Tumor heterogeneity analysis system and method based on magnetic resonance cell microstructure imaging

Abstract

This application discloses a tumor heterogeneity analysis system based on magnetic resonance cell microstructure imaging, including: a data acquisition module, a first image acquisition module, a second image acquisition module, a first prediction module, and a heterogeneity analysis module. Furthermore, this invention also provides a tumor heterogeneity analysis method based on magnetic resonance cell microstructure imaging. Based on a pre-trained prediction model, magnetic resonance cell microstructures are constructed using a magnetic resonance multimodal sequence obtained by combining magnetic resonance diffusion-weighted images and high-resolution T2-weighted images. Based on these magnetic resonance cell microstructures, the tumor heterogeneity of the patient to be predicted is predicted. This method not only quantitatively reflects the heterogeneity of tumor cell structure and function but also clearly reflects the heterogeneity of tumor morphology, enabling a comprehensive and accurate evaluation of the microscopic heterogeneity of rectal cancer and providing more accurate support for tumor treatment.

Description

Technical Field

[0001] This application relates to the field of bioinformatics, and in particular to a tumor heterogeneity analysis system and method based on magnetic resonance cell microstructure imaging. Background Technology

[0002] Rectal cancer is one of the most common malignant tumors and is highly heterogeneous. The degree of tumor differentiation, proliferative activity (Ki-67 expression level), nerve / microvascular invasion, and budding are important indicators of tumor heterogeneity, crucial for treatment and prognosis. However, currently, obtaining these important indicators before treatment for rectal cancer requires biopsy tissue obtained via colonoscopy. This is invasive and prone to complications; furthermore, biopsy tissue has limitations and is easily biased, failing to reflect the true overall condition of the tumor, especially regarding nerve and microvascular invasion and budding. This results in a time lag in the analysis of surgical specimens, hindering early treatment guidance, and further exacerbating the heterogeneous changes that occur in treated lesions.

[0003] In related technologies, magnetic resonance imaging (MRI) and computed tomography (CT) images can reflect the heterogeneity of tumor morphology. Radiomics can extract microscopic features from MRI and CT images to quantify the differences in tumor morphology. However, due to limitations in imaging resolution and feature extraction dimensions, the extracted microscopic features cannot reach the cellular level of microstructural and functional heterogeneity, making it difficult to accurately characterize the morphological heterogeneity of tumor margins. Summary of the Invention

[0004] In view of this, this application provides a tumor heterogeneity analysis system and method based on magnetic resonance cell microstructure imaging, which solves the problem that the existing process of tumor heterogeneity analysis based on magnetic resonance images cannot reach the microstructure and functional heterogeneity at the cellular level, and it is difficult to accurately characterize the morphological heterogeneity of tumor margins.

[0005] According to the first aspect of this application, a tumor heterogeneity analysis system based on magnetic resonance cell microstructure imaging is provided, comprising: a data acquisition module, a first image acquisition module, a second image acquisition module, a first prediction module, and a heterogeneity analysis module; The data acquisition module is used to acquire clinical data information of the patient to be predicted. The first image acquisition module is used to acquire magnetic resonance diffusion-weighted images with the tumor region of interest marked; The second image acquisition module is used to acquire high-resolution T2-weighted magnetic resonance images with the tumor region of interest marked. The first prediction module is used to input the clinical data information, the magnetic resonance diffusion-weighted image, and the magnetic resonance high-resolution T2-weighted image into a pre-trained prediction model, so that the prediction model can predict and analyze tumor heterogeneity based on the clinical data information, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to obtain tumor heterogeneity characterization information. The quantitative parameters of tumor cell structure and function are obtained by fitting the image features representing the region of interest of the tumor in the magnetic resonance diffusion-weighted image, and the tumor fractal dimension is obtained by measuring the box dimension method based on the magnetic resonance high-resolution T2-weighted image. The heterogeneity analysis module is used to perform quantitative analysis of cell structure and function based on the tumor heterogeneity characterization information predicted by the first prediction module.

[0006] Furthermore, the system also includes: a first image post-processing module; The first image post-processing module is used to analyze the tumor microscopic features in the magnetic resonance diffusion-weighted image based on the imaging model after the first image acquisition module acquires the magnetic resonance diffusion-weighted image, and to fit the tumor microscopic features to a preset number of least squares curves to obtain quantitative parameters of tumor cell structure and function. The quantitative parameters of tumor cell structure and function include the average cell diameter, intracellular volume fraction, extracellular diffusion rate, cell density, and water molecule transmembrane permeability in the tumor region of interest.

[0007] Furthermore, the system also includes: a second image post-processing module; The second image post-processing module is used to select multiple images from the high-resolution T2-weighted magnetic resonance images obtained by the second image acquisition module to extract the tumor region of interest and draw a binary map, and use the box dimension method to determine the fractal dimension of the binary map to obtain the tumor fractal dimension.

[0008] Furthermore, the second image post-processing module includes: an extraction unit and a measurement unit; The extraction unit is used to select three images from the highest-resolution T2-weighted magnetic resonance image: the largest tumor layer and two adjacent head and tail layers. The tumor region of interest is extracted from the three images, and the tumor region of interest is plotted as a binary image. The measurement unit is used to measure the fractal dimension of the binary image multiple times using the box dimension method, and select the mean of the fractal dimensions obtained from multiple measurements to obtain the tumor fractal dimension.

[0009] Furthermore, the system also includes: a comprehensive analysis module; The comprehensive analysis module is used to perform univariate regression analysis on the clinical data information of the patient to be predicted after the data acquisition module obtains the clinical data information to obtain the clinical factors that independently affect the characterization of tumor heterogeneity. The clinical factors, the quantitative parameters of tumor cell structure and function, and the fractal dimension of the tumor are subjected to multivariate logistic regression analysis to obtain the cell structure, functional parameters and clinical factors that independently affect the characterization of tumor heterogeneity.

[0010] Furthermore, the system also includes: a training sample acquisition module, a model training module, and a model building module; The training sample acquisition module is used to acquire training sample data of the machine learning model before inputting the clinical data information, the magnetic resonance diffusion-weighted image, and the magnetic resonance high-resolution T2-weighted image into the pre-trained prediction model. The training sample data includes sample input data and sample output data. The model training module is used to input the training sample data obtained by the training sample acquisition module into the machine learning model, so as to train the mapping relationship between the sample input data and the sample output data through the machine learning model; The model building module is used to build a prediction model based on the mapping relationship trained by the model training module when the training meets the iteration stopping condition.

[0011] Furthermore, the training sample acquisition module includes: an input data construction unit and an output data construction unit; The input data construction unit is used to construct sample input data based on clinical data of different patient samples, magnetic resonance diffusion-weighted images with tumor regions of interest marked, and magnetic resonance high-resolution T2-weighted images with tumor regions of interest marked. The output data construction unit is used to construct sample output data based on the tumor heterogeneity characterization information obtained from the analysis of patient samples through pathological specimens.

[0012] Furthermore, the machine learning model in the model training module includes, but is not limited to, any one of the following: decision tree method, random forest method, support vector machine method, and Naive Bayes method.

[0013] Furthermore, the verification modules include parameter acquisition, comparison, and evaluation. The verification parameter acquisition module is used to acquire tumor heterogeneity characterization information obtained from pathological specimen analysis of the patient to be verified after the prediction model is constructed. The comparison module is used to compare the tumor heterogeneity characterization information obtained by the verification parameter acquisition module through pathological specimen analysis of the patient to be verified with the tumor heterogeneity characterization information predicted by the prediction model, and obtain the receiver operating characteristic curve predicted by the model. The evaluation module is used to evaluate the prediction accuracy of the pre-trained prediction model based on the receiver operating characteristic curve predicted by the model.

[0014] According to a second aspect of this application, a method for tumor heterogeneity analysis based on magnetic resonance cell microstructure imaging is provided, comprising: Clinical data of the patient to be predicted, diffusion-weighted magnetic resonance images with regions of interest (ROIs) marked with tumors, and high-resolution T2-weighted magnetic resonance images with ROIs marked with tumors were obtained respectively. The clinical data, the diffusion-weighted magnetic resonance image, and the high-resolution T2-weighted magnetic resonance image are input into a pre-trained prediction model. The prediction model then uses the clinical data, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to predict and analyze tumor heterogeneity, thereby obtaining tumor heterogeneity characterization information. The quantitative parameters of tumor cell structure and function are obtained by fitting image features representing regions of interest in the tumor in the diffusion-weighted magnetic resonance image. The tumor fractal dimension is measured using the box-counting method based on the high-resolution T2-weighted magnetic resonance image. Quantitative analysis of cell structure and function was performed based on the predicted tumor heterogeneity characterization information. After acquiring a magnetic resonance diffusion-weighted image with the tumor region of interest labeled, the method further includes: Based on the imaging model, the tumor microscopic features in the magnetic resonance diffusion-weighted image are analyzed, and the tumor microscopic features are fitted with the least squares curve for a preset number of times to obtain quantitative parameters of tumor cell structure and function. The quantitative parameters of tumor cell structure and function include the average cell diameter, intracellular volume fraction, extracellular diffusion rate, cell density, and water molecule transmembrane permeability in the tumor region of interest. After acquiring high-resolution T2-weighted magnetic resonance images with tumor regions of interest labeled, the method further includes: Multiple images are selected from the high-resolution T2-weighted magnetic resonance images to extract the tumor region of interest and draw a binary map. The fractal dimension of the binary map is determined using the box-counting method to obtain the tumor fractal dimension. Specifically, this includes: selecting three images from the highest tumor level and its two adjacent head and tail levels in the high-resolution T2-weighted magnetic resonance images; extracting the tumor region of interest from the three images; drawing the tumor region of interest as a binary map; and repeatedly measuring the fractal dimension of the binary map using the box-counting method. The mean of the fractal dimensions obtained from the multiple measurements is then selected to obtain the tumor fractal dimension. After obtaining the clinical data of the patient to be predicted, the method further includes: Univariate regression analysis was performed on the clinical data to obtain the clinical factors that independently affect the characterization of tumor heterogeneity. Multivariate logistic regression analysis was then performed on the clinical factors, the quantitative parameters of tumor cell structure and function, and the fractal dimension of the tumor to obtain the cell structure, functional parameters, and clinical factors that independently affect the characterization of tumor heterogeneity. Before inputting the clinical data, the diffusion-weighted magnetic resonance imaging (MRI) image, and the high-resolution T2-weighted MRI image into the pre-trained prediction model, the method further includes: The training sample data for the machine learning model is obtained. The training sample data includes sample input data and sample output data. Specifically, it includes: constructing sample input data based on clinical data of different patient samples, magnetic resonance diffusion-weighted images with tumor regions of interest (MRI) labeled, and magnetic resonance high-resolution T2-weighted images with tumor MRI labeled; and constructing sample output data based on tumor heterogeneity characterization information obtained from pathological specimen analysis of patient samples. The training sample data obtained by the training sample acquisition module is input into the machine learning model to train the mapping relationship between the sample input data and the sample output data through the machine learning model. The machine learning model includes, but is not limited to, any one of the following: decision tree method, random forest method, support vector machine method, and Naive Bayes method. When the training meets the iteration stopping condition, a prediction model is constructed based on the mapping relationship obtained by the model training module. After constructing the prediction model, the method further includes: Obtain tumor heterogeneity characterization information obtained from pathological specimen analysis of patients to be validated; The tumor heterogeneity characterization information obtained from the pathological specimen analysis of the patient to be verified is used as the measured result and compared with the tumor heterogeneity characterization information predicted by the prediction model to obtain the receiver operating characteristic curve predicted by the model. The accuracy of the pre-trained prediction model is evaluated based on the receiver operating characteristic curve predicted by the model.

[0015] The beneficial effects of this invention are as follows: Compared with the existing tumor heterogeneity analysis process, this application uses magnetic resonance diffusion-weighted images to obtain structural information such as cell morphology, density, and arrangement in the tumor microenvironment, as well as functional information such as cell membrane water permeability, to reflect the microstructural and functional heterogeneity at the cellular level. Simultaneously, it combines high-resolution T2-weighted magnetic resonance images to obtain the tumor fractal dimension, reflecting the morphological heterogeneity of the tumor margin. Based on a pre-trained prediction model, a magnetic resonance multimodal sequence obtained by combining magnetic resonance diffusion-weighted images and high-resolution T2-weighted magnetic resonance images is used to construct magnetic resonance cell microstructures. Based on these magnetic resonance cell microstructures, the tumor heterogeneity of the patient to be predicted is predicted. This not only quantitatively reflects the structural and functional heterogeneity of tumor cells but also clearly reflects the morphological heterogeneity of the tumor, enabling a comprehensive and accurate evaluation of the microscopic heterogeneity of rectal cancer and providing more accurate assistance in tumor treatment.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a structural block diagram of tumor heterogeneity analysis based on magnetic resonance cell microstructure imaging provided in this application embodiment; Figure 2 This is a nodal plot of relevant parameters in the model prediction process provided in the embodiments of this application; Figure 3 This is a waterfall chart comparing the predicted results and measured results provided in the embodiments of this application; Figure 4 This is the receiver operating characteristic curve of the prediction model provided in the embodiments of this application; Figure 5 This is a structural block diagram of another tumor heterogeneity analysis system based on magnetic resonance cell microstructure imaging provided in this application embodiment. Detailed Implementation

[0018] The invention will now be discussed with reference to several exemplary embodiments. It should be understood that these embodiments are described merely to enable those skilled in the art to better understand and thus implement the invention, and are not intended to imply any limitation on the scope of the invention.

[0019] As used herein, the term "comprising" and its variations are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "at least partially based on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment". The term "another embodiment" is to be interpreted as "at least one other embodiment".

[0020] In related technologies, magnetic resonance imaging (MRI) and computed tomography (CT) images can reflect the heterogeneity of tumor morphology. Radiomics can extract microscopic features from MRI and CT images to quantify the differences in tumor morphology. However, due to limitations in imaging resolution and feature extraction dimensions, the extracted microscopic features cannot reach the cellular level of microstructural and functional heterogeneity, making it difficult to accurately characterize the morphological heterogeneity of tumor margins.

[0021] To address this issue, this embodiment provides a tumor heterogeneity analysis system based on magnetic resonance cellular microstructure imaging. Magnetic resonance cellular microstructure imaging is a non-invasive quantitative analysis method that utilizes magnetic resonance technology to analyze cellular microstructures, primarily based on differences in water molecule diffusion characteristics to reflect cell morphology, density, and other features. Its core principle is that magnetic resonance imaging uses a magnetic field to cause hydrogen nuclei within the tissue to resonate and records their energy release trajectories to reconstruct images, forming a cellular microstructure image reflecting the macroscopic structure of the tissue. Building upon this microstructure imaging, techniques such as oscillating gradient spin echo are further utilized to quantitatively analyze parameters such as cell diameter and density by measuring signal changes in water molecules at different diffusion times.

[0022] The specific system structure is as follows: Figure 1 As shown, it includes: a data acquisition module 11, a first image acquisition module 12, a second image acquisition module 14, a first prediction module 14, and a heterogeneity analysis module 15; The data acquisition module 11 is used to acquire clinical data information of the patient to be predicted; The first image acquisition module 12 is used to acquire magnetic resonance diffusion-weighted images with the tumor region of interest marked; The second image acquisition module 13 is used to acquire high-resolution T2-weighted magnetic resonance images with the tumor region of interest marked. The first prediction module 14 is used to input the clinical data information, the magnetic resonance diffusion-weighted image, and the magnetic resonance high-resolution T2-weighted image into a pre-trained prediction model, so that the prediction model can predict and analyze tumor heterogeneity based on the clinical data information, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to obtain tumor heterogeneity characterization information. The quantitative parameters of tumor cell structure and function are obtained by fitting the image features representing the region of interest of the tumor in the magnetic resonance diffusion-weighted image, and the tumor fractal dimension is obtained by measuring the box dimension method based on the magnetic resonance high-resolution T2-weighted image. The heterogeneity analysis module 15 is used to perform quantitative analysis of cell structure and function based on the tumor heterogeneity characterization information predicted by the first prediction module.

[0023] In the above system, the output terminals of the first image acquisition module, the second image acquisition module, and the data acquisition module are all connected to the input terminal of the first prediction module, and the output terminal of the first prediction module is connected to the input terminal of the heterogeneity analysis module.

[0024] In the aforementioned data acquisition module, the clinical data information refers to the patient's clinical data input by the operator, including but not limited to the patient's gender, age, height, weight, blood routine and serum tumor marker results, preoperative clinical stage, etc. The preoperative clinical stage here can be the preoperative TNM clinical stage of rectal cancer.

[0025] In the aforementioned first image acquisition module, magnetic resonance diffusion-weighted images can be retrieved via a path provided by the operator, pre-stored within the computer system space as images of all cytological imaging sequences from the patient's rectal magnetic resonance imaging scan. These images are stored in NII format and can be obtained based on the time-diffusion spectroscopy imaging function of the diffusion-weighted sequences, including two types of diffusion-weighted scan sequences with different diffusion times: pulsed gradient spin echo (PGSE) and oscillating gradient spin echo (OGSE). The tumor region of interest (ROI) marked here can be delineated by professional medical personnel in different magnetic resonance diffusion-weighted images, and the ROI in different MRI diffusion-weighted images is consistent. Further explanation is that the ROI is a part of the image, generated by selecting on the image or using methods such as setting a threshold or converting from other files (e.g., vectors). The ROI can be an irregular shape such as a point, line, or surface, and is typically used as a sample for image classification, a mask, a cropping region, and for other operations.

[0026] It should be noted that, considering the different technical characteristics and clinical goals of various diffusion-weighted sequences, in the parameter design process of magnetic resonance diffusion-weighted images, pulsed gradient spin echo (PGSE) focuses on the universality of routine diffusion assessment and usually uses a single sequence parameter combination, while oscillating gradient spin echo (OGSE) needs to cover different microscale diffusion using multiple sets of parameters, and usually requires at least two sequence parameter combinations. Accordingly, magnetic resonance diffusion-weighted images can be acquired using the acquisition parameters shown in Table 1 to obtain diffusion-weighted scanning sequences with different diffusion times.

[0027] Table 1. Acquisition parameters of diffusion-weighted magnetic resonance images

[0028] Where δ is the duration of each diffusion gradient, and Δ is the time interval between two diffusion gradients. The oscillation frequency of the diffuse gradient, This refers to the diffusion time. For PGSE sequences, =Δ-δ / 3. For OGSE sequences, =1 / (4 The values ​​are parameters used to quantify the sensitivity of the diffusion-sensitive gradient field to the diffusion motion of water molecules in tissues. Higher values ​​indicate greater sensitivity of the gradient field to differences in water molecule diffusion. Generally, the values ​​in Table 1 are not arbitrarily set, but rather correlated with the core parameters of the diffusion gradient (δ, Δ, ...). (Directly related)

[0029] In Table 1 above, the period number and sum are specific parameters describing the number of repetitions of the oscillating gradient, and they only exist in sequences where the gradient changes continuously and periodically. Considering that different sequences apply gradients in different ways, the gradient of the OGSE sequence is continuous and periodically repetitive, similar to a sine wave oscillation, requiring the period number to represent the number of gradient repetitions. Different period numbers can control the number of times the gradient samples the microscopic diffusion of water molecules. The diffuse gradient of the PGSE sequence is applied in the completely opposite way to the OGSE sequence; it is not a continuous oscillation. Therefore, the PGSE sequence does not require the period number and sum parameters.

[0030] In the second image acquisition module described above, high-resolution T2-weighted magnetic resonance images can be retrieved from the patient's rectal magnetic resonance imaging scan pre-stored in the computer system space via a path provided by the operator, and stored as .nii format images. Similarly, the marked tumor region of interest is delineated by professional medical personnel in the high-resolution T2-weighted magnetic resonance images.

[0031] In this embodiment, high-resolution T2-weighted magnetic resonance imaging (MRI) images are a type of MRI image capable of clearly displaying the fine structure of tissues and are generated based on T2-weighted imaging sequences. Due to their high spatial resolution and high sensitivity to changes in tissue water content, high-resolution T2-weighted MRI images play an irreplaceable role in the treatment of rectal cancer. In practical applications, specific oblique axial scanning orientations can be used, and correspondingly, high-resolution T2-weighted MRI images can be acquired using the scanning parameters shown in Table 2.

[0032] Table 2. Scanning parameters of high-resolution T2-weighted magnetic resonance images

[0033] Where TR is the repetition time, which is the time interval between two consecutive radio frequency pulse excitations in magnetic resonance imaging; TE is the echo time, which is the time interval between the generation of an echo signal after radio frequency pulse excitation and the acquisition of that signal; Matrix is ​​the matrix, the size of which directly determines the spatial resolution of the image. The more pixels there are, the larger the matrix, and the clearer the image details, but the scanning time will be correspondingly longer; FOV is the field of view, which refers to the extent of the imaging area and determines the spatial coverage of the image; ETL is the echo train length, which refers to the number of echo signals acquired consecutively after one radio frequency pulse excitation; NEX is the number of excitations, which refers to the number of times the same imaging layer is repeatedly excited, affecting the image signal-to-noise ratio; Thickness is the thickness of each imaging layer during magnetic resonance imaging scanning; and Gap is the interslice spacing, which refers to the distance between two adjacent imaging layers.

[0034] In the first prediction module mentioned above, the quantitative parameters of tumor cell structure and function are calculated based on an imaging model. This imaging model is established based on the physical properties of image signals. Specifically, a magnetic resonance diffusion-weighted image can be input into the imaging model. The model analyzes the correlation between the image signal and the tumor microscopic features, converting abstract image grayscale values ​​into quantitative indicators with clear biological significance. Based on the tumor microscopic features output by the model, a least-squares curve is used to fit the tumor microscopic features, and this is randomly initialized and repeated 100 times to avoid local minima. This yields the quantitative parameters of tumor cell structure and function, which include at least the average cell diameter (d), intracellular volume fraction (Vin), extracellular diffusion rate (Dex), cell density (Cellularity), and water molecule transmembrane permeability within the tumor region of interest. To facilitate observation, the quantitative parameters of tumor cell structure and function (d, Vin, Dex, Cellularity, ...) can be displayed using a habitat map. The aforementioned feature fitting, parameter calculation, and habitat map generation can all be performed using data analysis software.

[0035] Accordingly, the system further includes: a first image post-processing module; the first image post-processing module is used to analyze the tumor microscopic features in the magnetic resonance diffusion-weighted image based on the imaging model after the first image acquisition module acquires the magnetic resonance diffusion-weighted image, and to fit the tumor microscopic features to a preset number of least squares curves to obtain quantitative parameters of tumor cell structure and function, wherein the quantitative parameters of tumor cell structure and function include the average cell diameter, intracellular volume fraction, extracellular diffusion rate, cell density, and water molecule transmembrane permeability within the tumor region of interest.

[0036] It is understandable that quantitative parameters of tumor cell structure and function directly reflect the intrinsic characteristics of tumor cells. The essence of tumor heterogeneity is that tumor cells differ in morphology, function, and other aspects. Therefore, quantitative parameters of tumor cell structure and function can be used to describe the heterogeneity of tumor cells.

[0037] In the first prediction module mentioned above, fractal dimension (FD) is a non-integer dimensionality metric used to quantify the geometric complexity and spatial filling characteristics of an image or object. It transcends the traditional Euclidean geometric concept of dimension, reflecting structural irregularities and texture roughness by calculating the degree of self-similarity of the image's grayscale surface or boundary contours. In imaging classification, fractal dimension provides an objective and quantitative means to describe the morphological heterogeneity of lesion regions or tissue microstructures. To quantitatively assess the structural complexity and spatial irregularity of tumors in images, the specific fractal dimension can be calculated using box counting to determine the roughness of the image's grayscale or binary mask; a higher value indicates a more complex structure and more irregular spatial occupancy. In other words, tumor fractal dimension is a key indicator used to quantify the degree of irregularity and complexity of tumor morphology, structure, or generation patterns, enabling more accurate capture of the complexity of tumor spatial distribution, boundary morphology, and internal structure.

[0038] Accordingly, the system further includes: a second image post-processing module; the second image post-processing module is used to select multiple images from the high-resolution T2-weighted magnetic resonance images after the second image acquisition module has acquired them, extract the tumor region of interest from the high-resolution T2-weighted magnetic resonance images to draw a binary map, and use the box dimension method to determine the fractal dimension of the binary map to obtain the tumor fractal dimension.

[0039] Specifically, three images from three layers (the largest tumor layer and its two adjacent head and tail layers) can be selected from high-resolution T2-weighted magnetic resonance imaging. The region of interest (ROI) of the tumor in each image is extracted to create a binary map. The fractal dimension of the binary map is determined using a box-method software tool. Each image is measured three times, and the mean value is selected as the final tumor fractal dimension. The calculation process for the corresponding tumor fractal dimension can be expressed by the following formula:

[0040] in, This represents the number of boxes with side length 1 required to cover the entire binary graph.

[0041] Accordingly, the second image post-processing module includes an extraction unit and a measurement unit; the extraction unit is used to select three images from the highest-resolution T2-weighted magnetic resonance image, including the largest tumor layer and two adjacent head and tail layers, extract the region of interest of the tumor from the three images, and plot the region of interest of the tumor as a binary image; the measurement unit is used to measure the fractal dimension of the binary image multiple times using the box-counting method, and select the mean of the fractal dimensions obtained from multiple measurements to obtain the tumor fractal dimension.

[0042] Understandably, the morphological and structural complexity of tumors is an important manifestation of their heterogeneity; a higher fractal dimension indicates a more complex tumor morphology and a greater degree of heterogeneity. Therefore, fractal dimension can quantitatively describe the morphological complexity, internal texture, or vascular network characteristics of tumors, providing an objective and quantitative method for assessing tumor heterogeneity.

[0043] In the first prediction module mentioned above, although clinical data cannot directly reflect tumor heterogeneity, it can provide important macro-level background information. For example, different tumor stages may exhibit varying degrees of heterogeneity, with advanced tumors typically exhibiting higher heterogeneity. This clinical data can help the prediction model better understand the overall situation of tumor heterogeneity. Correspondingly, the prediction model can comprehensively analyze multi-dimensional information such as clinical data, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to uncover potential relationships and patterns between data, accurately predicting tumor heterogeneity characteristics. In other words, in the process of analyzing and predicting tumor heterogeneity, on the one hand, by using quantitative parameters of tumor cell structure and function, water molecule permeability can be incorporated into the analysis of tumor microstructure, reflecting the heterogeneity of tumor cell arrangement, size, density, and function; on the other hand, by introducing the fractal dimension of tumor into the fractal analysis of habitat images, the complexity of tumor morphology can be quantified. Combining quantitative parameters of tumor cell structure and function with the fractal dimension of tumor can more accurately assess tumor cell heterogeneity, enabling quantitative analysis of key factors such as tumor differentiation and Ki-67 expression, and providing a more reliable reference for individual treatment.

[0044] In this embodiment, tumor heterogeneity characterization information mainly refers to the differences in structure, function, and biological behavior of tumor cells, directly reflecting the heterogeneous characteristics of rectal cancer cells and serving as a key basis for clinically assessing tumor malignancy and invasion risk. Specific tumor heterogeneity characterization information may include, but is not limited to, tumor pathological differentiation grade, Ki-67 expression level, intratumoral neural and vascular invasion, and budding, among other related indicators. In other words, the essence of tumor heterogeneity is the multi-scale differences in tumor cells across genes, phenotypes, and clinical manifestations. Directly inputting all features into the prediction model may lead to overfitting and fail to clearly define the core characteristics of heterogeneity. To first identify the core clinical variables directly related to heterogeneity and exclude irrelevant clinical information, it is necessary to first screen for single clinical factors. Then, based on the core clinical variables, further screening is performed to identify cell structure / morphological parameters independently related to heterogeneity. Specifically, based on clinical data, single-factor regression analysis in the analysis software is used to screen for clinical factors independently related to tumor heterogeneity. Then, multivariate logistic regression analysis is performed, combining quantitative parameters of tumor cell structure and function and tumor fractal dimension, to obtain cell structure, functional parameters, and clinical factors independently related to tumor heterogeneity. Tumor heterogeneity can be characterized by indicators such as tumor pathological differentiation grade, Ki-67 expression level, intratumoral neural and vascular invasion, and budding. Furthermore, cellular structural and functional parameters and clinical factors independently related to tumor heterogeneity are input into the prediction model. The model then predicts tumor heterogeneity, yielding tumor heterogeneity characterization information with high accuracy and stability.

[0045] Accordingly, the system also includes a comprehensive analysis module; the comprehensive analysis module is used to perform univariate regression analysis on the clinical data information of the patient to be predicted after the data acquisition module obtains the clinical data information to obtain the clinical factors that independently affect the characterization of tumor heterogeneity, and to perform multivariate logistic regression analysis on the clinical factors, the quantitative parameters of tumor cell structure and function, and the fractal dimension of the tumor to obtain the cell structure, functional parameters and clinical factors that independently affect the characterization of tumor heterogeneity.

[0046] In practical applications, tumor heterogeneity analysis can be used for initial treatment selection, predicting prognosis and recurrence risk, and screening patients for specific treatments. It primarily focuses on the internal differences within tumor cells, revealing the heterogeneity within tumor cell populations—that is, the differences in genes, phenotypes, and functions among different cell subpopulations within the same tumor. For example, some cells exhibit more active growth (higher Ki-67 expression levels). Correspondingly, tumor heterogeneity analysis uses high-resolution techniques to capture the differences in individual tumor cells or small subpopulations, focusing on analyzing the microscopic characteristics of tumor cells. Ultimately, the results obtained from tumor heterogeneity analysis can directly target differences in tumor cells.

[0047] Correspondingly, in the process of quantitatively analyzing cell structure and function based on tumor heterogeneity characterization information, the main focus is on establishing a quantitative relationship between the heterogeneity characteristics of tumor cells and cell structure / function indicators. Specifically, this process can pre-construct the correlation targets between cell structure / function indicators and heterogeneity characterization, i.e., the heterogeneity characteristics corresponding to different cell structure / function indicators. For example, the heterogeneity dimension corresponding to tumor pathological differentiation grade is differentiation heterogeneity, and the corresponding matching cell structure indicators include cell morphology regularity and cell integrity, while the corresponding matching cell function indicators include cell metabolic activity and apoptosis rate. Then, based on statistical methods or models, correlation analysis is performed between the quantitative values ​​of heterogeneity characterization information and cell structure / function indicators. This correlation analysis includes whether there is a correlation, the strength of the correlation, and the direction of the correlation. Based on the correlation analysis results, statistical tests are performed on the structure / function indicators of different heterogeneous subgroups to clarify the significance of differences. In other words, by binding heterogeneity with pathological indicators familiar to clinicians, the above process gives cell structure and function indicators clear clinical orientation, providing auxiliary basis for clinical decision-making such as pathology reports, cell indicator analysis, and treatment plan formulation, thus realizing the clinical value of heterogeneity analysis.

[0048] Furthermore, the system also includes: a training sample acquisition module, a model training module, and a model building module; The training sample acquisition module is used to acquire training sample data of the machine learning model before inputting the clinical data information, the magnetic resonance diffusion-weighted image, and the magnetic resonance high-resolution T2-weighted image into the pre-trained prediction model. The training sample data includes sample input data and sample output data. The model training module is used to input the training sample data obtained by the training sample acquisition module into the machine learning model, so as to train the mapping relationship between the sample input data and the sample output data through the machine learning model; The model building module is used to build a prediction model based on the mapping relationship trained by the model training module when the training meets the iteration stopping condition.

[0049] Furthermore, the training sample acquisition module includes: an input data construction unit and an output data construction unit; The input data construction unit is used to construct sample input data based on clinical data of different patient samples, magnetic resonance diffusion-weighted images with tumor regions of interest marked, and magnetic resonance high-resolution T2-weighted images with tumor regions of interest marked. The output data construction unit is used to construct sample output data based on the tumor heterogeneity characterization information obtained from the analysis of patient samples through pathological specimens.

[0050] It should be noted that during the model training phase, the input data for the machine learning model includes not only clinical data, diffusion-weighted magnetic resonance imaging (MRI) images, and high-resolution T2-weighted MRI images from different patient samples, but also tumor heterogeneity characterization information obtained from pathological specimen analysis of patient samples. This allows the machine learning model to train the mapping relationship between the training input data and the training output data, thus constructing a predictive model. However, during the model prediction phase, the actual input data does not include tumor heterogeneity characterization information obtained from pathological specimen analysis of patient samples. This allows the predictive model to predict tumor heterogeneity characterization information based on the clinical data, MRI diffusion-weighted images, and high-resolution T2-weighted MRI images of the patient to be predicted.

[0051] Furthermore, the machine learning model in the model training module includes, but is not limited to, any one of the following: decision tree method, random forest method, support vector machine method, and Naive Bayes method.

[0052] To evaluate the prediction performance of the trained prediction model, the accuracy and stability of the trained prediction model can be verified using the receiver operating characteristic curve predicted by the model. Furthermore, the system also includes: a verification parameter acquisition module, a comparison module, and an evaluation module. The verification parameter acquisition module is used to acquire tumor heterogeneity characterization information obtained from pathological specimen analysis of the patient to be verified after the prediction model is constructed. The comparison module is used to compare the tumor heterogeneity characterization information obtained by the verification parameter acquisition module through pathological specimen analysis of the patient to be verified with the tumor heterogeneity characterization information predicted by the prediction model, and obtain the receiver operating characteristic curve predicted by the model. The evaluation module is used to evaluate the prediction accuracy of the pre-trained prediction model based on the receiver operating characteristic curve predicted by the model.

[0053] In the above comparison module, the measured results are the tumor heterogeneity characterization information obtained by analyzing the pathological specimens of the patient to be verified. Correspondingly, the clinical data information of the patient to be verified, the magnetic resonance diffusion-weighted image with the region of interest of the tumor, and the magnetic resonance high-resolution T2-weighted image with the region of interest of the tumor are input into the trained prediction model, and the tumor heterogeneity characterization information can be predicted. This section uses nomogram calculations to display relevant parameters during the model prediction process. For example, based on the clinical, imaging, and pathological data of 160 patients with locally advanced disease, a quantitative analysis of the structure and function of rectal cancer cells was constructed and externally validated. Patients were ranked according to treatment time and assigned to a training group (112 cases) and an external validation group (48 cases) in a 7:3 ratio. During training, the diffusion-weighted magnetic resonance images and high-resolution T2-weighted magnetic resonance images of the patients in the training group were stored in designated folders. Complete information was entered into the clinical data acquisition module. After the prediction model training was completed, the diffusion-weighted magnetic resonance images, high-resolution T2-weighted magnetic resonance images, and clinical data of the patients to be predicted were sequentially input into the prediction model. The prediction model then performed a prediction to obtain the prediction results. See the example below for details. Figure 2 The nomogram shown Figure 2 The nomograms illustrate specific patient characteristics, including poorly differentiated tumor pathology, Ki-67 expression >30%, intratumoral vascular / nerve invasion, and sprouting risk. During validation, measured results were obtained for each patient in the external validation group and correlated with the indicators in the nomograms. These measured results included poorly differentiated tumor pathology, Ki-67 expression >30%, intratumoral vascular / nerve invasion, and sprouting risk. Further comparison of the predicted results with the measured results showed that the accuracy of the predictive model was consistently greater than 80%. (See attached reference for details.) Figure 3 The waterfall picture, in Figure 3 In the training set, the predicted tumors were poorly differentiated, and the predicted Ki-67 expression rate was >30% in the validation set. The predicted values ​​of vascular / nerve invasion were 83.8%, 81.3%, and 82.0% respectively.

[0054] Since the training sample data for the validation group is obtained by inputting the sample input data into the prediction model, the sample output data can be directly used as the actual result to compare with the prediction result data for easier comparison. This yields the receiver operating characteristic (ROC) curve of the model's prediction, which is then used to verify the accuracy and stability of the prediction model. The specific ROC curve of the prediction model is shown below. Figure 4 As shown, the bias between the predicted results of the training set cases and the measured results of the validation set cases is compared using the receiver operating characteristic curve. Figure 4The results showed that the predictive power of the receiver operating characteristic (ROC) curves for the validation set cases reached 0.85-0.91. Specifically, the predictive power for the ROC curves for predicting Ki-67 expression ratio >30% was 0.858, for predicting vascular / nerve invasion it was 0.886, for predicting tumor budding it was 0.889, and for predicting poor tumor differentiation it was 0.912.

[0055] Furthermore, as Figure 1 In terms of specific system implementation, this application provides another tumor heterogeneity analysis system based on magnetic resonance cell microstructure imaging, such as... Figure 5 As shown, the system includes: a diffusion-weighted image acquisition module 21, a diffusion-weighted image processing module 22, a high-resolution image acquisition module 23, a high-resolution image processing module 24, a clinical data acquisition module 25, a comprehensive analysis module 26, a second prediction module 27, and a result display module 28. The output of the diffusion-weighted image acquisition module 21 is connected to the input of the diffusion-weighted image processing module 22; the output of the high-resolution image acquisition module 23 is connected to the input of the high-resolution image processing module 24; the outputs of the diffusion-weighted image processing module 22, the high-resolution image processing module 24, and the clinical data acquisition module 25 are all connected to the input of the comprehensive analysis module 26; the output of the comprehensive analysis module 26 is connected to the input of the second prediction module 27; and the output of the second prediction module 27 is connected to the input of the result display module 28.

[0056] The diffusion-weighted image acquisition module 21 is used to acquire magnetic resonance diffusion-weighted images with tumor regions of interest marked. The diffusion-weighted image processing module 22 is used to analyze the tumor micro-features in the magnetic resonance diffusion-weighted image based on the imaging model. It uses the least squares curve to fit the tumor micro-features a preset number of times to obtain the quantitative parameters of tumor cell structure and function, and sends the quantitative parameters of tumor cell structure and function to the comprehensive analysis module 26. The high-resolution image acquisition module 23 is used to acquire high-resolution T2-weighted magnetic resonance images with tumor regions of interest marked. The high-resolution image processing module 24 is used to select multiple images from the high-resolution T2-weighted magnetic resonance images to extract the tumor region of interest to draw a binary map, use the box dimension method to determine the fractal dimension of the binary map, obtain the tumor fractal dimension, and send the tumor fractal dimension to the comprehensive analysis module 26. The clinical data acquisition module 25 is used to acquire clinical data information of the patient to be predicted and send the clinical data information to the comprehensive analysis module 26; The comprehensive analysis module 26 is used to perform univariate regression analysis on clinical data to obtain the clinical factors that independently affect the characterization of tumor heterogeneity. It performs multivariate logistic regression analysis on clinical factors, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to obtain the cell structure, functional parameters and clinical factors that independently affect the characterization of tumor heterogeneity. The second prediction module 27 is used to predict and analyze tumor heterogeneity based on clinical data, quantitative parameters of tumor cell structure and function, and tumor fractal dimension through a prediction model, and obtain tumor heterogeneity characterization information. Display module 28 is used to display the tumor heterogeneity characterization information of the predicted patient in the form of a nomogram, and to perform quantitative analysis of cell structure and function based on the tumor heterogeneity characterization information.

[0057] It is understandable that the aforementioned diffusion-weighted image acquisition module 21 and diffusion-weighted image processing module 22 are combined to obtain quantitative parameters of tumor cell structure and function, and the aforementioned high-resolution image acquisition module 23 and high-resolution image processing module 24 are combined to obtain the tumor fractal dimension. The quantitative parameters of tumor cell structure and function, the tumor fractal dimension, and the clinical data information obtained by the data acquisition module are formed into three data branches. These three data branches are executed in parallel in actual application, without any order of data acquisition. Then, the data obtained from the three data branches are transmitted to the comprehensive analysis module 26, which merges and analyzes the acquired data to obtain the cell structure, functional parameters, and clinical factors that independently affect the characterization of tumor heterogeneity. Then, the second prediction module 27 predicts and analyzes the tumor heterogeneity of the patient to be predicted based on the data obtained from the comprehensive analysis module 26, and outputs the predicted data to the display module 28. The predicted tumor heterogeneity characterization information is displayed in a nomogram, and the cell structure and function are quantitatively analyzed based on the displayed tumor heterogeneity characterization information.

[0058] In addition, the present invention also provides a method for tumor heterogeneity analysis based on magnetic resonance cell microstructure imaging, comprising the following steps: S1. Acquire clinical data of the patient to be predicted, diffusion-weighted magnetic resonance images with tumor regions of interest marked, and high-resolution T2-weighted magnetic resonance images with tumor regions of interest marked. S2. Input the clinical data, the magnetic resonance diffusion-weighted image, and the magnetic resonance high-resolution T2-weighted image into a pre-trained prediction model. The prediction model then uses the clinical data, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to predict and analyze tumor heterogeneity, thereby obtaining tumor heterogeneity characterization information. The quantitative parameters of tumor cell structure and function are obtained by fitting the image features representing the region of interest of the tumor in the magnetic resonance diffusion-weighted image. The tumor fractal dimension is obtained by measuring the box dimension method based on the high-resolution T2-weighted image. S3. Quantitative analysis of cell structure and function based on the predicted tumor heterogeneity characterization information; After step S1, the method further includes: S4. Based on the imaging model, analyze the tumor micro-features in the magnetic resonance diffusion-weighted image, and use the least squares curve to fit the tumor micro-features a preset number of times to obtain quantitative parameters of tumor cell structure and function. The quantitative parameters of tumor cell structure and function include the average cell diameter, intracellular volume fraction, extracellular diffusion rate, cell density, and water molecule transmembrane permeability in the tumor region of interest. After step S1, the method further includes: S5. Select multiple images from the high-resolution T2-weighted magnetic resonance images to extract the region of interest of the tumor and draw a binary map. Use the box-counting method to determine the fractal dimension of the binary map to obtain the fractal dimension of the tumor. Specifically, this includes: S5.1 Select three images from the highest-resolution T2-weighted magnetic resonance image: the largest tumor layer and its two adjacent head and tail layers. Extract the region of interest (ROI) from the three images and plot the ROI as a binary image. S5.2. The fractal dimension of the binary image is measured multiple times using the box dimension method. The mean of the fractal dimensions obtained from multiple measurements is selected to obtain the fractal dimension of the tumor. After step S1, the method further includes: S6. Perform univariate regression analysis on the clinical data to obtain the clinical factors that independently affect the characterization of tumor heterogeneity. Perform multivariate logistic regression analysis on the clinical factors, the quantitative parameters of tumor cell structure and function, and the fractal dimension of the tumor to obtain the cell structure, functional parameters, and clinical factors that independently affect the characterization of tumor heterogeneity. Prior to step S2, the method further includes: S7. Obtain training sample data for the machine learning model, wherein the training sample data includes sample input data and sample output data, specifically including: S7.1 Construct sample input data based on clinical data of different patient samples, diffusion-weighted magnetic resonance images with tumor regions of interest (MRI) and high-resolution T2-weighted magnetic resonance images with tumor regions of interest (MRI). S7.2 Construct sample output data based on the tumor heterogeneity characterization information obtained from the analysis of pathological specimens of patient samples; S8. Input the training sample data obtained by the training sample acquisition module into the machine learning model, so as to train the mapping relationship between the sample input data and the sample output data through the machine learning model. The machine learning model includes, but is not limited to, any one of the following: decision tree method, random forest method, support vector machine method, and Naive Bayes method. S9. When the training meets the iteration stopping condition, a prediction model is constructed based on the mapping relationship obtained by the model training module. After step S9, the method further includes: S10. Obtain tumor heterogeneity characterization information obtained from pathological specimen analysis of the patient to be validated; S11. The tumor heterogeneity characterization information obtained by the pathological specimen analysis of the patient to be verified is used as the measured result and compared with the tumor heterogeneity characterization information predicted by the prediction model to obtain the receiver operating characteristic curve predicted by the model. S12. Evaluate the prediction accuracy of the pre-trained prediction model based on the receiver operating characteristic curve predicted by the model.

[0059] The tumor heterogeneity analysis method based on magnetic resonance cell microstructure imaging provided in this application, compared with the existing tumor heterogeneity analysis process, uses magnetic resonance diffusion-weighted images to obtain structural information such as cell morphology, density, and arrangement in the tumor microenvironment, as well as functional information such as cell membrane water permeability, to reflect the microstructural and functional heterogeneity at the cellular level. Simultaneously, it combines high-resolution T2-weighted magnetic resonance images to obtain the fractal dimension of the tumor, reflecting the morphological heterogeneity of the tumor margin. Based on a pre-trained prediction model, it uses magnetic resonance multimodal sequences obtained by combining magnetic resonance diffusion-weighted images and high-resolution T2-weighted magnetic resonance images to construct magnetic resonance cell microstructures. Based on these magnetic resonance cell microstructures, it predicts the tumor heterogeneity of the patient being tested. This not only quantitatively reflects the structural and functional heterogeneity of tumor cells but also clearly reflects the morphological heterogeneity of the tumor, enabling a comprehensive and accurate evaluation of the microscopic heterogeneity of rectal cancer and providing more accurate support for tumor treatment.

[0060] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), and includes several instructions to enable a computer device (such as a personal computer, server, or network device) to execute the tumor heterogeneity analysis method based on magnetic resonance cell microstructure imaging described in various implementation scenarios of this application.

[0061] Based on the above methods, and Figure 1 and Figure 5 To achieve the above objectives, the system embodiment shown in this application also provides a physical device for tumor heterogeneity analysis based on magnetic resonance cell microstructure imaging. Specifically, it can be a computer, smartphone, tablet computer, smartwatch, server, or network device, etc. The physical device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the above-mentioned tumor heterogeneity analysis method based on magnetic resonance cell microstructure imaging.

[0062] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0063] Those skilled in the art will understand that the physical device structure for tumor heterogeneity analysis based on magnetic resonance cell microstructure imaging provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0064] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device for tumor heterogeneity analysis based on magnetic resonance cell microstructure imaging, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0065] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented through hardware. By applying the technical solution of this application, compared with the existing methods, this application uses magnetic resonance diffusion-weighted images to obtain structural information such as cell morphology, density, and arrangement in the tumor microenvironment, as well as functional information such as cell membrane water permeability, to reflect the microstructural and functional heterogeneity at the cellular level. At the same time, it combines high-resolution T2-weighted magnetic resonance images to obtain the tumor fractal dimension to reflect the morphological heterogeneity of the tumor edge. Based on a pre-trained prediction model, it uses clinical data, quantitative parameters of tumor cell structure and function, and the tumor fractal dimension to predict the tumor heterogeneity of the patient to be predicted, so as to achieve a comprehensive and accurate evaluation of the microscopic heterogeneity of rectal cancer, which can more accurately assist in tumor treatment.

[0066] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0067] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A tumor heterogeneity analysis system based on magnetic resonance cellular microstructure imaging, characterized in that, include: The system includes a data acquisition module, a first image acquisition module, a second image acquisition module, a first prediction module, and a heterogeneity analysis module. The data acquisition module is used to acquire clinical data information of the patient to be predicted. The first image acquisition module is used to acquire magnetic resonance diffusion-weighted images with the tumor region of interest marked; The second image acquisition module is used to acquire high-resolution T2-weighted magnetic resonance images with the tumor region of interest marked. The first prediction module is used to input the clinical data information, the magnetic resonance diffusion-weighted image, and the magnetic resonance high-resolution T2-weighted image into a pre-trained prediction model, so that the prediction model can predict and analyze tumor heterogeneity based on the clinical data information, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to obtain tumor heterogeneity characterization information. The quantitative parameters of tumor cell structure and function are obtained by fitting the image features representing the region of interest of the tumor in the magnetic resonance diffusion-weighted image, and the tumor fractal dimension is obtained by measuring the box dimension method based on the magnetic resonance high-resolution T2-weighted image. The heterogeneity analysis module is used to perform quantitative analysis of cell structure and function based on the tumor heterogeneity characterization information predicted by the first prediction module.

2. The system according to claim 1, characterized in that, The system further includes: a first image post-processing module; The first image post-processing module is used to analyze the tumor microscopic features in the magnetic resonance diffusion-weighted image based on the imaging model after the first image acquisition module acquires the magnetic resonance diffusion-weighted image, and to fit the tumor microscopic features to a preset number of least squares curves to obtain quantitative parameters of tumor cell structure and function. The quantitative parameters of tumor cell structure and function include the average cell diameter, intracellular volume fraction, extracellular diffusion rate, cell density, and water molecule transmembrane permeability in the tumor region of interest.

3. The system according to claim 1, characterized in that, The system also includes: a second image post-processing module; The second image post-processing module is used to select multiple images from the high-resolution T2-weighted magnetic resonance images obtained by the second image acquisition module to extract the tumor region of interest and draw a binary map, and use the box dimension method to determine the fractal dimension of the binary map to obtain the tumor fractal dimension.

4. The system according to claim 3, characterized in that, The second image post-processing module includes: an extraction unit and a measurement unit; The extraction unit is used to select three images from the highest-resolution T2-weighted magnetic resonance image: the largest tumor layer and two adjacent head and tail layers. The tumor region of interest is extracted from the three images, and the tumor region of interest is plotted as a binary image. The measurement unit is used to measure the fractal dimension of the binary image multiple times using the box dimension method, and select the mean of the fractal dimensions obtained from multiple measurements to obtain the tumor fractal dimension.

5. The system according to claim 1, characterized in that, The system also includes: a comprehensive analysis module; The comprehensive analysis module is used to perform univariate regression analysis on the clinical data information of the patient to be predicted after the data acquisition module obtains the clinical data information to obtain the clinical factors that independently affect the characterization of tumor heterogeneity. The clinical factors, the quantitative parameters of tumor cell structure and function, and the fractal dimension of the tumor are subjected to multivariate logistic regression analysis to obtain the cell structure, functional parameters and clinical factors that independently affect the characterization of tumor heterogeneity.

6. The system according to any one of claims 1-5, characterized in that, The system also includes: a training sample acquisition module, a model training module, and a model building module; The training sample acquisition module is used to acquire training sample data of the machine learning model before inputting the clinical data information, the magnetic resonance diffusion-weighted image, and the magnetic resonance high-resolution T2-weighted image into the pre-trained prediction model. The training sample data includes sample input data and sample output data. The model training module is used to input the training sample data obtained by the training sample acquisition module into the machine learning model, so as to train the mapping relationship between the sample input data and the sample output data through the machine learning model; The model building module is used to build a prediction model based on the mapping relationship trained by the model training module when the training meets the iteration stopping condition.

7. The system according to claim 6, characterized in that, The training sample acquisition module includes: an input data construction unit and an output data construction unit; The input data construction unit is used to construct sample input data based on clinical data of different patient samples, magnetic resonance diffusion-weighted images with tumor regions of interest marked, and magnetic resonance high-resolution T2-weighted images with tumor regions of interest marked. The output data construction unit is used to construct sample output data based on the tumor heterogeneity characterization information obtained from the analysis of patient samples through pathological specimens.

8. The system according to claim 6, characterized in that, The machine learning models in the model training module include, but are not limited to, any one of the following: decision tree method, random forest method, support vector machine method, and Naive Bayes method.

9. The system according to claim 6, characterized in that, The system also includes: a verification parameter acquisition module, a comparison module, and an evaluation module; The verification parameter acquisition module is used to acquire tumor heterogeneity characterization information obtained from pathological specimen analysis of the patient to be verified after the prediction model is constructed. The comparison module is used to compare the tumor heterogeneity characterization information obtained by the verification parameter acquisition module through pathological specimen analysis of the patient to be verified with the tumor heterogeneity characterization information predicted by the prediction model, and obtain the receiver operating characteristic curve predicted by the model. The evaluation module is used to evaluate the prediction accuracy of the pre-trained prediction model based on the receiver operating characteristic curve predicted by the model.

10. A method for tumor heterogeneity analysis based on magnetic resonance cellular microstructure imaging, characterized in that, include: Clinical data of the patient to be predicted, diffusion-weighted magnetic resonance images with regions of interest marked on the tumor, and high-resolution T2-weighted magnetic resonance images with regions of interest marked on the tumor were obtained respectively. The clinical data, the diffusion-weighted magnetic resonance image, and the high-resolution T2-weighted magnetic resonance image are input into a pre-trained prediction model. The prediction model then uses the clinical data, quantitative parameters of tumor cell structure and function, and tumor fractal dimension to predict and analyze tumor heterogeneity, thereby obtaining tumor heterogeneity characterization information. The quantitative parameters of tumor cell structure and function are obtained by fitting image features representing regions of interest in the tumor in the diffusion-weighted magnetic resonance image. The tumor fractal dimension is measured using the box-counting method based on the high-resolution T2-weighted magnetic resonance image. Quantitative analysis of cell structure and function was performed based on the predicted tumor heterogeneity characterization information. After acquiring a magnetic resonance diffusion-weighted image with the tumor region of interest labeled, the method further includes: Based on the imaging model, the tumor microscopic features in the magnetic resonance diffusion-weighted image are analyzed, and the tumor microscopic features are fitted with the least squares curve for a preset number of times to obtain quantitative parameters of tumor cell structure and function. The quantitative parameters of tumor cell structure and function include the average cell diameter, intracellular volume fraction, extracellular diffusion rate, cell density, and water molecule transmembrane permeability in the tumor region of interest. After acquiring high-resolution T2-weighted magnetic resonance images with tumor regions of interest labeled, the method further includes: Multiple images are selected from the high-resolution T2-weighted magnetic resonance images to extract the tumor region of interest and draw a binary map. The fractal dimension of the binary map is determined using the box-counting method to obtain the tumor fractal dimension. Specifically, this includes: selecting three images from the highest tumor level and its two adjacent head and tail levels in the high-resolution T2-weighted magnetic resonance images; extracting the tumor region of interest from the three images; drawing the tumor region of interest as a binary map; and repeatedly measuring the fractal dimension of the binary map using the box-counting method. The mean of the fractal dimensions obtained from the multiple measurements is then selected to obtain the tumor fractal dimension. After obtaining the clinical data of the patient to be predicted, the method further includes: Univariate regression analysis was performed on the clinical data to obtain the clinical factors that independently affect the characterization of tumor heterogeneity. Multivariate logistic regression analysis was then performed on the clinical factors, the quantitative parameters of tumor cell structure and function, and the fractal dimension of the tumor to obtain the cell structure, functional parameters, and clinical factors that independently affect the characterization of tumor heterogeneity. Before inputting the clinical data, the diffusion-weighted magnetic resonance imaging (MRI) image, and the high-resolution T2-weighted MRI image into the pre-trained prediction model, the method further includes: The training sample data for the machine learning model is obtained. The training sample data includes sample input data and sample output data. Specifically, it includes: constructing sample input data based on clinical data of different patient samples, magnetic resonance diffusion-weighted images with tumor regions of interest (MRI) labeled, and magnetic resonance high-resolution T2-weighted images with tumor MRI labeled; and constructing sample output data based on tumor heterogeneity characterization information obtained from pathological specimen analysis of patient samples. The training sample data obtained by the training sample acquisition module is input into the machine learning model to train the mapping relationship between the sample input data and the sample output data through the machine learning model. The machine learning model includes, but is not limited to, any one of the following: decision tree method, random forest method, support vector machine method, and Naive Bayes method. When the training meets the iteration stopping condition, a prediction model is constructed based on the mapping relationship obtained by the model training module. After constructing the prediction model, the method further includes: Obtain tumor heterogeneity characterization information from pathological specimen analysis of patients to be validated; The tumor heterogeneity characterization information obtained from the pathological specimen analysis of the patient to be verified is used as the measured result and compared with the tumor heterogeneity characterization information predicted by the prediction model to obtain the receiver operating characteristic curve predicted by the model. The accuracy of the pre-trained prediction model is evaluated based on the receiver operating characteristic curve predicted by the model.

Images