetic resonance image belong to the same patient, and then a prediction operation is performed, thereby ensuring the accuracy of the diagnosis and avoiding medical errors.

[0157] Regarding the system 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.

[0158] For the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0159] Accordingly, the present application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned magnetic resonance imaging prediction method. Figure 6 As shown in FIG. 1 , a hardware structure diagram of a nuclear magnetic resonance image prediction system provided by an embodiment of the present invention is provided for any device with data processing capability, except Figure 6 In addition to the processor, memory, and network interface shown, any device with data processing capabilities in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0160] Accordingly, the present application also provides a computer-readable storage medium having computer instructions stored thereon, which implement the above-mentioned magnetic resonance imaging prediction method when executed by a processor. The computer-readable storage medium can be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), an SD card, a flash card (Flash Card), etc. equipped on the device. Furthermore, the computer-readable storage medium can also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and can also be used to temporarily store data that has been output or is to be output.

[0161] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the contents disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art that are not disclosed in this application.

[0162] It will be understood that the present application is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof.

Claims

1. A method for predicting nuclear magnetic resonance images, characterized in that: The steps include: Obtain patient basic information, historical imaging data, and current imaging data, perform multimodal feature comparison of the data, and output the comparison results; Locate the lesion areas of the current image and the historical image respectively, and fill the buffer area for the lesion area; Calculate the similarity weights of spatial features and temporal features, optimize the fusion ratio of historical data and current data according to the individual characteristics of the patient, obtain predicted data, and output it.

2. The method for predicting nuclear magnetic resonance images according to claim 1, wherein: In the steps of obtaining basic patient information, historical imaging data, and current imaging data, performing multimodal feature comparison of the data, and outputting the comparison results: Obtain basic patient information and current images, and obtain uploaded data of historical images; Characterize the patient's basic information and extract basic information features; Perform feature extraction on image data to obtain historical image features and current image features respectively; Compare multimodal features and output comparison results.

3. The method for predicting nuclear magnetic resonance images according to claim 2, wherein: In the step of characterizing the patient's basic information and extracting basic information features: Extract the numerical data from the basic information and obtain the numerical feature vector after normalization; Extract categorical data from basic information and obtain categorical feature vectors after dimension division; Extract text data from basic information and obtain text feature vectors by weighted averaging the vectors of text words.

4. The method for predicting nuclear magnetic resonance images according to claim 3, wherein: In the steps of extracting features from image data and obtaining historical image features and current image features: Calculate the image texture features and perform feature extraction to obtain the historical image feature vector and the current image feature vector respectively.

5. The method for predicting nuclear magnetic resonance images according to claim 4, wherein: In the step of comparing multimodal features and outputting comparison results: Each vector is spliced ​​together to obtain a comprehensive feature vector, and a similarity measure is calculated. Based on the result of the similarity measure, a comparison report is generated.

6. The method for predicting nuclear magnetic resonance images according to claim 1, wherein: In the steps of locating the lesion areas of the current image and the historical image respectively, and filling the buffer area for the lesion area: Perform pixel-level segmentation on the current image and historical images respectively, set a threshold, divide the pixels into lesions and non-lesions based on the threshold, and locate the lesion area.

7. The method for predicting nuclear magnetic resonance images according to claim 6, wherein: After performing pixel-level segmentation on the current image and the historical image, setting a threshold, dividing pixels into lesions and non-lesions based on the threshold, and locating the lesion area: The lesion area is expanded and filled with a buffer area.

8. The method for predicting nuclear magnetic resonance images according to claim 1, wherein: In the steps of calculating the similarity weights of spatial and temporal features, optimizing the fusion ratio of historical data and current data based on individual patient characteristics, obtaining predicted data, and outputting it: Extract spatial features and temporal features separately; The similarity weights of spatial features and temporal features are calculated separately, and the historical data and current data are fused to obtain the predicted data.

9. The method for predicting nuclear magnetic resonance images according to claim 8, wherein: In the steps of extracting spatial features and temporal features respectively: Extract spatial features of the lesion area and its buffer area of ​​the historical image and the current image respectively; Compare the changes of lesions in historical images and current images to extract temporal features.

10. A nuclear magnetic resonance image prediction system, applied to the nuclear magnetic resonance image prediction method according to claim 1, characterized in that: It includes a multimodal feature comparison module, a lesion area positioning module, and a fusion ratio optimization prediction module; among which: The multimodal feature comparison module is used to obtain basic patient information, historical image data and current image data, perform multimodal feature comparison of the data, and output the comparison results; The lesion area positioning module is used to respectively locate the lesion area of ​​the current image and the historical image, and fill the buffer area for the lesion area; The fusion ratio optimization prediction module is used to calculate the similarity weights of spatial features and temporal features, optimize the fusion ratio of historical data and current data according to individual patient characteristics, obtain predicted data, and output it.

Images