Medical image processing method and device, computer equipment and readable storage medium

Abstract

The invention relates to a medical image processing method and device, computer equipment and a computer readable storage medium. The method comprises the following steps: acquiring a magnetic resonance image of glioma; segmenting a region of interest corresponding to the glioma from the magnetic resonance image according to a pre-trained segmentation model; extracting radiomics characteristics of the region of interest; and inputting the radiomics characteristics into a predetermined target classifier to obtain a typing result of the glioma. By adopting the method, the glioma typing efficiency can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a medical image processing method, apparatus, computer device, and computer-readable storage medium. Background Technology

[0002] Gliomas are common primary intracranial tumors originating from glial cells in the brain. Statistics show an annual incidence of 5-8 per 100,000, and their 5-year mortality rate is second only to pancreatic and lung cancer among all cancers. Given that glioma diagnosis and classification can provide crucial clues for patient treatment selection and prognostic assessment, reliable diagnostic classification methods have received widespread attention.

[0003] In existing technologies, molecular detection can serve as an important indicator for glioma subtyping. Molecular detection techniques mainly include immunohistochemical staining, gene sequencing, and in situ hybridization. However, due to the large amount of manpower and time required, as well as the need for specialized detection equipment, the efficiency of molecular detection techniques is not high. Summary of the Invention

[0004] Therefore, it is necessary to provide a medical image processing method, device, computer equipment, computer-readable storage medium, and computer program product that can improve detection efficiency in response to the above-mentioned technical problems.

[0005] In a first aspect, this application provides a medical image processing method, comprising:

[0006] Obtain magnetic resonance images of gliomas;

[0007] The region of interest corresponding to the glioma is segmented from the magnetic resonance image based on a pre-trained segmentation model;

[0008] Extract the image omics features of the region of interest;

[0009] The radiomics features are input into a pre-defined target classifier to obtain the glioma classification result.

[0010] In one embodiment, segmenting the region of interest corresponding to the glioma from the magnetic resonance image based on a pre-trained segmentation model includes:

[0011] The magnetic resonance images are registered according to a pre-set image template to obtain the registered magnetic resonance images.

[0012] The resolution and grayscale value of the registered magnetic resonance image are adjusted to obtain the preprocessed magnetic resonance image.

[0013] The preprocessed magnetic resonance image is input into the pre-trained segmentation model to segment the region of interest from the preprocessed magnetic resonance image.

[0014] In one embodiment, prior to acquiring the magnetic resonance image of the glioma, the method further includes:

[0015] Obtain magnetic resonance image samples and corresponding sample labels for the magnetic resonance image samples; each magnetic resonance image sample corresponds to at least one sample type; the sample label is a region in the magnetic resonance image sample that includes tumors and edema.

[0016] Magnetic resonance image samples of different sample types are combined into multi-channel samples, and the multi-channel samples are input into the segmentation model to be trained to obtain the sample segmentation results of the multi-channel samples;

[0017] Based on the difference between the sample segmentation result and the sample label, the segmentation model to be trained is trained to obtain the pre-trained segmentation model.

[0018] In one embodiment, extracting the image omics features of the region of interest includes:

[0019] In the registered magnetic resonance image, the region outside the region of interest is defined as the reference region;

[0020] Determine the histogram mapping parameters between the reference region and the image template;

[0021] Based on the histogram mapping parameters, the registered magnetic resonance image is mapped to obtain a standard magnetic resonance image.

[0022] Based on the standard magnetic resonance imaging, the radiomics features of the region of interest are extracted.

[0023] In one embodiment, extracting the radiomics features of the region of interest based on the standard magnetic resonance image includes:

[0024] Based on the region of interest, the standard region of interest of the standard magnetic resonance image is determined;

[0025] Extract at least one of the first-order statistical features, two-dimensional shape features, three-dimensional shape features, and gray-scale matrix features from the standard region of interest to obtain the target extraction features of the standard region of interest.

[0026] The extracted features of the target are standardized to obtain the image omics features.

[0027] In one embodiment, prior to acquiring the magnetic resonance image of the glioma, the method further includes:

[0028] Obtain the region of interest (ROI) in the magnetic resonance image sample, and obtain the reference region based on the magnetic resonance image samples outside the ROI;

[0029] Based on the sample reference region, the magnetic resonance image sample is standardized to obtain a standard magnetic resonance image sample;

[0030] Based on the standard magnetic resonance image samples, extract the sample image omics features of the region of interest in the samples;

[0031] Each candidate classifier in at least one candidate classifier is used to classify the sample image omics features;

[0032] Based on the classification results, the target classifier is determined from the at least one candidate classifier.

[0033] In one embodiment, the sample radiomics features correspond to at least one sample radiomics feature; the step of extracting the sample radiomics features of the region of interest based on the standard magnetic resonance image sample includes:

[0034] If the sample imaging features are missing, then the median corresponding to the sample imaging features is obtained;

[0035] The sample image omics features are obtained based on the median.

[0036] Secondly, this application also provides a medical image processing device, comprising:

[0037] The acquisition module is used to acquire magnetic resonance images of gliomas;

[0038] The segmentation module is used to segment the region of interest corresponding to the glioma from the magnetic resonance image based on a pre-trained segmentation model.

[0039] The extraction module is used to extract the radiomics features of the region of interest;

[0040] The classification module is used to input the radiomics features into a pre-determined target classifier to obtain the classification result of the glioma.

[0041] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0042] Obtain magnetic resonance images of gliomas;

[0043] Based on a pre-trained segmentation model, the region of interest corresponding to the glioma is segmented from the magnetic resonance image;

[0044] Extract the image omics features of the region of interest;

[0045] The radiomics features are input into a pre-defined target classifier to obtain the glioma classification result.

[0046] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0047] Obtain magnetic resonance images of gliomas;

[0048] Based on a pre-trained segmentation model, the region of interest corresponding to the glioma is segmented from the magnetic resonance image;

[0049] Extract the image omics features of the region of interest;

[0050] The radiomics features are input into a pre-defined target classifier to obtain the glioma classification result.

[0051] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0052] Obtain magnetic resonance images of gliomas;

[0053] Based on a pre-trained segmentation model, the region of interest corresponding to the glioma is segmented from the magnetic resonance image;

[0054] Extract the image omics features of the region of interest;

[0055] The radiomics features are input into a pre-defined target classifier to obtain the glioma classification result.

[0056] The aforementioned medical image processing methods, devices, computer equipment, computer-readable storage media, and computer program products acquire magnetic resonance images of gliomas, segment the corresponding regions of interest (ROIs) from the ROIs based on a pre-trained segmentation model, extract radiomics features of the ROIs, and input the radiomics features into a pre-defined target classifier to obtain the glioma classification result. This allows for the efficient segmentation of ROIs from ROIs and the classification of gliomas based on their radiomics features. Since molecular detection is unnecessary, classification can be performed solely through medical images, thus improving the efficiency of glioma classification. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating a medical image processing method in one embodiment;

[0059] Figure 2 This is a flowchart illustrating a glioma image prediction method in one embodiment;

[0060] Figure 3 This is a flowchart illustrating a medical image processing method in another embodiment;

[0061] Figure 4 This is a structural block diagram of a medical image processing device in one embodiment;

[0062] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0065] In one exemplary embodiment, such as Figure 1 As shown, a medical image processing method is provided. This embodiment illustrates the method applied to a terminal. It is understood that this method can also be applied to a server, and furthermore, to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0066] Step S102: Obtain magnetic resonance images of the glioma.

[0067] The magnetic resonance imaging may include, but is not limited to, T1 (longitudinal relaxation time) sequence images, T2 (lateral relaxation time) sequence images, T2 Flair (Fluid Attenuated Inversion Recovery) sequence images, and T1ce (T1-weighted Contrast Enhanced) sequence images.

[0068] In practice, a magnetic resonance imaging (MRI) device can be used to scan glioma patients to obtain MRI images of the glioma, which are then input into a terminal.

[0069] In practical applications, after the terminal acquires the magnetic resonance image of the glioma, it can preprocess the magnetic resonance image. For example, the magnetic resonance image can be registered according to a pre-set anatomical template, and the background, skull, scalp and other tissue structures can be removed from the magnetic resonance image. The resolution of the magnetic resonance image can be adjusted to a specified resolution. The magnetic resonance image can be normalized by first subtracting the average gray value of the brain region and then dividing by the standard deviation. Finally, the preprocessed magnetic resonance image can be obtained.

[0070] Step S104: Based on the pre-trained segmentation model, segment the region of interest corresponding to the glioma from the magnetic resonance image.

[0071] The segmentation model can be a model for segmenting the region of interest from magnetic resonance imaging. The region of interest can be the glioma body and the peritumoral edema area.

[0072] In practice, the terminal can pre-train a segmentation model, input the magnetic resonance image into the segmentation model, and segment the magnetic resonance image through the segmentation model to obtain the region of interest corresponding to the glioma.

[0073] In practical applications, a segmentation model based on UNet (U-shaped network) can be pre-trained. The pre-processed magnetic resonance images can be input into the UNet-based segmentation model to segment the glioma body and peritumoral edema area from the pre-processed magnetic resonance images.

[0074] Step S106: Extract the image omics features of the region of interest.

[0075] In practice, the terminal can extract features from the region of interest in the magnetic resonance image to obtain first-order statistical features, two-dimensional shape features, three-dimensional shape features, gray-scale matrix features, etc. The first-order statistical features, two-dimensional shape features, three-dimensional shape features, gray-scale matrix features, etc. are then standardized to obtain the radiomics features of the region of interest.

[0076] In practical applications, after segmenting the glioma and peritumoral edema area from preprocessed MRI images, the registered MRI images can be standardized. For example, the regions containing gray matter, white matter, cerebrospinal fluid, and brainstem can be segmented from the registered MRI images. The mapping parameters between the gray-level histogram of this region and the gray-level histogram of the anatomical template can be determined. Based on these mapping parameters, the registered MRI images are mapped to obtain standardized MRI images. Then, radiomics features are extracted from the glioma and peritumoral edema areas in the standardized MRI images.

[0077] Step S107: Input the radiomics features into a pre-determined target classifier to obtain the glioma classification results.

[0078] The target classifier can be a classifier used to predict glioma type. The classification result can be, but is not limited to, glioma types such as glioblastoma, astrocytoma, and oligodendroglioma.

[0079] In practice, the terminal can predetermine the target classifier, input the radiomics features into the target classifier, and classify the radiomics features through the target classifier to obtain the corresponding subtype of glioma.

[0080] In practical applications, the terminal can pre-store classifiers such as support vector machines, logistic regression, decision trees, random forests, and XGBoost (a distributed gradient boosting library), and automatically select the best-performing classifier as the target classifier. The extracted radiomics features are then input into the target classifier to automatically determine the glioma classification result.

[0081] The aforementioned medical image processing method acquires magnetic resonance images of gliomas, segments the corresponding region of interest (ROI) from the ROI based on a pre-trained segmentation model, extracts the radiomics features of the ROI, and inputs the radiomics features into a pre-defined target classifier to obtain the glioma classification result. It can effectively segment the ROI from the ROI of gliomas and classify the gliomas based on the radiomics features of the ROI. Since molecular detection is not required, classification can be performed solely through medical images, thus improving the efficiency of glioma classification.

[0082] In an exemplary embodiment, step S104 may specifically include: registering the magnetic resonance image according to a pre-set image template to obtain a registered magnetic resonance image; adjusting the resolution and grayscale value of the registered magnetic resonance image to obtain a pre-processed magnetic resonance image; and inputting the pre-processed magnetic resonance image into a pre-trained segmentation model to segment the region of interest from the pre-processed magnetic resonance image.

[0083] The image template can be an anatomical image of normal tissue without glioma.

[0084] In practice, the terminal can register the magnetic resonance image to a pre-set image template to obtain the registered magnetic resonance image. It can also adjust the resolution of the registered magnetic resonance image to a preset resolution and normalize the grayscale values ​​to obtain the pre-processed magnetic resonance image. The terminal can pre-train a segmentation model and input the pre-processed magnetic resonance image into the pre-trained segmentation model to obtain the region of interest in the pre-processed magnetic resonance image.

[0085] In practical applications, MRI images can be registered to a pre-defined anatomical template, removing background, skull, scalp, and other tissue structures. Rigid registration can be used to align MRI images of different modalities, and the resolution can be adjusted to a preset resolution, such as 1 mm. Alternatively, the MRI images can be normalized by subtracting the average grayscale value of the brain region and then dividing by the standard deviation. The terminal can pre-train a UNet-based segmentation model, inputting the pre-processed MRI images into this model to automatically segment the glioma body and peritumoral edema area within the MRI images.

[0086] In this embodiment, the magnetic resonance image is registered according to a pre-set image template to obtain a registered magnetic resonance image. The resolution and grayscale value of the registered magnetic resonance image are adjusted to obtain a pre-processed magnetic resonance image. The pre-processed magnetic resonance image is then input into a pre-trained segmentation model to segment the region of interest from the pre-processed magnetic resonance image. By pre-processing the magnetic resonance image first and then segmenting it according to the pre-trained segmentation model, the accuracy of the segmented region of interest can be improved, thereby increasing the reliability of the classification results.

[0087] In an exemplary embodiment, prior to step S102, the method may further include: obtaining magnetic resonance image samples and corresponding sample labels; each magnetic resonance image sample corresponds to at least one sample type; the sample label is a region in the magnetic resonance image sample including tumors and edema; combining magnetic resonance image samples of different sample types into multi-channel samples; inputting the multi-channel samples into a segmentation model to be trained to obtain the sample segmentation results of the multi-channel samples; and training the segmentation model to be trained based on the difference between the sample segmentation results and the sample labels to obtain a pre-trained segmentation model.

[0088] The magnetic resonance imaging (MRI) samples can be multiple MRI images obtained from multiple MRI scans performed on at least one glioma patient, used for model training of a segmentation model. Sample labels can be manually annotated MRI images of the tumor and peritumoral edema areas. Sample types can include T1-series, T2-series, T2 Flair-Track, or T1ce-series images. Multichannel samples can be samples synthesized by combining at least two of the following sequences: T1-series, T2-series, T2 Flair-Track, and T1ce-series images.

[0089] In practice, the terminal can acquire multiple magnetic resonance imaging samples and their labels. Different magnetic resonance imaging samples may belong to different sample types. Magnetic resonance imaging samples of different sample types are combined into multi-channel samples. The multi-channel samples are input into the segmentation model to be trained to obtain the sample segmentation results. The sample segmentation results are compared with the sample labels. Based on the difference between the sample segmentation results and the sample labels, the model parameters of the segmentation model to be trained are adjusted. The above process is repeated until the preset number of iterations is reached, or the iteration converges (for example, the difference between the sample segmentation results and the sample labels is calculated, and the difference between two adjacent differences is less than a preset value), thus obtaining the pre-trained segmentation model.

[0090] In practical applications, taking dual-channel as an example, T1ce and T2 Flair image samples can be acquired and synthesized into dual-channel samples, which are then input into the UNet network to be trained. The UNet network has an encoder-decoder structure, where the encoder can downsample 5 times using convolution operations, and correspondingly, the decoder follows the same structure, upsampling 5 times using transposed convolutions. The encoder's feature maps are then concatenated to the corresponding level's decoder feature maps via skip links, forming richer feature information. The feature maps are normalized, and Leaky ReLU (Leakage Corrected Linear Unit) is used as the activation function to calculate the UNet network's output. The initial number of convolutional filters is 32, with a maximum of 320 at the top layer. The loss function is defined as the sum of cross-entropy loss and Dice loss. Network training is stopped after a fixed number of epochs or multiple consecutive training sessions when performance no longer improves, resulting in the final segmentation model.

[0091] In this embodiment, by acquiring magnetic resonance image samples and their corresponding sample labels, magnetic resonance image samples of different types are synthesized into multi-channel samples. The multi-channel samples are then input into the segmentation model to be trained to obtain the sample segmentation results of the multi-channel samples. Based on the difference between the sample segmentation results and the sample labels, the segmentation model to be trained is trained to obtain a pre-trained segmentation model. The segmentation model can be pre-trained and used to automatically segment magnetic resonance images, thereby improving segmentation efficiency and accuracy.

[0092] In an exemplary embodiment, step S106 may specifically include: in the registered magnetic resonance image, determining the region outside the region of interest as a reference region; determining the histogram mapping parameters between the reference region and the image template; performing mapping processing on the registered magnetic resonance image according to the histogram mapping parameters to obtain a standard magnetic resonance image; and extracting the radiomics features of the region of interest according to the standard magnetic resonance image.

[0093] The reference region can be the region used for histogram equalization. The histogram mapping parameters can be the mapping parameters between the histogram of the reference region and the histogram of the image template. The standard magnetic resonance image can be the magnetic resonance image after histogram equalization.

[0094] In specific implementation, the terminal can define the region outside the region of interest in the registered magnetic resonance image as the reference region, and fit the gray-level distribution histogram of the reference region with a normal distribution to obtain the normal distribution fitting result of the gray-level distribution histogram of the reference region, denoted as the first fitting result. The terminal can also perform similar processing on the image template, that is, fit the gray-level distribution histogram of the region outside the region of interest in the image template with a normal distribution to obtain the normal distribution fitting result of the gray-level distribution histogram of the region outside the region of interest in the image template, denoted as the second fitting result. The mapping parameter between the second fitting result and the first fitting result is determined as the histogram mapping parameter between the reference region and the image template. The registered magnetic resonance image is mapped according to the histogram mapping parameter to obtain the standard magnetic resonance image. Then, the terminal can perform feature extraction on the standard magnetic resonance image to obtain the radiomics features of the region of interest.

[0095] In practical applications, a trained segmentation model can be used to segment the tumor and peritumoral edema region from the registered MRI image. The tumor and peritumoral edema region are then excluded from the registered MRI image, resulting in the regions containing gray matter, white matter, cerebrospinal fluid, and the brainstem. The intensity distribution of these regions is statistically analyzed to obtain intensity distribution histograms for the gray matter, white matter, cerebrospinal fluid, and brainstem regions. These histograms are then fitted using a normal distribution model to obtain the first fitting result. The same operation can then be performed on an anatomical template to obtain a normal distribution fitting result for the intensity distribution histograms of the gray matter, white matter, cerebrospinal fluid, and brainstem regions in the anatomical template, denoted as the second fitting result. Finally, the mapping parameters between the first and second fitting results are determined. Then, the tumor and peritumoral edema areas can be taken into account. The registered magnetic resonance images can be mapped using mapping parameters to obtain standard magnetic resonance images. Radiomic features can be extracted from the tumor and peritumoral edema areas in the standard magnetic resonance images. Specifically, first-order statistical features, two-dimensional shape features, three-dimensional shape features, gray-level matrix features, etc., of the tumor and peritumoral edema areas can be extracted. Z-scores are then used to standardize each feature to obtain radiomic features.

[0096] In this embodiment, by defining the region outside the region of interest as the reference region in the registered magnetic resonance image, determining the histogram mapping parameters between the reference region and the image template, and performing mapping processing on the registered magnetic resonance image according to the histogram mapping parameters to obtain a standard magnetic resonance image, and extracting the radiomics features of the region of interest from the standard magnetic resonance image, it is possible to extract features from the registered and histogram-equalized magnetic resonance image, accurately extract the radiomics features of the magnetic resonance image, and improve the accuracy of classification.

[0097] In an exemplary embodiment, the step of extracting radiomics features of a region of interest from a standard magnetic resonance image may specifically include: determining a standard region of interest from the standard magnetic resonance image based on the region of interest; extracting at least one of first-order statistical features, two-dimensional shape features, three-dimensional shape features, and gray-scale matrix features from the standard region of interest to obtain target extraction features of the standard region of interest; and standardizing the target extraction features to obtain radiomics features.

[0098] The standard region of interest (ROI) can be the tumor and peritumoral edema area in a standard magnetic resonance imaging (MRI) image. The target extraction features can be the features extracted from the standard ROI.

[0099] In practice, since standard magnetic resonance images are obtained by processing registered magnetic resonance images, the region of interest (ROI) in the registered ROI can be defined as the standard ROI of the standard ROI. One or more of the following features—first-order statistical features, two-dimensional shape features, three-dimensional shape features, and grayscale matrix features—are extracted from the standard ROI to obtain target extraction features. The extracted target extraction features are then standardized using Z-scores to obtain the radiomics features of the standard ROI. It is understood that, for different types of sequences or images, the target extraction features are not limited to the first-order statistical features, two-dimensional shape features, three-dimensional shape features, and grayscale matrix features mentioned above.

[0100] In this embodiment, a standard region of interest (ROI) is determined based on the region of interest (ROI). At least one of the first-order statistical features, two-dimensional shape features, three-dimensional shape features, and gray-scale matrix features of the standard ROI is extracted to obtain the target extraction features of the standard ROI. The target extraction features are then standardized to obtain radiomics features. Radiomics features can be extracted from ROIs that have been registered and histogram equalized. Using radiomics features for typing can further improve the accuracy of typing.

[0101] In an exemplary embodiment, prior to step S102, the method may further include: obtaining a region of interest (ROI) from a magnetic resonance imaging (MRI) sample; obtaining a reference region from MRI samples outside the ROI; standardizing the MRI samples based on the reference region to obtain a standard MRI sample; extracting image omics features from the ROI based on the standard MRI sample; determining the classification result of each candidate classifier among at least one candidate classifier for classifying the image omics features; and determining a target classifier from the at least one candidate classifier based on the classification result.

[0102] The region of interest (ROI) for the sample can be the tumor and peritumoral edema area in the MRI sample. The reference region for the sample can be the region used for histogram equalization, such as cerebrospinal fluid, brainstem, gray matter, and white matter in the MRI sample. The standard MRI sample can be an MRI sample that has undergone histogram equalization. The radiomics features of the sample can be the radiomics features of the MRI sample. Candidate classifiers include, but are not limited to, support vector machines, logistic regression, decision trees, random forests, and XGBoost. The classification result can be the predicted result of glioma subtyping.

[0103] In practice, the terminal can sequentially input each magnetic resonance imaging (MRI) image sample into a pre-trained segmentation model. For each MRI image sample, a corresponding region of interest (ROI) can be obtained through the segmentation model. MRI images outside the ROI are defined as reference regions. Based on these reference regions, the MRI images are standardized (histogram equalization) to obtain standard MRI images. Feature extraction is performed on the ROI of these standard MRI images to obtain a set of image omics features. Each of at least one candidate classifier is used to classify these image omics features, resulting in a classification result for each candidate classifier. This process is repeated for each MRI image sample, yielding multiple classification results for each candidate classifier. Based on these multiple classification results, a target classifier is selected from at least one candidate classifier. In practical applications, the candidate classifier with the highest accuracy can be selected as the target classifier based on the classification results.

[0104] In this embodiment, the region of interest (ROI) in the magnetic resonance imaging (MRI) sample is obtained. Based on the MRI samples outside the ROI, a reference region is obtained. Based on the reference region, the MRI samples are standardized to obtain standard MRI samples. Based on the standard MRI samples, the ROI features are extracted. Each candidate classifier in at least one candidate classifier is determined. The classification results of the ROI features are obtained. Based on the classification results, the target classifier is determined from at least one candidate classifier. This can identify a target classifier with better performance and improve the accuracy of the classification results.

[0105] In an exemplary embodiment, the above-mentioned sample radiomics features correspond to at least one sample feature; the step of extracting sample radiomics features of the region of interest of a sample based on a standard magnetic resonance image sample may specifically include: if a sample feature is missing, obtaining the median corresponding to the sample feature; and obtaining the sample radiomics features based on the median.

[0106] Among them, the sample features can be any one of the following: first-order statistical features, two-dimensional shape features, three-dimensional shape features, and gray-scale matrix features.

[0107] In practice, for each magnetic resonance imaging sample, a set of sample radiomics features can be obtained. The sample radiomics features include first-order statistical features, two-dimensional shape features, three-dimensional shape features, gray-scale matrix features, and other sample features. If any sample feature is missing, the median of that sample feature among all magnetic resonance imaging samples other than that magnetic resonance imaging sample can be used as the sample feature of that magnetic resonance imaging sample, thereby obtaining the sample radiomics features of that magnetic resonance imaging sample.

[0108] For example, suppose there are n magnetic resonance imaging (MRI) images, where the first-order statistical characteristics of the 1st to (n-1th)th MRI images are s1, s2, ..., sn2, ..., sn3, respectively. n-1 If the first-order statistical features of the nth magnetic resonance imaging sample are missing, then s1, s2, ..., s n-1 The median was determined as the first-order statistical characteristic of the nth magnetic resonance image sample.

[0109] In this embodiment, if sample features are missing, the median corresponding to the sample features is obtained, and the sample radiomics features are obtained based on the median. This allows the sample radiomics features to still be obtained even when sample features are missing, thus avoiding inaccurate classifier training due to missing sample features.

[0110] To facilitate a deeper understanding of the embodiments of this application by those skilled in the art, a specific example will be used for illustration below.

[0111] This application provides a method for predicting glioma images. By segmenting gliomas, extracting MRI image information of tumors and peritumoral edema, and combining this with molecular diagnostic results, a model of glioma molecular subtypes is constructed, and glioma subtype prediction is performed based on the images, achieving efficient, accurate, and intuitive diagnostic results.

[0112] Specifically, this application, based on the World Health Organization (WHO) 2021 classification of central nervous system tumors, achieves the identification and diagnosis of the new classification of diffuse gliomas in adults. Compared with existing technologies that study glioblastoma or single genotypes of gliomas, this application uses more comprehensive data sources and yields more universally applicable conclusions. This application uses clinical magnetic resonance imaging (MRI) sequences for automated segmentation and subtype prediction of gliomas. Compared with existing methods that use pathological images to predict molecular subtypes, this improves diagnostic performance while saving doctors time spent analyzing and interpreting histopathological slides. This application utilizes conventional MRI sequences (T1ce, T2...) The MRI signal standardization method can non-invasively predict glioma molecular subtypes, reducing imaging costs and shortening imaging time. For patients who cannot acquire comprehensive MRI multimodal data, molecular subtype diagnosis of glioma can be performed using only basic MRI scan images. This application also proposes an automatic calculation method for MRI signal value standardization. This method can overcome the problem of large differences in the range of image values ​​among different patients and obtain stable numerical characteristics. Based on the image with standardized signal values, more stable prediction results can be obtained. It can also quantitatively compare the nature of edema areas and analyze prognosis.

[0113] refer to Figure 2This application proposes a method and system for predicting the subtype of diffuse gliomas (hereinafter referred to as gliomas) in adults based on conventional MRI sequences. It utilizes the radiological characteristics of MRI images to predict glioma subtypes (glioblastoma, astrocytoma, and oligodendroglioma) according to the WHO 2021 classification, providing important guidance for clinical glioma subtype diagnosis. Specifically, this application includes two parts: glioma segmentation and glioma subtype prediction. For the glioma segmentation task, it includes image preprocessing, UNet segmentation model establishment, and tumor segmentation modules; for the glioma subtype prediction task, it includes image relative signal value normalization, feature extraction, and classifier establishment modules.

[0114] In practical applications, the first step is to acquire imaging data from glioma patients. This data includes standard MRI sequences, which should include T1, T2, T2 Flair, or T1ce sequences. The tumor and peritumoral edema areas in the imaging data are labeled and confirmed by the physician, ultimately forming annotation labels. These labels include the tumor core area and the peritumoral edema / infiltrated tissue area.

[0115] For glioma segmentation tasks, the following processing steps may be included:

[0116] Step S211, Image Preprocessing. All image data are registered to the same anatomical template, background, skull, scalp, etc., are removed, and rigid registration is used to align images of different modalities. The resolution is adjusted to 1mm. For each image, normalization is performed by subtracting the average grayscale value of the brain region and then dividing by the standard deviation.

[0117] Step S212: UNet segmentation model establishment. The T1ce and T2 Flair images are synthesized into a dual-channel image and used as input to train the UNet network. The network used in this application has an encoder-decoder structure. The encoder downsamples 5 times using convolution operations, and correspondingly, the decoder follows the same structure, upsampling 5 times using transposed convolutions. Feature maps from the encoder are concatenated to corresponding decoder feature maps at the same level via skip links, forming richer feature information. Instance normalization is used to normalize the feature maps, and Leaky ReLU is used as the activation function to calculate the output. The initial number of convolutional filters is 32, and the maximum number of convolutional filters at the top layer is 320. The loss function is defined as the sum of cross-entropy loss and Dice loss. Network training is stopped after a fixed number of epochs or multiple consecutive training sessions when performance no longer improves, resulting in the final segmentation model.

[0118] Step S213, Tumor Segmentation. The constructed segmentation model is applied to the glioma MRI image to segment the tumor and the peritumoral edema area.

[0119] The following processing steps may be included in the task of predicting glioma subtypes:

[0120] Step S221, image signal value standardization. This involves three steps: First, based on the results of registration with the template in the previous steps, gray matter, white matter, cerebrospinal fluid, and brainstem are segmented. Then, the segmentation model obtained above is used to segment the tumor and peritumoral edema regions. Next, after excluding the tumor and peritumoral edema regions, the intensity distribution histograms of gray matter, white matter, cerebrospinal fluid, and brainstem are obtained respectively, and fitted with a normal distribution model. The same operation is performed on the template image. Finally, the histograms of the image's gray matter, white matter, and other structures are matched with the histograms of the corresponding structures in the template to obtain mapping parameters. Taking the tumor and peritumoral edema regions into account, the mapping parameters are used to map the target image's gray matter, white matter, cerebrospinal fluid, and brainstem separately to obtain the standardized image.

[0121] Step S222, Feature Extraction. In the standardized image obtained above, radiomics features were extracted from the segmented tumor core and whole tumor (tumor core and edema) regions, including first-order statistical features, two- and three-dimensional shape features, gray-scale matrix features, etc. Z-scores were used to standardize each feature. For data where there may be missing feature values, the median was used instead.

[0122] Step S223: Classifier Construction. The input data consists of radiomics features of the tumor core and the entire tumor region, labeled with the WHO 2021 glioma subtype. Any classifier can be used, such as Support Vector Machine (SVM), Logistic Regression, Decision Tree, Random Forest, or XGBoost. The model with the highest average cross-validation performance is selected as the final subtyping classifier using a hyperparameter random search method and 5-fold cross-validation.

[0123] In one specific embodiment, a system for predicting glioma subtypes using features of multimodal MRI sequence images of gliomas is provided.

[0124] In another embodiment, a non-invasive model training system for predicting glioma subtypes is also provided. This system includes: a data acquisition module, an image preprocessing module, a tumor segmentation module, a feature extraction module, a classification training module, and a model evaluation module; wherein,

[0125] The data acquisition module is used to acquire MRI images and glioma subtype information of glioma patients, including T1 sequence images, T2 sequence image data, T2 Flair sequence image data, and T1ce sequence image data.

[0126] The image preprocessing module is used to preprocess the patient's MRI multimodal image data to obtain a normalized image after registration and alignment;

[0127] The tumor segmentation module is used to obtain the tumor core region and peritumoral edema region under each modality of MRI image data without processing multimodal MRI image data.

[0128] The image standardization module is used to standardize the signal values ​​of image data to obtain a standardized image, which is then used for subsequent feature extraction calculations.

[0129] The feature extraction module takes the data from the tumor segmentation module as input data and extracts radiomics features from the image data of each region, including first-order statistics, shape-based (3D), gray-level co-occurrence matrix, gray-level run length matrix, gray-level size region matrix, adjacent gray-level tone difference matrix, gray-level dependency matrix and other feature information.

[0130] The classification training module uses radiomics features as input data and glioma subtype information as label information. It can be trained using various classifiers, including Support Vector Machines, Logistic Regression, Decision Trees, Random Forests, and XGBoost. The trained model can be evaluated, yielding results including the Area Under the Receiver Operating Characteristic (AUC), accuracy, sensitivity, and specificity. Based on the evaluation results, the network with the best performance is selected as the glioma subtype prediction model.

[0131] In one embodiment, such as Figure 3 As shown, a medical image processing method is provided. Taking the application of this method to a terminal as an example, the method includes the following steps:

[0132] Step S301: Obtain magnetic resonance image samples and corresponding sample labels, synthesize magnetic resonance image samples of different sample types into multi-channel samples, input the multi-channel samples into the segmentation model to be trained, obtain the sample segmentation results of the multi-channel samples, and train the segmentation model to be trained based on the difference between the sample segmentation results and the sample labels to obtain the pre-trained segmentation model.

[0133] Step S302: Obtain the region of interest (ROI) in the magnetic resonance image sample; obtain the reference region based on the magnetic resonance image sample outside the ROI; standardize the magnetic resonance image sample based on the reference region to obtain a standard magnetic resonance image sample; extract the image omics features of the ROI based on the standard magnetic resonance image sample; determine the classification result of each candidate classifier in at least one candidate classifier for classifying the image omics features; and determine the target classifier from at least one candidate classifier based on the classification result.

[0134] Step S303: Obtain magnetic resonance images of the glioma;

[0135] Step S304: Based on the pre-trained segmentation model, segment the region of interest corresponding to the glioma from the magnetic resonance image;

[0136] Step S305: Extract the image omics features of the region of interest;

[0137] Step S306: Input the radiomics features into a pre-determined target classifier to obtain the glioma classification results.

[0138] In practice, a segmentation model and a classifier can be pre-trained. During the training of the segmentation model, magnetic resonance image samples and their labels can be acquired. The magnetic resonance image samples are then synthesized according to their types to obtain multi-channel samples. These multi-channel samples are input into the segmentation model to be trained. Based on the difference between the obtained sample segmentation results and the sample labels, the segmentation model is trained to obtain the pre-trained segmentation model. For the classifier, magnetic resonance image samples can be input into the pre-trained segmentation model to obtain the region of interest (ROI) and reference region. The magnetic resonance image samples are then standardized based on the reference region. The process involves obtaining standard magnetic resonance imaging (MRI) samples, extracting the region of interest (ROI) features from these samples, and classifying the ROI features using at least one candidate classifier. A target classifier is then determined from these candidate classifiers. After training the segmentation model and classifier, MRI images of gliomas are acquired. These images are then input into the pre-trained segmentation model to segment the ROI and extract its ROI features. The extracted ROI features are then input into the target classifier to obtain the glioma classification.

[0139] The aforementioned medical image processing method can effectively segment the region of interest from the magnetic resonance image of glioma and classify the glioma according to the radiomics characteristics of the region of interest. Since molecular detection is not required, classification can be performed solely through medical images, thus improving the efficiency of glioma classification.

[0140] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0141] Based on the same inventive concept, this application also provides a medical image processing apparatus for implementing the medical image processing method described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, the specific limitations in one or more medical image processing apparatus embodiments provided below can be found in the limitations of the medical image processing method described above, and will not be repeated here.

[0142] In one exemplary embodiment, such as Figure 4 As shown, a medical image processing device is provided, including: an acquisition module 402, a segmentation module 404, an extraction module 406, and a classification module 408, wherein:

[0143] The acquisition module 402 is used to acquire magnetic resonance images of gliomas;

[0144] The segmentation module 404 is used to segment the region of interest corresponding to the glioma from the magnetic resonance image according to a pre-trained segmentation model;

[0145] Extraction module 406 is used to extract the image omics features of the region of interest;

[0146] The classification module 408 is used to input the radiomics features into a pre-determined target classifier to obtain the classification result of the glioma.

[0147] In an exemplary embodiment, the segmentation module 404 is further configured to register the magnetic resonance image according to a pre-set image template to obtain a registered magnetic resonance image; adjust the resolution and grayscale value of the registered magnetic resonance image to obtain a pre-processed magnetic resonance image; and input the pre-processed magnetic resonance image into the pre-trained segmentation model to segment the region of interest from the pre-processed magnetic resonance image.

[0148] In an exemplary embodiment, the medical image processing device further includes a segmentation model training module, used to acquire magnetic resonance image samples and sample labels corresponding to the magnetic resonance image samples; the magnetic resonance image samples correspond to at least one sample type; the sample labels are regions in the magnetic resonance image samples including tumors and edema; magnetic resonance image samples of different sample types are synthesized into multi-channel samples, and the multi-channel samples are input into the segmentation model to be trained to obtain the sample segmentation results of the multi-channel samples; the segmentation model to be trained is trained according to the difference between the sample segmentation results and the sample labels to obtain the pre-trained segmentation model.

[0149] In an exemplary embodiment, the extraction module 406 is further configured to: determine the region outside the region of interest as a reference region in the registered magnetic resonance image; determine the histogram mapping parameters between the reference region and the image template; perform mapping processing on the registered magnetic resonance image according to the histogram mapping parameters to obtain a standard magnetic resonance image; and extract the radiomics features of the region of interest according to the standard magnetic resonance image.

[0150] In an exemplary embodiment, the extraction module 406 is further configured to determine a standard region of interest (ROI) of the standard magnetic resonance image based on the ROI; extract at least one of the first-order statistical features, two-dimensional shape features, three-dimensional shape features, and grayscale matrix features of the standard ROI to obtain target extraction features of the standard ROI; and perform standardization processing on the target extraction features to obtain the radiomics features.

[0151] In an exemplary embodiment, the medical image processing apparatus further includes a classifier training module, configured to: acquire a region of interest (ROI) in the magnetic resonance imaging (MRI) sample; obtain a reference region based on MRI samples outside the ROI; standardize the MRI sample based on the reference region to obtain a standard MRI sample; extract radiomics features of the ROI based on the standard MRI sample; determine the classification result of each candidate classifier among at least one candidate classifier for classifying the radiomics features; and determine the target classifier from the at least one candidate classifier based on the classification result.

[0152] In an exemplary embodiment, the classifier training module is further configured to obtain the median corresponding to the sample feature if the sample feature is missing; and obtain the sample image omics feature based on the median.

[0153] Each module in the aforementioned medical image processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0154] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a medical image processing method. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0155] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0156] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0157] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0158] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0159] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0160] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0161] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0162] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A medical image processing method, characterized in that, The method includes: Obtain magnetic resonance images of gliomas; Based on a pre-trained segmentation model, the region of interest corresponding to the glioma is segmented from the magnetic resonance image; Extract the image omics features of the region of interest; The radiomics features are input into a pre-defined target classifier to obtain the glioma classification result.

2. The method according to claim 1, characterized in that, The step of segmenting the region of interest corresponding to the glioma from the magnetic resonance image based on the pre-trained segmentation model includes: The magnetic resonance images are registered according to a pre-set image template to obtain the registered magnetic resonance images. The resolution and grayscale value of the registered magnetic resonance image are adjusted to obtain the preprocessed magnetic resonance image. The preprocessed magnetic resonance image is input into the pre-trained segmentation model to segment the region of interest from the preprocessed magnetic resonance image.

3. The method according to claim 2, characterized in that, Before obtaining MRI images of the glioma, the following steps are also required: Obtain magnetic resonance image samples and corresponding sample labels for the magnetic resonance image samples; each magnetic resonance image sample corresponds to at least one sample type; the sample label is a region in the magnetic resonance image sample that includes tumors and edema. Magnetic resonance image samples of different sample types are combined into multi-channel samples, and the multi-channel samples are input into the segmentation model to be trained to obtain the sample segmentation results of the multi-channel samples; Based on the difference between the sample segmentation result and the sample label, the segmentation model to be trained is trained to obtain the pre-trained segmentation model.

4. The method according to claim 2, characterized in that, The extraction of image omics features from the region of interest includes: In the registered magnetic resonance image, the region outside the region of interest is defined as the reference region; Determine the histogram mapping parameters between the reference region and the image template; Based on the histogram mapping parameters, the registered magnetic resonance image is mapped to obtain a standard magnetic resonance image. Based on the standard magnetic resonance imaging, the radiomics features of the region of interest are extracted.

5. The method according to claim 4, characterized in that, The step of extracting the radiomics features of the region of interest based on the standard magnetic resonance image includes: Based on the region of interest, the standard region of interest of the standard magnetic resonance image is determined; Extract at least one of the first-order statistical features, two-dimensional shape features, three-dimensional shape features, and gray-scale matrix features from the standard region of interest to obtain the target extraction features of the standard region of interest. The extracted features of the target are standardized to obtain the image omics features.

6. The method according to claim 3, characterized in that, Before obtaining MRI images of the glioma, the following steps are also required: Obtain the region of interest (ROI) in the magnetic resonance image sample, and obtain the reference region based on the magnetic resonance image samples outside the ROI; Based on the sample reference region, the magnetic resonance image sample is standardized to obtain a standard magnetic resonance image sample; Based on the standard magnetic resonance image samples, extract the sample image omics features of the region of interest in the samples; Each candidate classifier in at least one candidate classifier is used to classify the sample image omics features; Based on the classification results, the target classifier is determined from the at least one candidate classifier.

7. The method according to claim 6, characterized in that, The sample image omics features correspond to at least one sample feature; the step of extracting the sample image omics features of the region of interest of the sample based on the standard magnetic resonance image sample includes: If the sample feature is missing, then obtain the median corresponding to the sample feature; The image omics features of the sample are obtained based on the median.

8. A medical image processing device, characterized in that, The device includes: The acquisition module is used to acquire magnetic resonance images of gliomas; The segmentation module is used to segment the region of interest corresponding to the glioma from the magnetic resonance image based on a pre-trained segmentation model. The extraction module is used to extract the radiomics features of the region of interest; The classification module is used to input the radiomics features into a pre-determined target classifier to obtain the classification result of the glioma.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images