Model construction method and device based on MRI habitat imaging, equipment and storage medium

Abstract

The embodiment of the invention provides a model construction method and device based on MRI habitat imaging, equipment and a storage medium, and the method comprises the steps: obtaining an MRI image of a glioma patient, and the MRI image comprises a DWI image and a T1-CE image; image preprocessing operation is carried out on the MRI image to obtain an ADC target image and a T1-CE target image marked with a region of interest, the ADC target image is obtained by converting a DWI image, and the T1-CE target image marked with the region of interest is obtained by converting a T1-CE image; a clustering method is adopted to group pixel gray values in the ADC target image and pixel gray values belonging to the region of interest in the T1-CE target image marked with the region of interest, a plurality of spatial habitats related to the glioma are obtained, and the spatial habitats are habitats displayed in an MRI image space; and constructing a prediction model for predicting the IDH state of the glioma according to the plurality of spatial habitats. The purpose of accurately and noninvasively predicting the IDH state of the glioma can be achieved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of medical image processing, and in particular to a model building method, device, equipment and storage medium based on MRI habitat imaging. Background Art

[0002] The status of isocitrate dehydrogenase (IDH) is associated with the prognosis, chemotherapy response and overall survival of glioma patients. Accurate and non-invasive preoperative prediction of IDH status can help glioma patients develop personalized treatment plans, prolong life, improve prognosis and improve quality of life. At present, the gold standard for detecting whether IDH is mutated is pathological analysis of tissue specimens obtained by surgical resection or puncture biopsy, including immunohistochemistry and DNA sequencing technology. Immunohistochemistry combines antibodies with specific antigens (peptides and proteins) in tissue cells to form antigen-antibody complexes, and uses specific stains to develop colors to achieve the purpose of showing the location and expression level of specific proteins. However, approximately 15% of IDH mutations will be missed using immunohistochemistry; and these processes for obtaining tumor tissues are invasive and may cause risks such as bleeding and infection. In addition, affected by genetic changes (single nucleotides and copy numbers), epigenetic changes (chromatin patterns and DNA methylation), and microenvironmental factors (drugs, hypoxia, and cell interactions), cells within the same tumor differ in genetic, epigenetic, and phenotypic characteristics, i.e., intra-tumor heterogeneity, which results in errors in specimen sampling, false negative results, and ultimately incorrect diagnosis of glioma grade and IDH status. Therefore, it is necessary to use a non-invasive method that comprehensively displays the spatial heterogeneity of gliomas to predict the IDH status of gliomas before surgery. Summary of the invention

[0003] The embodiment of the present invention proposes a model construction method, device, equipment and storage medium based on MRI habitat imaging to achieve non-invasive prediction of the IDH status of glioma.

[0004] In a first aspect, an embodiment of the present invention provides a model construction method based on MRI habitat imaging, comprising:

[0005] Acquiring MRI images of a glioma patient, wherein the MRI images include DWI images and T1-CE images;

[0006] Performing an image preprocessing operation on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon, wherein the ADC target image is obtained by converting the DWI image, and the T1-CE target image with a region of interest marked thereon is obtained by converting the T1-CE image;

[0007] A clustering method is used to group the grayscale values ​​of pixels in the ADC target image and the grayscale values ​​of pixels in the T1-CE target image marked with the region of interest, so as to obtain a plurality of spatial habitats related to glioma, wherein the spatial habitats are habitats displayed in the MRI image space;

[0008] A prediction model for predicting the IDH status of glioma is constructed based on the several spatial habitats.

[0009] In a second aspect, an embodiment of the present invention further provides a model building device based on MRI habitat imaging, comprising:

[0010] A data acquisition module, used to acquire MRI images of glioma patients, wherein the MRI images include DWI images and T1-CE images;

[0011] A data preprocessing module, used for performing image preprocessing operation on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon, wherein the ADC target image is obtained by converting the DWI image, and the T1-CE target image with a region of interest marked thereon is obtained by converting the T1-CE image;

[0012] A clustering module, for grouping the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​in the T1-CE target image marked with the region of interest by a clustering method, to obtain a plurality of spatial habitats related to glioma, wherein the spatial habitats are habitats displayed in the MRI image space;

[0013] A model building module is used to build a prediction model for predicting the IDH status of glioma based on the multiple spatial habitats.

[0014] In a third aspect, an embodiment of the present invention further provides a computer device, the computer device comprising:

[0015] one or more processors;

[0016] a memory for storing one or more programs,

[0017] When the one or more programs are executed by the one or more processors, the one or more processors implement the model building method as described in any one of the first aspects.

[0018] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the model building method as described in any one of the first aspects is implemented.

[0019] In an embodiment of the present application, an MRI image of a glioma patient is obtained, the MRI image including a DWI image and a T1-CE image; an image preprocessing operation is performed on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon, the ADC target image is converted from the DWI image, and the T1-CE target image with a region of interest marked thereon is converted from the T1-CE image; a clustering method is used to group the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​belonging to the region of interest in the T1-CE target image with the region of interest marked thereon, to obtain several spatial habitats related to the glioma, the spatial habitat being a habitat displayed in the MRI image space; and a prediction model for predicting the IDH state of glioma is constructed based on the several spatial habitats. Among them, T1-CE image is a conventional T1 enhanced image in magnetic resonance imaging (MRI) image, which can enhance the display effect of tumors. DWI image is diffusion-weighted imaging in MRI image, which is often used to detect the diffusion movement of water molecules in the human body. Some malignant tumor cells grow densely and have less free water, which can easily lead to restricted diffusion in malignant tumors and form high signals, that is, DWI images can also be used to diagnose tumors. Based on T1-CE images and DWI images obtained from glioma patients, the images are preprocessed and the areas of interest of tumor tissues are marked. The ecological environmental parameters related to glioma blood vessels and cell density are visualized, and the clustering method is used to group the pixel grayscale values ​​in the image, which can divide different habitats and realize habitat imaging. It can be understood that habitat imaging is an imaging genomics technology based on biological environment. Tumor cell groups with similar characteristics are clustered by quantitative imaging markers, so that the internal heterogeneity of tumors can be displayed and quantified. This application uses several spatial habitats to construct a prediction model for predicting the IDH status of glioma. Subsequently, the volume proportion, average T1-CE value and average ADC value of gliomas in different IDH states in different habitats can be further calculated, and the habitats that are more sensitive to changes in the IDH status of gliomas can be further inferred. The prediction model constructed by multiple habitats will be more comprehensive in quantifying tumor heterogeneity, which can improve the accuracy of IDH mutation prediction, and ultimately achieve the purpose of accurately and non-invasively predicting the IDH status of gliomas. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a model building method based on MRI habitat imaging provided in Example 1 of the present invention;

[0021] Figure 2 A structural block diagram of a model building device based on MRI habitat imaging provided in Embodiment 2 of the present invention;

[0022] Figure 3 A schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0024] Embodiment 1

[0025] Figure 1 A flowchart of a model building method based on MRI habitat imaging provided in Embodiment 1 of the present invention. This embodiment can be applied to non-invasive visualization of the tumor ecological environment of brain tumor patients and prediction of the IDH status of gliomas. The method can be executed by a model building device, which can be implemented by software and / or hardware and can be configured in a computer device, such as a server, a workstation, a personal computer, a medical system, a scanner, a medical detection device, etc., and specifically includes the following steps:

[0026] S110 , obtaining MRI images of glioma patients, where the MRI images include DWI images and T1-CE images.

[0027] In the embodiment of the present application, an MRI image of a glioma patient can be obtained by performing an MRI scan on the brain of the glioma patient, and an MRI image refers to image data obtained using magnetic resonance imaging (MRI) technology. For example, an MRI scanner can be configured to scan an object (such as a glioma patient) to obtain MRI image data, and the MRI scanner can detect at least two echo signals by applying an MRI pulse sequence to the object, which is not specifically limited in the embodiment of the present application.

[0028] It is understood that the relaxation parameters of an MRI scan object in the art generally include T1 (longitudinal relaxation time), T2 (transverse relaxation time), T2 * (the inverse of the transverse relaxation time), etc., T1 can be the time constant for the regrowth of the longitudinal magnetization (e.g., along the main magnetic field), T2 relaxation can be the process by which the transverse component of the magnetization decays or dephases, and T2 can be the time constant for the decay or dephasing of the transverse magnetization. In magnetic resonance imaging (MRI), relaxation parameters are a measure of how quickly hydrogen nuclei in tissue return to their original state, and are usually assessed by measuring the longitudinal relaxation time T1 and the transverse relaxation time T2 of the tissue.

[0029] In this application, DWI images are magnetic resonance imaging of the DWI sequence. The DWI sequence is a special sequence of magnetic resonance imaging. It is a diffusion-weighted imaging in MRI images and is often used to detect the diffusion movement of water molecules in the human body. Some malignant tumor cells grow densely and have less free water. Diffusion restriction is easily produced in malignant tumors, forming high signals, that is, DWI images can also be used to diagnose tumors. T1-CE images are magnetic resonance imaging of the T1-CE sequence. T1-CE is a special sequence in magnetic resonance imaging, also known as dynamic enhanced magnetic resonance imaging. This technology enhances the display effect of tumors or other tissues by injecting contrast agents (usually a gadolinium-based contrast agent); after the contrast agent is injected, by acquiring T1-weighted images at different time points, the uptake of the contrast agent by the tissue can be observed, thereby providing information about tissue blood supply, vascular permeability, and tumor angiogenesis. The T1-CE sequence is widely used in clinical practice, especially in tumor diagnosis and treatment evaluation. It can help doctors understand the growth and blood supply of tumors more accurately, and provide an important basis for formulating treatment plans. That is, the T1-CE image is a T1 enhanced image in magnetic resonance imaging, which can enhance the display effect of the tumor.

[0030] In the embodiment of the present application, the MRI images of glioma patients may be obtained from multiple MRI images collected from different hospitals and different patients, and may be obtained to obtain an MRI image data set, which may include T1-CE images and DWI images from different hospitals and different patients, that is, T1-CE images and DWI images may include multiple images. The embodiment of the present application does not limit the specific number.

[0031] S120 , performing image preprocessing operations on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon.

[0032] The ADC target image is obtained by converting the DWI image, and the T1-CE target image with the region of interest marked is obtained by converting the T1-CE image.

[0033] In a specific implementation, S120 may include the following steps:

[0034] S1201 , converting the native format of the MRI image into a target format to obtain a format-converted MRI image.

[0035] Since the MRI images acquired in the embodiments of the present application may be collected from different hospitals and different patients, each MRI image may differ in parameters such as signal intensity and scanning layer thickness due to different scanning devices. After the MRI images of all glioma patients are exported from the image archiving and communication system in DICOM format, in order to reduce the differences in image scanning layer thickness, signal intensity, etc. caused by different scanning devices and parameters, all MRI images in DICOM format may be converted into the neuroimaging informatics technology initiative (NIfTI) format.

[0036] S1202 , determining from the format-converted MRI image a DWI image with a diffusion gradient weighted value of 1000 as a first DWI image and a DWI image with a diffusion gradient weighted value of 0 as a second DWI image.

[0037] S1203 . Determine an ADC image according to the first DWI image and the second DWI image.

[0038] In this embodiment, the ADC image is calculated from the DWI image with diffusion gradient weighted values ​​b=1000 and b=0 s / mm2. For example, the ADC image can be calculated based on the formula: ADC=-ln[(S / S0) / b], which is used to calculate the surface diffusion coefficient (Apparent Diffusion Coefficient, ADC); wherein S0 is the non-diffusion gradient weighted data, S is the diffusion gradient weighted data, and b is the b value of the diffusion gradient.

[0039] The ADC image in the embodiment of the present application refers to an Surface Diffusion Coefficient (Apparent Diffusion Coefficient) image, which is a technique in magnetic resonance imaging and is used to describe the speed of diffusion of water molecules in tissue structures.

[0040] S1204, perform three-dimensional rigid registration of the ADC image and the T1-CE image in the format-converted MRI image with the standard T1 anatomical template respectively, and resample based on the registration results to obtain a T1-CE three-dimensional image and an ADC three-dimensional image with a target resolution.

[0041] The standard T1 anatomical template may be a SRI24 atlas standard T1 anatomical template, and the target resolution may be 1 mm×1 mm×1 mm.

[0042] S1205 , annotate the region of interest on the T1-CE three-dimensional image to obtain a T1-CE three-dimensional image with the region of interest annotated.

[0043] In this embodiment, the region of interest (ROI) of the image can be annotated manually by a manual annotation method. For example, a junior radiologist (3 years of experience in neuroimaging diagnosis) uses ITK-SNAP to manually outline the ROI along the contour of the largest tumor area on the T1-CE three-dimensional image, including the contrast-enhanced part (active enhanced tumor) and the non-enhanced tumor central component (necrosis), and is confirmed by another senior radiologist (15 years of experience in neuroimaging diagnosis). Both doctors are blinded to the clinical and pathological information.

[0044] In other implementations, a pre-trained image annotation model may be used to automatically annotate the region of interest on the T1-CE three-dimensional image to obtain a T1-CE three-dimensional image annotated with the region of interest.

[0045] S1206 , normalize the ADC three-dimensional image and the T1-CE three-dimensional image with the region of interest marked, respectively, to obtain an ADC target image and a T1-CE target image with the region of interest marked.

[0046] In a specific implementation of S1206, the signal intensity of the ADC three-dimensional image and the signal intensity of the T1-CE three-dimensional image with the region of interest marked can be normalized by using a kernel density estimation method to obtain an ADC target image and a T1-CE target image with the region of interest marked. More specifically, various types of images (such as ADC three-dimensional images or T1-CE three-dimensional images) can be unified to the same scale by scaling the pixel value size, thereby further reducing the difference in image signal intensity.

[0047] S130, using a clustering method to group the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​in the region of interest in the T1-CE target image marked with the region of interest, to obtain several spatial habitats related to glioma.

[0048] The spatial habitat is the habitat displayed in the MRI image space. "Habitat imaging" is to divide the regions with different biological characteristics on the image into several sub-regions with the same or similar characteristics, to ensure the homogeneity within the sub-regions to the greatest extent, and to achieve the quantification and visualization of intratumor heterogeneity.

[0049] In a specific implementation, S130 may include the following steps:

[0050] S1301: Determine the target number of clusters.

[0051] S1302. According to the target cluster number, the K-means clustering method is used to group the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​in the region of interest in the T1-CE target image marked with the region of interest, so as to obtain several spatial habitats related to glioma.

[0052] In one embodiment, the ADC target image includes multiple layers, pixels located at the same position in all layers constitute a voxel, and the ADC target image includes multiple voxels; the T1-CE target image with annotated regions of interest includes multiple layers, pixels located at the same position in all layers constitute a voxel, and the T1-CE target image includes multiple voxels; in order to determine the number of tumor habitats, that is, to obtain the optimal number of clusters, the range of the number of clusters can be set to 2 to 10. In one example, S1301 can include the following specific steps:

[0053] S13011. Determine an error curve formed by plotting the sum of squares of errors between the voxel grayscale value of each voxel in the ADC target image and the voxel grayscale value of each voxel in the region of interest in the T1-CE target image.

[0054] S13012. Determine the target number of clusters based on the error curve.

[0055] In one embodiment, S1302 may include the following specific steps:

[0056] Summarize the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​in the region of interest in the T1-CE target image marked with the region of interest;

[0057] Determine the pixel gray values ​​of all voxels in the T1-CE target image that belong to the region of interest as candidate pixel gray values;

[0058] The grayscale values ​​of the candidate pixels are sorted in descending order to obtain regions classified as high T1-CE voxel regions and low T1-CE voxel regions;

[0059] The pixel grayscale values ​​of each voxel in the high T1-CE voxel region are further sorted and classified in descending order according to the value of the surface diffusion coefficient ADC, and further classified into low ADC voxel regions and high ADC voxel regions, thereby obtaining a first spatial habitat of high enhancement-low ADC value and a fourth spatial habitat of high enhancement-high ADC value;

[0060] The pixel grayscale values ​​of each voxel in the above-mentioned low T1-CE voxel area are further sorted and classified in descending order according to the value of the surface diffusion coefficient ADC, and further classified into low ADC voxel areas and high ADC voxel areas, to obtain a third spatial habitat of low enhancement-low ADC value and a second spatial habitat of low enhancement-high ADC value.

[0061] S140. Construct a prediction model for predicting the IDH status of glioma based on several spatial habitats.

[0062] In this embodiment, due to the biomedical properties of the ADC target image and the T1-CE target image, and the fact that both types of images are MRI images, a number of spatial habitats are generated based on the MRI image representing the entire tumor area in combination with a clustering method.

[0063] In this embodiment, several prediction models can be constructed according to the characteristics of several spatial habitats. For example, if clustering generates 4 spatial habitats, four prediction models can be constructed, namely, a full tumor model, a 4-habitat combination model, a full tumor + 4 habitat combination model, and a sensitive habitat model.

[0064] In a specific implementation, S140 may include the following steps:

[0065] S1401 . Determine MRI images representing a plurality of spatial habitats and an MRI image representing the entire tumor region based on the ADC target image and the T1-CE target image.

[0066] The ADC target image and the T1-CE target image can be wavelet transformed respectively to obtain the ADC target image and the T1-CE target image after the wavelet transform, and the MRI images representing several spatial habitats and the MRI image representing the entire tumor area can be determined based on the ADC target image and the T1-CE target image before the wavelet transform and the ADC target image and the T1-CE target image after the wavelet transform.

[0067] S1402 . Extract features from the MRI images representing the plurality of spatial habitats and the MRI image representing the entire tumor region.

[0068] Among them, the features include first-order features, shape features, and texture features. The first-order features are used to quantitatively describe the voxel intensity distribution in the MRI image, the shape features are used to reflect the shape and size of the lesion area; and the texture features are used to reflect the homogeneity inside the tumor.

[0069] In one example, MRI images representing several spatial habitats and MRI images representing the entire tumor area may both include ADC target images and T1-CE target images after wavelet transformation. Specifically, wavelet filters are used to decompose the original ADC target images and T1-CE target images into several decomposition graphs, and first-order features, shape features, and texture features are extracted from each region of each decomposition graph. The Z-score normalization method is used to normalize each extracted feature so that each feature conforms to a standard normal distribution with a mean of 0 and a standard deviation of 1, so as to eliminate dimensional differences in the data and improve data comparability and analyticity.

[0070] In one embodiment, texture features may include at least five types, such as Gray Level Co-occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), Gray-Level Run-Length Matrix (GLRLM), Gray Level Size Zone Matrix (GLSZM), and Neighborhood Gray-tone Difference Matrix (NGTDM), etc.

[0071] S1403. Screen the features to obtain target features.

[0072] In this embodiment, in order to improve the accuracy and efficiency of subsequent model construction, it is necessary to identify and filter out irrelevant and redundant features, and finally obtain target features that can be used to build the model.

[0073] In one implementation, S1403 may include the following steps:

[0074] S14031. Use a two-sample t-test to select features that have significant differences between IDH wild-type cases and IDH mutant cases as the first candidate features. The significant difference means that the probability value in the t-test is less than a first preset threshold.

[0075] Among them, the first preset threshold may be 0.05.

[0076] S14032. Perform Pearson correlation analysis on each pair of first candidate features. If the absolute value of the correlation coefficient between the two first candidate features exceeds a second preset threshold, retain the first candidate feature with a smaller probability value in the t-test stage of significant difference among the two first candidate features as the second candidate feature.

[0077] The second preset threshold may be 0.9. That is, if the absolute value of the correlation coefficient between the two first candidate features exceeds 0.9, the feature with the smaller p-value in the significant difference t-test stage is retained; because the absolute value of the correlation coefficient exceeds 0.9, it can be indicated that the two first candidate features are highly correlated features, and highly correlated features may provide similar information, which may cause the model to overfit, so deleting one of the two can reduce redundancy.

[0078] S14033. Use the maximum relevance minimum redundancy algorithm to perform dimensionality reduction analysis on the second candidate feature to obtain a third candidate feature.

[0079] S14034. Perform a logistic regression analysis on the third candidate feature, and select the third candidate feature whose regression coefficient is not zero as the target feature.

[0080] For example, the least absolute shrinkage and selection operator (Lasso) with a penalty coefficient can be used to perform logistic regression analysis on the third candidate feature, and the optimal penalty parameter λ can be selected through 10-fold cross validation to select the third candidate feature with a non-zero regression coefficient as the target feature.

[0081] S1404. Construct a prediction model for predicting the IDH status of glioma based on the target features.

[0082] The embodiment of the present invention obtains an MRI image of a glioma patient, wherein the MRI image includes a DWI image and a T1-CE image; performs image preprocessing operations on the MRI image to obtain an ADC target image and a T1-CE target image with annotated regions of interest, wherein the ADC target image is converted from the DWI image, and the T1-CE target image with annotated regions of interest is converted from the T1-CE image; uses a clustering method to group pixel grayscale values ​​in the ADC target image and pixel grayscale values ​​in the region of interest in the T1-CE target image with annotated regions of interest, thereby obtaining a number of spatial habitats related to the glioma, wherein the spatial habitats are habitats displayed in the MRI image space; and constructs a prediction model for predicting the IDH state of glioma based on the number of spatial habitats. Among them, T1-CE image is a conventional T1 enhanced image in magnetic resonance imaging (MRI) image, which can enhance the display effect of tumors. DWI image is diffusion-weighted imaging in MRI image, which is often used to detect the diffusion movement of water molecules in the human body. Some malignant tumor cells grow densely and have less free water, which can easily lead to restricted diffusion in malignant tumors and form high signals, that is, DWI images can also be used to diagnose tumors. Based on T1-CE images and DWI images obtained from glioma patients, the images are preprocessed and the areas of interest of tumor tissues are marked. The ecological environmental parameters related to glioma blood vessels and cell density are visualized, and the clustering method is used to group the pixel grayscale values ​​in the image, which can divide different habitats and realize habitat imaging. It can be understood that habitat imaging is an imaging genomics technology based on biological environment. Tumor cell groups with similar characteristics are clustered by quantitative imaging markers, so that the internal heterogeneity of tumors can be displayed and quantified. This application uses several spatial habitats to construct a prediction model for predicting the IDH status of glioma. Subsequently, the volume proportion, average T1-CE value and average ADC value of gliomas in different IDH states in different habitats can be further calculated, and the habitats that are more sensitive to changes in the IDH status of gliomas can be further inferred. The prediction model constructed by multiple habitats will be more comprehensive in quantifying tumor heterogeneity, which can improve the accuracy of IDH mutation prediction, and ultimately achieve the purpose of accurately and non-invasively predicting the IDH status of gliomas.

[0083] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0084] Embodiment 2

[0085] Figure 2 A structural block diagram of a model building device based on MRI habitat imaging provided in the second embodiment of the present invention may specifically include the following modules:

[0086] A data acquisition module 210 is used to acquire MRI images of glioma patients, wherein the MRI images include DWI images and T1-CE images;

[0087] A data preprocessing module 220 is used to perform image preprocessing operations on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon, wherein the ADC target image is obtained by converting the DWI image, and the T1-CE target image with a region of interest marked thereon is obtained by converting the T1-CE image;

[0088] A clustering module 230 is used to group the grayscale values ​​of pixels in the ADC target image and the grayscale values ​​of pixels in the T1-CE target image marked with the region of interest by using a clustering method to obtain a plurality of spatial habitats related to glioma, wherein the spatial habitats are habitats displayed in the MRI image space;

[0089] The model building module 240 is used to build a prediction model for predicting the IDH status of glioma according to the plurality of spatial habitats.

[0090] In one embodiment of the present application, the data preprocessing module 220 may include:

[0091] A format conversion submodule, used for converting the native format of the MRI image into a target format to obtain a format-converted MRI image;

[0092] A diffusion gradient weighted selection submodule, used to determine, from the format-converted MRI image, a DWI image with a diffusion gradient weighted value of 1000 as the first DWI image, and a DWI image with a diffusion gradient weighted value of 0 as the second DWI image;

[0093] an ADC image determination submodule, configured to determine an ADC image according to the first DWI image and the second DWI image;

[0094] An image registration submodule, for performing three-dimensional rigid registration of the ADC image and the T1-CE image in the format-converted MRI image with a standard T1 anatomical template, respectively, and resampling based on the registration results to obtain a T1-CE three-dimensional image and an ADC three-dimensional image with a target resolution;

[0095] An image annotation submodule, used for annotating a region of interest on the T1-CE three-dimensional image to obtain a T1-CE three-dimensional image annotated with the region of interest;

[0096] The normalization submodule is used to normalize the ADC three-dimensional image and the T1-CE three-dimensional image with the region of interest marked, respectively, to obtain an ADC target image and a T1-CE target image with the region of interest marked.

[0097] In one embodiment of the present application, the normalization submodule may include:

[0098] The normalization unit is used to normalize the signal intensity of the ADC three-dimensional image and the signal intensity of the T1-CE three-dimensional image marked with the region of interest by using a kernel density estimation method to obtain an ADC target image and a T1-CE target image marked with the region of interest.

[0099] In one embodiment of the present application, the clustering module 230 may include:

[0100] A cluster number determination submodule is used to determine the target cluster number;

[0101] The clustering submodule is used to group the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​in the T1-CE target image marked with the region of interest using the K-means clustering method according to the target cluster number, so as to obtain several spatial habitats related to glioma.

[0102] In one embodiment of the present application, the ADC target image includes multiple layers, pixels located at the same position in all the layers constitute a voxel, and the ADC target image includes multiple voxels; the T1-CE target image marked with a region of interest includes multiple layers, pixels located at the same position in all the layers constitute a voxel, and the T1-CE target image includes multiple voxels; the cluster number determination submodule may include:

[0103] an error curve drawing unit, used for determining an error curve drawn by the sum of squares of errors between the voxel grayscale values ​​of each voxel in the ADC target image and the voxel grayscale values ​​of each voxel in the region of interest in the T1-CE target image;

[0104] A target cluster number determination unit is used to determine the target cluster number based on the error curve.

[0105] In one embodiment of the present application, the model building module 240 may include:

[0106] a target image determination submodule, configured to determine, based on the ADC target image and the T1-CE target image, MRI images representing the plurality of spatial habitats and an MRI image representing the entire tumor area;

[0107] A feature extraction submodule, for extracting features from the MRI images representing the plurality of spatial habitats and the MRI image representing the entire tumor area; the features include first-order features, shape features, and texture features, the first-order features are used to quantitatively describe the voxel intensity distribution in the MRI image, the shape features are used to reflect the shape and size of the lesion area, and the texture features are used to reflect the homogeneity phenomenon inside the tumor;

[0108] A feature screening submodule is used to screen the features to obtain target features;

[0109] The prediction model construction submodule is used to construct a prediction model for predicting the IDH status of glioma based on the target features.

[0110] In one embodiment of the present application, the feature screening submodule may include:

[0111] A first candidate feature determination unit is used to select a feature that has a significant difference between IDH wild-type cases and IDH mutant cases from the features using a two-sample t-test as a first candidate feature, wherein the significant difference is a probability value in the t-test that is less than a first preset threshold;

[0112] a second candidate feature determination unit, configured to perform a Pearson correlation analysis on the two first candidate features, and if the absolute value of the correlation coefficient between the two first candidate features exceeds a second preset threshold, retaining the first candidate feature with a smaller probability value in the t-test stage of the significant difference among the two first candidate features as the second candidate feature;

[0113] A third candidate feature determination unit, configured to perform dimensionality reduction analysis on the second candidate feature by using a maximum relevance minimum redundancy algorithm to obtain a third candidate feature;

[0114] The target feature determination unit is used to perform a logistic regression analysis on the third candidate feature and select the third candidate feature whose regression coefficient is not zero as the target feature.

[0115] The model building device provided in the embodiment of the present invention can execute the model building method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0116] Embodiment 3

[0117] Figure 3 A schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. Figure 3 A block diagram of an exemplary computer device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 3 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0118] like Figure 3 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0119] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0120] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0121] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 3 not shown, usually called a "hard drive"). Although Figure 3 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0122] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0123] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the computer device 12, and / or may communicate with any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the computer device 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the computer device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the computer device 12, including, but not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0124] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the model building method provided in the embodiment of the present invention.

[0125] Embodiment 4

[0126] Embodiment 4 of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the above-mentioned model building method are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0127] Among them, computer-readable storage media may include, for example, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.

[0128] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A model construction method based on MRI habitat imaging, characterized in that: include: Acquiring MRI images of a glioma patient, wherein the MRI images include DWI images and T1-CE images; Performing an image preprocessing operation on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon, wherein the ADC target image is obtained by converting the DWI image, and the T1-CE target image with a region of interest marked thereon is obtained by converting the T1-CE image; A clustering method is used to group the grayscale values ​​of pixels in the ADC target image and the grayscale values ​​of pixels in the T1-CE target image marked with the region of interest, so as to obtain a plurality of spatial habitats related to glioma, wherein the spatial habitats are habitats displayed in the MRI image space; A prediction model for predicting the IDH status of glioma is constructed based on the several spatial habitats.

2. The method according to claim 1, characterized in that The performing of image preprocessing operation on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon includes: Converting the native format of the MRI image into a target format to obtain a format-converted MRI image; Determining respectively from the format-converted MRI images a DWI image with a diffusion gradient weighted value of 1000 as a first DWI image and a DWI image with a diffusion gradient weighted value of 0 as a second DWI image; determining an ADC image according to the first DWI image and the second DWI image; Performing three-dimensional rigid registration of the ADC image and the T1-CE image in the format-converted MRI image with a standard T1 anatomical template, respectively, and resampling based on the registration results to obtain a T1-CE three-dimensional image and an ADC three-dimensional image with a target resolution; Annotating a region of interest on the T1-CE three-dimensional image to obtain a T1-CE three-dimensional image with the region of interest annotated; The ADC three-dimensional image and the T1-CE three-dimensional image with the region of interest marked are normalized respectively to obtain an ADC target image and a T1-CE target image with the region of interest marked.

3. The method according to claim 2, characterized in that The step of normalizing the ADC three-dimensional image and the T1-CE three-dimensional image with the region of interest marked thereon to obtain an ADC target image and a T1-CE target image with the region of interest marked thereon comprises: The signal intensity of the ADC three-dimensional image and the signal intensity of the T1-CE three-dimensional image with the region of interest marked are normalized respectively by using a kernel density estimation method to obtain an ADC target image and a T1-CE target image with the region of interest marked.

4. The method according to claim 1, characterized in that: The clustering method is used to group the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​in the T1-CE target image marked with the region of interest, to obtain several spatial habitats related to glioma, including: Determine the target number of clusters; According to the target cluster number, the K-means clustering method is used to group the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​belonging to the region of interest in the T1-CE target image marked with the region of interest, so as to obtain several spatial habitats related to glioma.

5. The method according to claim 4, characterized in that The ADC target image includes a plurality of layers, and pixels located at the same position in all the layers constitute a voxel, and the ADC target image includes a plurality of the voxels; the T1-CE target image with annotated regions of interest includes a plurality of layers, and pixels located at the same position in all the layers constitute a voxel, and the T1-CE target image includes a plurality of the voxels; determining the number of target clusters includes: Determine an error curve formed by plotting the sum of squares of errors between the voxel grayscale value of each voxel in the ADC target image and the voxel grayscale value of each voxel located in the region of interest in the T1-CE target image; A target number of clusters is determined based on the error curve.

6. The method according to claim 1, characterized in that The prediction model for predicting the IDH status of glioma is constructed according to the plurality of spatial habitats, comprising: Determine, based on the ADC target image and the T1-CE target image, an MRI image representing the plurality of spatial habitats and an MRI image representing the entire tumor region; Extracting features from the MRI images representing the plurality of spatial habitats and the MRI image representing the entire tumor area; the features include first-order features, shape features, and texture features, the first-order features are used to quantitatively describe the voxel intensity distribution in the MRI image, the shape features are used to reflect the shape and size of the lesion area; and the texture features are used to reflect the homogeneity phenomenon inside the tumor; Screening the features to obtain target features; A prediction model for predicting the IDH status of glioma is constructed based on the target features.

7. The method according to claim 1, characterized in that The screening of the features to obtain target features includes: Using a two-sample t-test, a feature that is significantly different between IDH wild-type cases and IDH mutant cases is selected from the features as a first candidate feature, wherein the significant difference is a probability value in the t-test that is less than a first preset threshold; Performing Pearson correlation analysis on the two first candidate features in pairs, if the absolute value of the correlation coefficient between the two first candidate features exceeds a second preset threshold, retaining the first candidate feature with a smaller probability value in the t-test stage of the significant difference among the two first candidate features as the second candidate feature; Using a maximum relevance minimum redundancy algorithm to perform dimensionality reduction analysis on the second candidate feature to obtain a third candidate feature; Perform a logistic regression analysis on the third candidate feature, and select the third candidate feature whose regression coefficient is not zero as the target feature.

8. A model building device based on MRI habitat imaging, characterized in that: include: A data acquisition module, used to acquire MRI images of glioma patients, wherein the MRI images include DWI images and T1-CE images; A data preprocessing module, used for performing image preprocessing operation on the MRI image to obtain an ADC target image and a T1-CE target image with a region of interest marked thereon, wherein the ADC target image is obtained by converting the DWI image, and the T1-CE target image with a region of interest marked thereon is obtained by converting the T1-CE image; A clustering module, for grouping the pixel grayscale values ​​in the ADC target image and the pixel grayscale values ​​in the T1-CE target image marked with the region of interest by a clustering method, to obtain a plurality of spatial habitats related to glioma, wherein the spatial habitats are habitats displayed in the MRI image space; A model building module is used to build a prediction model for predicting the IDH status of glioma based on the multiple spatial habitats.

9. A computer device, characterized in that: The computer device comprises: 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 model building method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the model building method according to any one of claims 1 to 7 is implemented.

Images