MRI (Magnetic Resonance Imaging) image-based early-stage cerebral ventricular paraphrine softening risk prediction method and system

Abstract

The invention relates to an MRI image-based early-stage paraphrase softening risk prediction method and an MRI image-based early-stage paraphrase softening risk prediction system. The method comprises the following steps: acquiring a clinical MRI image and a clinical medical history of a to-be-predicted risk patient; preprocessing the clinical MRI image to obtain a to-be-detected MRI image; and inputting a to-be-detected MRI image and the clinical medical history into the ventricular paraphrine softening prediction model to obtain a ventricular paraphrine softening risk prediction result. By adopting the method, the disease risk of the child patient can be reflected more comprehensively through the risk prediction result, and more accurate support is provided for clinical decision making.

Description

Technical Field

[0001] The present invention belongs to the field of auxiliary medical diagnosis, and in particular relates to a method and system for predicting the risk of early periventricular leukomalacia based on MRI images. Background Art

[0002] With the development of medical imaging technology, magnetic resonance imaging (MRI) has been widely used in clinical diagnosis. MRI has the characteristics of multi-parameter and multi-sequence imaging, which can clearly display the anatomical structure and functional information of human soft tissue. In the diagnosis of pediatric neurological diseases, especially pediatric brain lesions, MRI provides clinicians with rich imaging basis.

[0003] The diagnosis of periventricular leukomalacia (PVL) in children mainly relies on clinical symptom observation and the doctor's subjective visual assessment of MRI images, combined with the child's motor development and clinical experience to determine the possibility of the child having PVL.

[0004] However, in clinical symptom observation, the early symptoms of children with PVL are often atypical, and some mild neurological abnormalities are easily overlooked, leading to delayed and missed diagnosis. Due to differences in clinical experience and judgment criteria among different doctors, it is difficult to accurately judge early subtle lesions, which is not conducive to early detection and intervention of PVL in children, thereby affecting the prognosis and quality of life of the children. Summary of the Invention

[0005] Based on this, it is necessary to provide an early ventricular leukomalacia risk prediction method and system based on MRI images, which can provide PVL risk warnings and assist doctors in strengthening PVL supervision in children, in order to address the above technical problems.

[0006] In a first aspect, the present application provides a method for predicting the risk of early periventricular leukomalacia based on MRI images, comprising:

[0007] Obtain clinical MRI images and clinical history of patients whose risk is to be predicted;

[0008] Preprocessing the clinical MRI image to obtain the MRI image to be tested;

[0009] The MRI images to be tested and the clinical history are input into the periventricular leukomalacia prediction model to obtain the periventricular leukomalacia risk prediction results.

[0010] In one embodiment, preprocessing a clinical MRI image to obtain an MRI image to be detected includes:

[0011] Performing standardization correction on clinical MRI images to obtain standardized clinical MRI images; wherein the clinical MRI images include T1WI sequence MRI images, T2WI sequence MRI images and FLAEIR sequence MRI images;

[0012] The target region is segmented on the standardized clinical MRI image to obtain the MRI image to be detected; the target region includes the white matter region including the lateral ventricle level.

[0013] In one embodiment, the periventricular leucomelia prediction model is based on the following method to obtain the periventricular leucomelia risk prediction result:

[0014] Perform feature extraction on the MRI image to be tested and the clinical history to obtain the final feature vector;

[0015] The final feature vector is screened for periventricular leukomalacia-related features to obtain relevant feature vectors;

[0016] Logistic regression prediction was performed based on the relevant eigenvectors to obtain the risk prediction results of periventricular leukomalacia; the risk prediction results of periventricular leukomalacia included the risk of periventricular leukomalacia and the risk of no periventricular leukomalacia.

[0017] In one embodiment, feature extraction is performed on the MRI image to be detected and the clinical history to obtain a final feature vector, including:

[0018] Perform radiomics feature extraction on the MRI image to be tested to obtain image features; image features include morphological features, first-order statistical features and texture features;

[0019] Extract the data features of periventricular leukomalacia from the clinical history to obtain clinical characteristics;

[0020] The imaging features are fused with the clinical features to obtain the final feature vector.

[0021] In one embodiment, radiomics feature extraction is performed on the MRI image to be detected to obtain image features, including:

[0022] Calculate the volume, surface area and sphericity of the white matter region of the MRI image to be tested to obtain morphological features;

[0023] Extracting the MRI signal intensity distribution characteristics of the MRI image to be detected to obtain first-order statistical features; the MRI signal intensity distribution characteristics include the MRI signal intensity mean, the MRI signal intensity standard deviation, and the MRI signal intensity distribution skewness;

[0024] Calculate the gray level co-occurrence matrix and gray level run length matrix of the MRI image to be detected to obtain texture features;

[0025] The image features are obtained by fusing morphological features, first-order statistical features and texture features;

[0026] The morphological characteristics are obtained by the following formula:

[0027]

[0028]

[0029]

[0030] S={V,A,Sphericity}

[0031] Among them, S is the morphological feature; V is the volume of the white matter area; v i is the volume of each voxel; N is the number of voxels in the white matter region; A is the surface area of ​​the white matter region; a j is the area of ​​each grid cell on the surface of the white matter region; M is the total number of grid cells; Sphericity is the sphericity, and the sphericity value range is (0,1], the closer to 1, the closer to a sphere;

[0032] The first-order statistical characteristics are obtained through the following formula:

[0033]

[0034]

[0035]

[0036] Z={μ,σ,Skewness}

[0037] Where Z is the first-order statistical feature; μ is the mean of MRI signal intensity; σ is the standard deviation of MRI signal intensity; Skewness is the skewness of MRI signal intensity distribution; x i is the signal intensity value of the i-th voxel in the white matter area.

[0038] In one embodiment, the final feature vector is screened for periventricular leukomalacia-related features to obtain relevant feature vectors, including:

[0039] The mutual information was used to calculate the information sharing degree between the final feature vector and the periventricular leukomalacia trait;

[0040] The final feature vector corresponding to the information sharing degree value exceeding the preset threshold is determined as the relevant feature vector.

[0041] In one embodiment, a logistic regression prediction is performed based on the relevant feature vectors to obtain a risk prediction result of periventricular leukomalacia, including:

[0042] Use the logistic regression decision function to output the relevant feature vector to represent the probability of the predicted result category;

[0043] The risk prediction results of periventricular leukomalacia are obtained according to the probability of the prediction result category;

[0044] Use the following logistic regression decision function to get the probability of the predicted result category:

[0045] P(Y=1|X)=ρ(W T X+b)

[0046]

[0047] z=W T X+b

[0048] Where P(Y = 1|X) is the probability of the predicted result category; X is the relevant eigenvector; W is the weight parameter of the periventricular leucomalacia prediction model; b is the bias term; ρ(z) is the Sigmiod function; and z is a linear combination.

[0049] In one embodiment, the periventricular leukomalacia prediction model is constructed by the following method:

[0050] Obtain clinical MRI images, clinical history, and tracking labels of children clinically suspected of having movement abnormalities; construct a one-to-one corresponding training set of clinical MRI images, clinical history, and tracking labels based on preset inclusion and exclusion criteria;

[0051] The training set was used to iteratively train the logistic regression model according to the gradient descent algorithm to obtain the parameters of the periventricular leucomalacia prediction model;

[0052] The parameters of the periventricular leukomalacia prediction model were adjusted through the loss function and optimization algorithm until the loss function converged to obtain the periventricular leukomalacia prediction model.

[0053] In a second aspect, the present application also provides an early-stage periventricular leukomalacia risk prediction system based on MRI images, comprising:

[0054] A data acquisition module, used to obtain clinical MRI images and clinical medical history of patients whose risk is to be predicted;

[0055] A data processing module is used to pre-process clinical MRI images to obtain MRI images to be tested;

[0056] The model algorithm module is used to input the MRI image to be tested and the clinical history into the periventricular leukomalacia prediction model to obtain the periventricular leukomalacia risk prediction results.

[0057] In a third aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of any of the above-mentioned methods for predicting the risk of early periventricular leukomalacia based on MRI images.

[0058] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-mentioned methods for predicting the risk of early periventricular leukomalacia based on MRI images.

[0059] The above-mentioned MRI-based early PVL risk prediction method and system obtains clinical MRI images and clinical history, and uses the PVL prediction model to derive PVL risk prediction results. It can assess the potential risk of PVL. Compared with the traditional method of relying on symptoms to judge, which often does not become noticeable until the disease develops to a certain extent, this application more comprehensively reflects the disease risk of children through risk prediction results, provides more accurate support for clinical decision-making, alerts doctors to conduct further observation, and helps children with PVL detect the disease early to gain valuable treatment time. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0061] Figure 1 Schematic diagram of the process of the method for predicting the risk of early periventricular leukomalacia based on MRI images of the present invention;

[0062] Figure 2 Schematic diagram of the step-by-step process of step S103;

[0063] Figure 3 Schematic diagram of the step-by-step process of step S201;

[0064] Figure 4 This is a structural diagram of the early periventricular leukomalacia risk prediction system based on MRI images of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0066] In one embodiment, Figure 1 As shown, a method for predicting the risk of early periventricular leukomalacia based on MRI images is provided. This embodiment uses the method applied to a terminal as an example. It is understood that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0067] S101. Obtain clinical MRI images and clinical medical history of a patient whose risk is to be predicted.

[0068] Clinical MRI images can visually reveal information about brain tissue structure, including its morphology, structure, and organizational characteristics. They also have strong soft tissue resolution, helping to detect abnormal changes in the periventricular white matter. Furthermore, some MRI imaging techniques, including diffusion tensor imaging, can reveal the integrity and directionality of white matter fibers. PVL, which disrupts white matter fiber bundles, can provide a strong basis for predicting the risk of periventricular leucomalacia. Furthermore, MRI uses no ionizing radiation, making it suitable for examining children and for repeated monitoring when there is a risk of periventricular leucomalacia.

[0069] Clinical history mainly includes birth weight, prematurity, neonatal Apgar score and blood oxygen level. Birth weight and prematurity are related to the risk of PVL. The risk of PVL in premature infants with a gestational age of less than 32 weeks is significantly increased, and the incidence of PVL is higher in infants weighing less than 1500 grams. The neonatal Apgar score is the history of neonatal asphyxia. Asphyxia can cause cerebral ischemia, directly damage the periventricular white matter, and increase the risk of PVL.

[0070] Visualizing white matter damage through clinical MRI images can directly reflect the risk of PVL. Further using clinical history to provide relevant information, it helps identify high-risk patients and improves prediction accuracy.

[0071] S102: Preprocess the clinical MRI image to obtain an MRI image to be detected.

[0072] Since the original clinical MRI images may have quality problems due to factors such as equipment differences, patient movement, and imaging environment, including image blur, inconsistent signal intensity, and geometric deformation, the preprocessing of clinical MRI images can eliminate or reduce interference factors, so that the image quality reaches a standard suitable for subsequent analysis and prediction, and the accuracy and reliability of feature extraction are improved, thereby improving the performance of the prediction model. For example, motion correction is performed using a rigid registration algorithm. Using the image of a certain stable layer as a reference, the images of other layers that are offset due to patient movement are aligned to correct the spatial position deviation of the image. Then, the histogram matching method is used for intensity correction to unify the grayscale values ​​of images obtained by different MRI devices into the same standard range to ensure the comparability of signal intensity between images. Finally, image standardization is performed to resample the image resolution to a unified standard of 1mm×1mm×1mm to facilitate subsequent feature extraction.

[0073] S103 , inputting the MRI image to be detected and the clinical history into the periventricular leukomalacia prediction model to obtain a periventricular leukomalacia risk prediction result.

[0074] The PVLM prediction model is trained on a large amount of existing clinical MRI image and clinical history data. By learning from this data, the model automatically extracts features related to PVLM and establishes a relationship between these features and disease risk. When the model inputs the MRI image to be tested and the clinical history, it outputs a risk prediction result for PVLM based on this mapping relationship.

[0075] The above-mentioned MRI-based early PVL risk prediction method uses MRI image analysis to detect white matter damage characteristics related to PVL, and combines clinical history to predict the risk of PVL, identify high-risk patients early, and guide high-risk patients and doctors to conduct continuous supervision, so as to carry out intervention treatment as soon as possible and reduce the long-term neurological damage caused by PVL. By integrating MRI imaging genomics features and clinical history data, combined with comprehensive analysis of the PVL prediction machine learning model, the accuracy of PVL risk prediction is improved, the objectivity of PVL risk assessment is improved, and doctors' diagnosis is assisted, which improves diagnostic efficiency.

[0076] In one embodiment, preprocessing a clinical MRI image to obtain an MRI image to be detected includes:

[0077] S21. Perform standardization correction on the clinical MRI image to obtain a standardized clinical MRI image; wherein the clinical MRI image includes a T1WI sequence MRI image, a T2WI sequence MRI image, and a FLAEIR sequence MRI image.

[0078] T1WI sequence MRI images primarily reflect differences in the longitudinal relaxation times of tissues. Specifically, adipose tissue appears as a high signal, or white, while cerebrospinal fluid appears as a low signal, or black. T1WI can clearly display anatomical details, distinguishing gray matter from white matter, allowing for further observation of normal brain structure and the presence of structural malformations and space-occupying lesions. For example, in the early stages of PVL, the periventricular white matter region exhibits a decreased signal on T1WI due to increased local brain tissue water content and cytotoxic edema. Furthermore, if softening lesions are present, these areas will exhibit a lower signal, contrasting more clearly with surrounding normal tissue, and their boundaries may gradually become clearer.

[0079] T2WI (T2-weighted MRI) sequences highlight differences in transverse relaxation times of tissues. Specifically, water appears as a high signal, or white, while fat appears as a low signal, or black. T2WI is particularly sensitive in diagnosing brain diseases. For example, PVL (polysaccharidosis) (PVL) appears as a high signal on T2WI. Due to increased water content in the PVL lesion, the high signal may be scattered and small in the early stages. As the disease progresses, the high signal areas may expand and merge.

[0080] FLAEIR sequence MRI images are fluid-attenuated inversion recovery sequences, or special T2WI sequence MRI images. They primarily suppress cerebrospinal fluid signals, making lesions near the cerebrospinal fluid appear more clearly. Schematically, FLAEIR sequence MRI images suppress the high signal of the cerebrospinal fluid, making lesions in the periventricular white matter more prominent. In the early stages of PVL, FLAIR images can be found as scattered, poorly defined, high-signal lesions in the periventricular white matter. These lesions may be difficult to distinguish on T2WI due to interference from the high signal of the cerebrospinal fluid. As the disease worsens, the high-signal areas become more pronounced, and the relationship between the lesions and the ventricles can be clearly shown.

[0081] Clinical MRI images may have inconsistent image intensity and spatial resolution due to factors such as different equipment, different scanning parameters, and individual differences. The purpose of standardization correction is to eliminate these differences, make images from different sources comparable, and facilitate subsequent feature extraction and analysis. For MRI images of T1WI, T2WI, and FLAIR sequences, they each reflect different tissue characteristics, but before comprehensive analysis, their intensity and spatial scale need to be unified. Schematically, the histogram matching method is used for intensity normalization. For each T1WI, T2WI, and FLAIR sequence image to be corrected, its grayscale histogram is calculated and matched with the histogram of the reference image. Furthermore, the spatial resolution of all images is resampled to a unified standard, such as 1mm×1mm×1mm. The images are resampled using linear interpolation to ensure image continuity and accuracy.

[0082] S22. Segment the target region of the standardized clinical MRI image to obtain the MRI image to be detected; the target region includes the white matter region including the lateral ventricle level.

[0083] Target region segmentation aims to extract specific areas relevant to the study from standardized clinical MRI images, specifically the white matter region encompassing the lateral ventricles. As the primary site of PVL lesions, accurately segmenting this region facilitates subsequent lesion feature extraction and analysis. Segmentation reduces interference from irrelevant regions and improves the accuracy and specificity of the analysis.

[0084] Exemplarily, a semi-automatic segmentation method is combined with threshold segmentation and region growing. Specifically, a suitable threshold range is set based on the grayscale characteristics of the white matter region in different sequence images, and pixels in the image that meet the threshold conditions are preliminarily marked as white matter regions. Furthermore, using the preliminarily marked pixels as seed points, a region growing algorithm is used to merge adjacent pixels with similar grayscale values ​​into the white matter region to obtain a white matter region including the lateral ventricle level.

[0085] For example, fully automated segmentation can be performed using a deep learning model. A convolutional neural network model, such as U-Net, can be trained on a large dataset of annotated MRI images to learn the characteristics and segmentation patterns of white matter regions. The trained model can then quickly and accurately segment new standardized MRI images.

[0086] In one embodiment, Figure 2 As shown in Figure 2, the periventricular leukomalacia prediction model is based on the following method to obtain the risk prediction results of periventricular leukomalacia:

[0087] S201 , extract features from the MRI image to be detected and the clinical history to obtain a final feature vector.

[0088] Multiple features are extracted from the MRI image and clinical history to reflect the patient's condition from different perspectives. MRI image features, such as texture, shape, and intensity, can reveal differences in brain tissue microstructure, morphology, and signal intensity. Clinical history features include information closely related to the development and progression of PVL, such as the patient's birth date and past medical history. MRI image features and clinical history features are integrated to form a comprehensive final feature vector, providing a comprehensive data foundation for subsequent analysis and prediction.

[0089] S202: Perform paraventricular leukomalacia-related screening on the final feature vector to obtain a relevant feature vector.

[0090] Schematically, the features included in the final feature vector have varying degrees of relevance to PVL, potentially including redundant or irrelevant low-relevance features. These low-relevance features can interfere with the model's predictive accuracy and efficiency. Through a specific screening method, features with low relevance to PVL are removed, while features that are important for predicting PVL are retained. This reduces feature dimensionality, improves model performance, and makes subsequent predictions more accurate and efficient.

[0091] S203. Perform logistic regression prediction based on the relevant eigenvectors to obtain a periventricular leukomalacia risk prediction result; the periventricular leukomalacia risk prediction result includes the presence of periventricular leukomalacia risk and the absence of periventricular leukomalacia risk.

[0092] Logistic regression is a statistical model for binary classification problems. It links feature vectors to probability values ​​by constructing a logistic function. In PVL risk prediction, the logistic regression model learns a probability model for PVL occurrence based on the relevant feature vectors. When a new relevant feature vector is input, the model calculates the probability that the patient sample being predicted belongs to the PVL group or the non-PVL group, and determines the final risk prediction result based on a set probability threshold.

[0093] In one embodiment, Figure 3 As shown, feature extraction is performed on the MRI image to be detected and the clinical history to obtain the final feature vector, including:

[0094] S301 , performing radiomics feature extraction on the MRI image to be detected to obtain image features; the image features include morphological features, first-order statistical features, and texture features.

[0095] Schematically, the MRI images being examined contain a wealth of information related to PVL. Radiomics aims to extract this information through quantitative analysis. Morphological features reflect the shape and size of the target region, visually demonstrating the morphological changes in the periventricular white matter. First-order statistical features, based on the distribution of image grayscale values, describe the overall grayscale characteristics of the image. Texture features reflect the spatial distribution of grayscale within the image, reflecting differences in tissue microstructure. These imaging features reflect tissue characteristics at different levels, aiding subsequent models in discovering imaging markers associated with PVL.

[0096] S302. Extract data features related to periventricular leukomalacia from the clinical history to obtain clinical features.

[0097] Similarly, clinical histories contain information related to the development and progression of PVL. By combing and quantifying these clinical histories, we can extract features that reflect the patient's risk of developing the disease. These clinical features can be direct influencing factors, such as gestational age and birth weight, or indirect correlates, such as a history of neonatal asphyxia or infection. These factors are quantified and converted into numerical features that can be used for analysis, providing a basis for subsequent predictions. For example, gestational age and birth weight are represented as actual numerical values, while binary features such as a history of neonatal asphyxia and infection are recorded as 1 if present and 0 if absent, forming a clinical feature vector.

[0098] S303: Fuse the image features with the clinical features to obtain a final feature vector.

[0099] Imaging features and clinical characteristics reflect the likelihood of PVL in patients from different perspectives. Fusion of the two provides more comprehensive information. Schematically, by combining imaging and clinical features in a certain order, a comprehensive final feature vector is formed, providing richer and more accurate data for subsequent model predictions.

[0100] In one embodiment, radiomics feature extraction is performed on the MRI image to be detected to obtain image features, including:

[0101] S41. Calculate the volume, surface area, and sphericity of the white matter region of the MRI image to be detected to obtain morphological features.

[0102] Schematically, for the white matter area in the MRI image, the volume reflects the size of the white matter area, the surface area reflects the range of its surface, and the sphericity measures the similarity of its shape to an ideal sphere. This is to detect the morphological differences between the white matter area and the normal state, evaluate the size and shape changes of the pathology, and thus assist in determining whether there are lesions related to periventricular leucomelia.

[0103] S42. Extracting MRI signal intensity distribution characteristics of the MRI image to be detected to obtain first-order statistical features; the MRI signal intensity distribution characteristics include MRI signal intensity mean, MRI signal intensity standard deviation, and MRI signal intensity distribution skewness.

[0104] Schematically, the first-order statistical features are calculated based on the distribution of signal intensity values ​​in the MRI image. The mean of MRI signal intensity reflects the overall signal intensity level of the white matter area, the standard deviation measures the discreteness of the signal intensity values, and reflects the fluctuation of the signal intensity; the skewness describes the asymmetry of the signal intensity distribution. The first-order statistical features can reflect the uniformity of the tissue in the white matter area and the trend of changes in signal intensity. When the white matter is diseased, its signal intensity distribution will change. By analyzing these features, potential lesion information can be discovered. For example, in the T2W1 sequence MRI images, the water signal is higher, and the mean of the lesion area is higher; the larger the standard deviation, the greater the change in signal intensity, indicating that there is heterogeneity in the white matter area and there may be lesions;

[0105] S43. Calculate the gray-level co-occurrence matrix and the gray-level run length matrix of the MRI image to be detected to obtain texture features.

[0106] Schematically, the grayscale co-occurrence matrix is ​​used to describe the spatial distribution of different grayscale pairs in an image, reflecting the texture information of the image, including features such as contrast, correlation, energy, and entropy. Contrast reflects the degree of grayscale difference in the image. The higher the contrast, the greater the grayscale variation distance between pixels and the unclear white matter boundary. Correlation reflects the degree of linear correlation between grayscale pairs. Energy indicates the uniformity of the image grayscale distribution, and entropy measures the complexity of the image texture. The grayscale run length matrix records the length information of consecutive pixels with the same grayscale value in the image, and can obtain features such as short run advantage and long run advantage to reflect the thickness and distribution pattern of the texture in the image. Through texture features, changes in the microstructure of the white matter region can be found. For example, PVL will cause changes in the white matter microstructure, which in turn causes changes in texture features.

[0107] S44. The morphological features, first-order statistical features and texture features are integrated to obtain image features.

[0108] Morphological, first-order statistical, and texture features describe the imaging characteristics of white matter regions in MRI images from different perspectives. Morphological features reflect macroscopic shape, first-order statistical features reveal signal intensity distribution, and texture features reveal microscopic structure. Fusion of multiple features can construct a more comprehensive and integrated image feature vector that more accurately represents the state of the white matter region.

[0109] The morphological characteristics are obtained by the following formula:

[0110]

[0111]

[0112]

[0113] S={V,A,Sphericity}

[0114] Among them, S is the morphological feature; V is the volume of the white matter area; v i is the volume of each voxel; N is the number of voxels in the white matter region; A is the surface area of ​​the white matter region; a j is the area of ​​each grid cell on the surface of the white matter region; M is the total number of grid cells; Sphericity is the sphericity, and the sphericity value range is (0,1], and the closer to 1, the closer to a sphere.

[0115] The first-order statistical characteristics are obtained through the following formula:

[0116]

[0117]

[0118]

[0119] Z={μ,σ,Skewness}

[0120] Where Z is the first-order statistical feature; μ is the mean of MRI signal intensity; σ is the standard deviation of MRI signal intensity; Skewness is the skewness of MRI signal intensity distribution; x i is the signal intensity value of the i-th voxel in the white matter area.

[0121] In one embodiment, the final feature vector is screened for periventricular leukomalacia-related features to obtain relevant feature vectors, including:

[0122] S51. Calculate the information sharing degree between the final feature vector and the paraventricular leucomalacia trait using mutual information.

[0123] Mutual information is an indicator used to measure the degree of mutual dependence between two random variables. In radiomics analysis, it is used to evaluate the correlation between a feature and a disease label and to select the most discriminative feature. Where X is the final feature vector; Y is the target vector for whether PVL occurs; p(x,y) is the joint probability distribution of X and Y; p(x) is the marginal probability distribution of X; p(y) is the marginal probability distribution of Y; if MI(X,Y)=0, it means that a part of the final feature vector is not helpful for PVL prediction and can be eliminated; if MI(X,Y)>0, it means that a part of the final feature vector is helpful for PVL prediction and can be retained, and the higher the MI(X,Y), the greater its contribution to PVL prediction.

[0124] S52: Determine the final feature vector corresponding to the information sharing degree value exceeding the preset threshold as the relevant feature vector.

[0125] Illustratively, after calculating the mutual information value between each feature in the final feature vector and the PVL trait, it is compared with a preset threshold. If the mutual information value of a feature exceeds the preset threshold, it indicates a high degree of information sharing between the feature and the PVL trait and contains a large amount of PVL-related information, and is retained as part of the relevant feature vector. Features with mutual information values ​​below the preset threshold are considered to have a weak association with the PVL trait and may introduce noise or interfere with model accuracy, so they are removed. This method can reduce the dimensionality of the feature vector, improving model training efficiency and prediction accuracy.

[0126] In one embodiment, a logistic regression prediction is performed based on the relevant feature vectors to obtain a risk prediction result of periventricular leukomalacia, including:

[0127] S61. Use the logistic regression decision function to output the relevant feature vector to represent the probability of the prediction result category.

[0128] Schematically, the logistic regression decision function maps the relevant feature vector to a probability value in the range [0,1], which represents the probability of the predicted result category.

[0129] S62. Obtain a risk prediction result of periventricular leukomalacia based on the probability of the prediction result category.

[0130] Exemplarily, the probability of the prediction result category corresponds to a binary classification decision. Exemplarily, if P(Y=1|X)>0.5, it indicates that there is a risk of periventricular leucomelia; if P(Y=1|X)≤0.5, it indicates that there is no risk of periventricular leucomelia.

[0131] Use the following logistic regression decision function to get the probability of the predicted result category:

[0132] P(Y=1|X)=ρ(W T X+b)

[0133]

[0134] z=W T X+b

[0135] Where P(Y = 1|X) is the probability of the predicted result category; X is the relevant eigenvector; W is the weight parameter of the periventricular leucomalacia prediction model; b is the bias term; ρ(z) is the Sigmiod function; and z is a linear combination.

[0136] In one embodiment, the periventricular leukomalacia prediction model is constructed by the following method:

[0137] S71. Obtain clinical MRI images, clinical history, and tracking labels of children clinically suspected of having movement abnormalities.

[0138] Children clinically suspected of having movement abnormalities may be at risk for periventricular leucorrhea (PVL). Clinical MRI images, clinical histories, and tracking labels are collected to provide a comprehensive data foundation for subsequent model training. Clinical MRI images can intuitively present brain structure and tissue information, and clinical histories contain various health-related information about the child from birth to the present, which is closely related to the occurrence and development of PVL. Tracking labels are determined based on the child's subsequent diagnostic results and are used to clarify the actual category of the sample (whether or not it has PVL), so that model training has a clear goal.

[0139] For example, in a children's hospital, pediatricians use a professional movement assessment scale to screen out children who are clinically suspected of having movement abnormalities. For these children, MRI examinations are arranged, and MRI images of T1WI, T2WI, and FLAIR sequences are collected using equipment such as Siemens, GE, or Philips, and the image quality is ensured to meet the requirements. At the same time, medical staff extract the child's clinical history from the hospital's electronic medical record system, including information such as the mother's pregnancy status, Apgar score at birth, whether respiratory distress syndrome occurred in the neonatal period, and whether there is a history of infection. The tracking label is determined based on the MRI follow-up results after the child is 2 years old. If PVL is confirmed, it is marked as 1; if PVL is not diagnosed, it is marked as 0. These images, medical history, and label information are stored one by one in the database to provide data support for subsequent steps.

[0140] S72. According to the preset exclusion criteria, the clinical MRI images, clinical medical history and tracking labels are used to form a one-to-one corresponding training set.

[0141] Pre-set inclusion and exclusion criteria were used to screen appropriate data and ensure the quality and reliability of the training set. The inclusion criteria selected cases potentially associated with PVL, while the exclusion criteria removed cases that would interfere with model training or were irrelevant to the study objectives. This resulting training set more accurately reflects the relationship between PVL and related factors, improving the effectiveness of model training.

[0142] For example, the inclusion criteria are children with clinically suspected movement abnormalities, and the exclusion criteria include patients with cerebral infarction or encephalomalacia involving more than one lobe, abnormal signals involving the cerebral cortex or neuronal migration disorders, incomplete MRI imaging or artifacts, PVL diagnosed on the initial MRI examination, incomplete follow-up, or lost to follow-up. The collected data are screened according to these inclusion and exclusion criteria to obtain cases that meet the requirements. The clinical MRI images, clinical history, and tracking labels of the qualified cases are mapped one by one to form a training set. The training set is divided into training and validation subsets according to a certain ratio for subsequent model training and evaluation.

[0143] S73. Use the training set to iteratively train the logistic regression model according to the gradient descent algorithm to obtain the parameters of the periventricular leucomalacia prediction model.

[0144] S74. Adjust the parameters of the periventricular leukomalacia prediction model using a loss function and an optimization algorithm until the loss function converges, thereby obtaining a periventricular leukomalacia prediction model.

[0145] During the training process, the model parameters, including weights and biases, must be determined. The gradient descent algorithm is used to optimize these parameters. By calculating the gradient of the loss function with respect to the model parameters, the parameters are gradually adjusted in the opposite direction of the gradient, resulting in a continuous decrease in the loss function. In each iteration, the gradient is calculated based on the samples in the training set, and the parameters are updated. After multiple iterations, the parameters gradually converge to an optimal value, thereby obtaining the parameters for the periventricular leucomelia prediction model and confirming the prediction model.

[0146] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0147] Based on the same inventive concept, embodiments of the present application also provide an MRI-based early ventricular leucomelia risk prediction system for implementing the aforementioned MRI-based early ventricular leucomelia risk prediction method. The solution provided by this system is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more of the following embodiments of the MRI-based early ventricular leucomelia risk prediction system can be found in the aforementioned limitations of the MRI-based early ventricular leucomelia risk prediction method and will not be further elaborated here.

[0148] In an exemplary embodiment, Figure 4 As shown, a risk prediction system for early periventricular leukomalacia based on MRI images is provided, comprising:

[0149] The data acquisition module is used to obtain clinical MRI images and clinical medical history of patients whose risk is to be predicted.

[0150] The data processing module is used to preprocess the clinical MRI image to obtain the MRI image to be detected.

[0151] The model algorithm module is used to input the MRI image to be tested and the clinical history into the periventricular leukomalacia prediction model to obtain the periventricular leukomalacia risk prediction results.

[0152] In one embodiment, the system further includes an image segmentation module for performing target region segmentation on the standardized clinical MRI image to obtain the MRI image to be detected.

[0153] In one embodiment, a feature extraction module is further included for extracting features from the MRI image to be detected and the clinical history to obtain a final feature vector;

[0154] The method further comprises a feature screening module for screening the final feature vector for paraventricular leukomalacia to obtain a relevant feature vector.

[0155] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0156] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0157] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0158] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A method for predicting the risk of early periventricular leukomalacia based on MRI imaging, characterized in that: The method comprises: Obtain clinical MRI images and clinical history of patients whose risk is to be predicted; Preprocessing the clinical MRI image to obtain an MRI image to be detected; The MRI image to be detected and the clinical history are input into a periventricular leukomalacia prediction model to obtain a periventricular leukomalacia risk prediction result.

2. The method according to claim 1, characterized in that The preprocessing of the clinical MRI image to obtain the MRI image to be detected includes: Performing standardization correction on the clinical MRI image to obtain a standardized clinical MRI image; wherein the clinical MRI image includes a T1WI sequence MRI image, a T2WI sequence MRI image, and a FLAEIR sequence MRI image; The standardized clinical MRI image is segmented into a target region to obtain an MRI image to be detected; the target region includes a white matter region including a lateral ventricle level.

3. The method according to claim 2, characterized in that The periventricular leukomalacia prediction model is based on the following method to obtain the periventricular leukomalacia risk prediction results: Performing feature extraction on the MRI image to be detected and the clinical history to obtain a final feature vector; Performing periventricular leukomalacia-related screening on the final feature vector to obtain a related feature vector; Logistic regression prediction is performed based on the relevant eigenvectors to obtain a periventricular leukomalacia risk prediction result; the periventricular leukomalacia risk prediction result includes the presence of periventricular leukomalacia risk and the absence of periventricular leukomalacia risk.

4. The method according to claim 3, characterized in that The feature extraction of the MRI image to be detected and the clinical history to obtain a final feature vector includes: Performing radiomics feature extraction on the MRI image to be detected to obtain image features; the image features include morphological features, first-order statistical features, and texture features; Extracting data features related to periventricular leukomalacia from the clinical history to obtain clinical features; The image features are fused with the clinical features to obtain a final feature vector.

5. The method according to claim 4, characterized in that The extracting of radiomics features from the MRI image to be detected to obtain image features includes: Calculating the volume, surface area, and sphericity of the white matter region of the MRI image to be detected to obtain morphological features; Extracting the MRI signal intensity distribution characteristics of the MRI image to be detected to obtain first-order statistical features; the MRI signal intensity distribution characteristics include the MRI signal intensity mean, the MRI signal intensity standard deviation, and the MRI signal intensity distribution skewness; Calculating the gray level co-occurrence matrix and the gray level run length matrix of the MRI image to be detected to obtain texture features; fusing the morphological features, the first-order statistical features, and the texture features to obtain image features; The morphological characteristics are obtained by the following formula: S={V,A,Sphericity} Among them, S is the morphological feature; V is the volume of the white matter area; v i is the volume of each voxel; N is the number of voxels in the white matter region; A is the surface area of ​​the white matter region; a j is the area of ​​each grid cell on the surface of the white matter region; M is the total number of grid cells; Sphericity is the sphericity, and the sphericity value range is (0,1], the closer to 1, the closer to a sphere; The first-order statistical characteristics are obtained through the following formula: Z={μ,σ,Skewness} Where Z is the first-order statistical feature; μ is the mean of MRI signal intensity; σ is the standard deviation of MRI signal intensity; Skewness is the skewness of MRI signal intensity distribution; x i is the signal intensity value of the i-th voxel in the white matter area.

6. The method according to claim 3, characterized in that The final feature vector is subjected to paraventricular leukomalacia-related screening to obtain a related feature vector, including: The mutual information was used to calculate the information sharing degree between the final feature vector and the periventricular leukomalacia trait; The final feature vector corresponding to the information sharing degree value exceeding a preset threshold is determined as the relevant feature vector.

7. The method according to claim 3, characterized in that The logistic regression prediction is performed based on the relevant eigenvectors to obtain a risk prediction result of periventricular leukomalacia, including: Outputting the relevant feature vectors to represent the probability of the predicted result category using a logistic regression decision function; Obtaining a periventricular leukomalacia risk prediction result according to the probability of the prediction result category; Use the following logistic regression decision function to get the probability of the predicted result category: P(Y=1|X)=ρ(W T X+b) z=W T X+b Where P(Y = 1|X) is the probability of the predicted result category; X is the relevant eigenvector; W is the weight parameter of the periventricular leucomalacia prediction model; b is the bias term; ρ(z) is the Sigmiod function; and z is a linear combination.

8. The method according to any one of claims 1 to 6, characterized in that The periventricular leukomalacia prediction model was constructed by the following method: Obtain clinical MRI images, clinical history, and tracking labels of children clinically suspected of having movement abnormalities; According to a preset exclusion criterion, the clinical MRI image, the clinical history, and the tracking label are used to form a one-to-one corresponding training set; Iteratively training a logistic regression model using the training set according to a gradient descent algorithm to obtain periventricular leukomalacia prediction model parameters; The parameters of the periventricular leukomalacia prediction model are adjusted by using a loss function and an optimization algorithm until the loss function converges to obtain a periventricular leukomalacia prediction model.

9. An early-stage periventricular leukomalacia risk prediction system based on MRI images, characterized in that: The system comprises: A data acquisition module, used to obtain clinical MRI images and clinical medical history of patients whose risk is to be predicted; A data processing module, configured to pre-process the clinical MRI image to obtain an MRI image to be detected; The model algorithm module is used to input the MRI image to be detected and the clinical history into the periventricular leukomalacia prediction model to obtain a periventricular leukomalacia risk prediction result.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images