A diagnostic aid system for craniocerebral injury based on medical images

Abstract

The present invention discloses a diagnosis auxiliary system for craniocerebral injury degree based on medical images, which relates to the field of medical technology. The initial judgment module can accurately judge whether the patient has the risk of interstitial brain edema by analyzing the degree of interstitial fluid accumulation in the white matter of the brain and quantitatively calculating the permeability abnormality coefficient Syxs, providing the ability of early risk screening, and can timely identify potential problems before the condition worsens, prompting clinicians to intervene quickly, thereby significantly improving the patient's prognosis. When there is a risk of interstitial brain edema, the type diagnosis module further distinguishes the type of lesion (such as brain contusion or other injuries) by extracting relevant distinguishing features from the multimodal feature set, and the calculation of the locking coefficient Sdxs provides a scientific quantitative basis for the diagnosis of the lesion type based on the specific characteristics of the lesion. This precise type diagnosis method can reduce diagnostic errors and provide reliable support for clinical decision-making.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a medical image-based diagnosis auxiliary system for the degree of craniocerebral injury. Background Art

[0002] In the field of diagnosis of the degree of craniocerebral injury, the core is to quantify the characteristics of craniocerebral lesions through accurate identification and comprehensive analysis of abnormal lesions in medical images, and to assist doctors in judging the severity of the injury and possible clinical intervention. Specifically speaking of the pathological state of interstitial cerebral edema, it often overlaps with the characteristics of other injury types such as cerebral ischemia and cerebral contusion, especially in terms of imaging manifestations, and is easily misdiagnosed as other pathological states. This misdiagnosis may ignore the potential pathological associations of interstitial cerebral edema (such as the spread of edema may cause intracranial hypertension). Therefore, it is necessary to conduct a comprehensive analysis combined with the characteristics of other injury types to improve the accuracy of diagnosis and the reliability of clinical judgment.

[0003] There are many deficiencies in the current clinical diagnosis of interstitial brain edema and its pathological differentiation from other types of injuries. On the one hand, traditional diagnosis mainly relies on the doctor's imaging experience, making subjective judgments on abnormalities in the white matter area of ​​the brain, and lacking systematic quantitative standards; in addition, the current clinical evaluation model focuses on the analysis of single lesion characteristics, resulting in insufficient systematic and accurate diagnosis. The abnormal consequences brought about by the above limitations include: high misdiagnosis rate, delayed treatment, and failure to identify potentially dangerous lesions in a timely manner, all of which may directly lead to worsening of the patient's condition, increased treatment costs, and even increased risk of death for patients. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a medical image-based auxiliary system for diagnosing the degree of craniocerebral injury, which solves the problems in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a craniocerebral injury degree diagnosis auxiliary system based on medical images, including an image data acquisition module, a preliminary judgment module, a type diagnosis module and a craniocerebral injury assessment module;

[0006] The image data acquisition module is used to acquire a group of cranial and brain medical images of the patient, and to identify the lesion area in the group of cranial and brain medical images through image preprocessing, and to obtain a multimodal feature set in the white matter of the patient's lesion area based on the lesion area;

[0007] The preliminary judgment module is used to preliminarily analyze the accumulation degree of interstitial fluid in the white matter of the brain according to the multimodal feature set to obtain the abnormal permeability coefficient Syxs, and diagnose whether there is a risk of interstitial cerebral edema in the current patient's brain based on the abnormal permeability coefficient Syxs, and if so, issue a recognition instruction;

[0008] The type diagnosis module is used to extract relevant distinguishing features in the multimodal feature set after receiving the recognition instruction, and obtain the locking coefficient Sdxs based on the relevant distinguishing features to diagnose the type of injury in the current patient's brain again;

[0009] The craniocerebral injury assessment module is used to construct a craniocerebral injury degree diagnosis model using convolutional neural network technology, and input the locking coefficient Sdxs and the permeability abnormality coefficient Syxs into the craniocerebral injury degree diagnosis model. After dimensionless processing, the injury assessment index Spzs is fitted and output, and displayed in a visual form.

[0010] Preferably, the image data acquisition module includes an image acquisition unit, an image preprocessing unit and a feature extraction unit;

[0011] The image acquisition unit is used to perform a cranial brain scan on the patient using a computer tomography scanner and a magnetic resonance imaging device in advance to acquire a cranial brain medical image set of the patient;

[0012] The image preprocessing unit is used to import the cranial brain medical image group into a DICOM file, and convert the DICOM file into a NIfTI format, and at the same time remove noise from the No. 1 cranial brain medical image obtained by a computer tomography scanner through a bilateral filtering method, and remove noise from the No. 2 cranial brain medical image obtained by a magnetic resonance imaging device through a non-local mean filtering method, and then normalize the grayscale value of the cranial brain medical image group, adjust the grayscale value range, and automatically segment the normalized cranial brain medical image group to identify and extract the lesion area in the cranial brain medical image group;

[0013] The cranial brain medical image group includes a first cranial brain medical image acquired by a computer tomography scanner and a second cranial brain medical image acquired by a magnetic resonance imaging device;

[0014] The feature extraction unit is used to process the cranial medical image group in advance to generate a gray level co-occurrence matrix, and extract a multimodal feature set in the white matter of the patient's lesion area according to the lesion area, wherein the multimodal feature set includes relevant fluid penetration features and relevant distinguishing features in the white matter of the brain, and the relevant fluid penetration features include fluid accumulation volume and the total volume of the lesion area ; The relevant distinguishing features include the density of the corresponding pixels in the lesion area , the total area of ​​the lesion , the total length of the lesion area boundary , the gray level Hdj of the corresponding pixel in the gray level co-occurrence matrix, and the average T1 signal intensity of the lesion area , the average T1 signal intensity of normal brain tissue and lesion diffusion .

[0015] Preferably, the preliminary judgment module includes a non-specific image performance analysis unit and a preliminary judgment unit;

[0016] The non-specific image performance analysis unit is used to analyze the abnormal accumulation of interstitial fluid infiltrating into the white matter area of ​​the brain according to the relevant fluid infiltration characteristics to obtain the water content ratio Shb. The water content ratio Shb is obtained by the following formula: In the formula, Expressed as the liquid accumulation volume, It is expressed as the total volume of the lesion area; combined with the cranial medical images of the lesion area, the average density of the lesion area is analyzed and obtained , the average density The specific method is as follows: In the formula, It is represented as the density of the i-th pixel in the lesion area, N is represented as the total number of pixels in the lesion area, i=1, 2, 3, ..., N.

[0017] Preferably, by combining the water content ratio Shb and the average density The permeability anomaly coefficient Syxs is calculated and obtained after dimensionless processing. The permeability anomaly coefficient Syxs is obtained by the following formula: In the formula, and are weight values, represents the first correction constant, where and The specific value is set by the user according to the situation.

[0018] Preferably, the preliminary judgment unit is used to extract a group of normal people's cranial and brain medical images from the hospital, and based on the group of normal people's cranial and brain medical images, set a standard threshold H, and compare and analyze it with the abnormal permeability coefficient Syxs to diagnose whether there is a risk of interstitial brain edema in the current patient's cranial and brain. The specific diagnosis content is as follows:

[0019] If the abnormal permeability coefficient Syxs exceeds the standard threshold value H, it is diagnosed that the patient has a risk of interstitial cerebral edema in the brain, and an identification instruction is sent to the type diagnosis module;

[0020] If the abnormal permeability coefficient Syxs does not exceed the standard threshold value H, it is diagnosed that there is no risk of interstitial cerebral edema in the current patient's brain. At this time, no identification instruction will be issued to the type diagnosis module, and the analysis process of the risk of interstitial cerebral edema in the patient's brain will be ended.

[0021] Preferably, the type diagnosis module includes an image feature analysis unit and a type recognition unit;

[0022] The image feature analysis unit is used to receive the recognition instruction, analyze the boundary characteristics of the image in the cranial medical image group according to the relevant distinguishing features in the multimodal feature set, so as to calculate and obtain the boundary clarity Bqcd and the texture contrast Wdbd respectively. The boundary clarity Bqcd and the texture contrast Wdbd are respectively obtained by the following formulas: In the formula, It is expressed as the total area of ​​the lesion region; It is expressed as the total length of the boundary of the lesion area; It is expressed as pi; In the formula, It is expressed as the joint probability of gray levels between adjacent pixels in the gray-level co-occurrence matrix; Represented as the gray level of the j-th pixel, It is represented as the gray level of the j+1th pixel, and j represents each pixel in the gray level co-occurrence matrix.

[0023] Preferably, the type identification unit is used to obtain a locking coefficient Sdxs by correlating the boundary definition Bqcd and the texture contrast Wdbd obtained in the image feature analysis unit and performing dimensionless processing, and the locking coefficient Sdxs is obtained by the following formula: In the formula, Expressed as signal intensity ratio, , and are weight values, among which, , and The specific value is set by the user according to the situation; among them, the signal strength ratio Obtained by the following formula: In the formula, It is expressed as the average T1 signal intensity of the lesion area. It is expressed as the average T1 signal intensity of normal brain tissue.

[0024] Preferably, a division threshold E is preset, and the locking coefficient Sdxs is compared with the division threshold E to diagnose the type of intracranial injury of the current patient again, and the specific contents are as follows:

[0025] If the locking coefficient Sdxs exceeds the division threshold E, it will be diagnosed that the current patient's intracranial injury type is not interstitial cerebral edema;

[0026] If the locking coefficient Sdxs does not exceed the division threshold E, the current patient's intracranial injury type will be diagnosed as interstitial cerebral edema.

[0027] Preferably, the craniocerebral injury assessment module includes a modeling unit and an evaluation output unit;

[0028] The modeling unit is used to construct an initial model using convolutional neural network technology, and use a multimodal feature set as the input of the initial model for training and testing, and use the trained initial model as a state recognition model to obtain feature information in the state recognition model respectively, and use the acquired feature information to train and test the state recognition model, and combined with the current patient's craniocerebral injury type, use the trained state recognition model as a craniocerebral injury degree diagnosis model.

[0029] Preferably, the evaluation output unit is used to fit and output an injury assessment index Spzs according to a craniocerebral injury degree diagnosis model, and the injury assessment index Spzs is obtained by the following formula: Where Syxs is the permeability anomaly coefficient, Expressed as the lesion diffusion in the relevant distinguishing feature, It is expressed as the dynamic adjustment factor of the locking coefficient, and V represents the second correction constant;

[0030] Among them, the dynamic adjustment factor of the locking coefficient Obtained by the following formula:

[0031]

[0032] In the formula, e represents the base of natural logarithm, and its approximate value is 2.71828;

[0033] By presetting an assessment threshold K and comparing it with the injury assessment index Spzs, a comprehensive assessment is made of the degree of craniocerebral injury of the current patient. If the injury assessment index Spzs exceeds the assessment threshold K, it is determined that the craniocerebral injury of the current patient is in a dangerous state. At this time, the doctor will be notified immediately to perform artificial intervention on the current patient, transfer the patient to the intensive care unit, and monitor the patient's vital signs in real time. The vital signs monitoring includes blood pressure, heart rate, respiratory rate and intracranial pressure. If the injury assessment index Spzs does not exceed the assessment threshold K, it is determined that the craniocerebral injury of the current patient is not in a dangerous state. At this time, the patient's vital signs will be monitored regularly, and a CT and MRI review plan will be adopted within 24 hours to observe changes in lesions.

[0034] The present invention provides a medical image-based craniocerebral injury degree diagnosis auxiliary system, which has the following beneficial effects:

[0035] (1) The system can efficiently and accurately identify the lesion area in the cranial medical imaging group through the image data acquisition module and the image preprocessing technology. Based on the lesion area, the multimodal feature set in the white matter is extracted. The acquisition of this multimodal feature not only improves the precise positioning of the lesion area, but also lays a solid foundation for the quantitative analysis of complex pathological mechanisms. The initial judgment module can accurately judge whether the patient has the risk of interstitial brain edema by analyzing the degree of interstitial fluid accumulation in the white matter and quantitatively calculating the permeability abnormality coefficient Syxs, providing the ability of early risk screening, and can timely identify potential problems before the condition worsens, prompting clinicians to intervene quickly, thereby significantly improving the patient's prognosis. When there is a risk of interstitial brain edema, the type diagnosis module further distinguishes the lesion type (such as brain contusion or other injuries) by extracting relevant distinguishing features from the multimodal feature set, and the calculation of the locking coefficient Sdxs provides a scientific quantitative basis for the diagnosis of the lesion type based on the specific characteristics of the lesion. This precise type diagnosis method can reduce diagnostic errors and provide reliable support for clinical decision-making. The craniocerebral injury assessment module uses convolutional neural network technology to build a diagnostic model, realizes intelligent analysis of the degree of injury, and fits the output injury assessment index Spzs to provide clinicians with clear and quantitative diagnostic results. The visual output further enhances the intuitiveness and readability of the results, helping doctors to quickly understand the condition.

[0036] (2) The initial judgment module is based on the permeability abnormality coefficient and can detect the risk of interstitial cerebral edema at an early stage, providing a guarantee for timely intervention. At the same time, further analysis is triggered according to the initial judgment results, reducing unnecessary deep diagnosis links and optimizing resource utilization. The type diagnosis module refines the injury classification by locking the coefficient, providing a basis for formulating personalized treatment plans. According to the value of the injury assessment index, it can be judged whether the patient's injury level exceeds the assessment threshold, and then corresponding intervention measures (such as enhanced monitoring or surgical treatment) are taken. The convolutional neural network is used to construct an injury assessment model, which enhances the system's ability to fit complex features and improves diagnostic accuracy. The linkage design of the initial judgment module and the type diagnosis module ensures that the diagnosis process is flexible and dynamic to meet the personalized needs of different patients.

[0037] (3) The boundary regularity and morphological characteristics of the lesion area are quantified by boundary clarity, which clearly reflects the complexity of the lesion boundary. The grayscale difference and texture complexity in the lesion area are described based on texture contrast. The image feature analysis unit combines boundary clarity and texture contrast to comprehensively analyze the lesion characteristics from two dimensions: geometric morphology and texture characteristics. By accurately extracting and quantifying features, the limitations of traditional reliance on doctors' subjective judgment are eliminated, and the objectivity and accuracy of injury type diagnosis are further improved. Early warning is provided for lesion areas with uniform texture and low contrast (indicating interstitial cerebral edema), guiding doctors to take dehydration, decompression and other treatments. The output boundary clarity and texture contrast provide high-quality input for the subsequent injury assessment module. Through the construction of the type diagnosis module, the system can realize efficient and accurate analysis of the type of craniocerebral injury based on the boundary characteristics and texture characteristics of the image, providing strong technical support for personalized treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The figure is a flowchart diagram of the block diagram of the auxiliary system for diagnosing the degree of craniocerebral injury based on medical images of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Example 1

[0041] See also Figure 1 , the present invention provides a craniocerebral injury degree diagnosis auxiliary system based on medical images, including an image data acquisition module, a preliminary judgment module, a type diagnosis module and a craniocerebral injury assessment module;

[0042] The image data acquisition module is used to acquire a group of cranial and brain medical images of the patient, and to identify the lesion area in the group of cranial and brain medical images through image preprocessing, and to obtain a multimodal feature set in the white matter of the patient's lesion area based on the lesion area;

[0043] The preliminary judgment module is used to preliminarily analyze the accumulation degree of interstitial fluid in the white matter of the brain according to the multimodal feature set to obtain the abnormal permeability coefficient Syxs, and diagnose whether there is a risk of interstitial cerebral edema in the current patient's brain based on the abnormal permeability coefficient Syxs, and if so, issue a recognition instruction;

[0044] The type diagnosis module is used to extract relevant distinguishing features in the multimodal feature set after receiving the recognition instruction, and obtain the locking coefficient Sdxs based on the relevant distinguishing features to diagnose the type of injury in the current patient's brain again;

[0045] The craniocerebral injury assessment module is used to construct a craniocerebral injury degree diagnosis model using convolutional neural network technology, and input the locking coefficient Sdxs and the permeability abnormality coefficient Syxs into the craniocerebral injury degree diagnosis model. After dimensionless processing, the injury assessment index Spzs is fitted and output, and displayed in a visual form.

[0046] During the operation of this system, the system realizes the hierarchical diagnosis process from data acquisition to final evaluation through the division of labor and cooperation among the image data acquisition module, the preliminary judgment module, the type diagnosis module and the craniocerebral injury assessment module. By using the multimodal feature set in the lesion area, the in-depth analysis of the patient's white matter area is realized, and the risk of cerebral edema and the type of injury are accurately identified. The degree of interstitial fluid accumulation is preliminarily analyzed through the preliminary judgment module, and the type diagnosis module is further triggered when the risk exists, ensuring that the diagnosis process is dynamic and efficient, and reducing omissions. The convolutional neural network technology is used to automatically fit the injury assessment index to reduce the workload of doctors relying on manual interpretation in traditional diagnosis. The diagnosis results are presented in a visual form, which is convenient for doctors to quickly understand the degree of craniocerebral injury in patients and further improve the efficiency of diagnosis. From the preliminary analysis of interstitial fluid accumulation to the in-depth diagnosis of injury type, the whole process of craniocerebral injury from lesion characteristics to type classification is covered. And based on the provision of injury degree assessment Spzs, the risk and severity of craniocerebral injury of patients are comprehensively assessed.

[0047] Example 2

[0048] Please refer to Figure 1 ,Specifically: the image data acquisition module includes an image acquisition unit, an image preprocessing unit and a feature extraction unit;

[0049] The image acquisition unit is used to perform a cranial brain scan on the patient using a computer tomography scanner and a magnetic resonance imaging device in advance to acquire a cranial brain medical image set of the patient;

[0050] The image preprocessing unit is used to import the cranial brain medical image group into a DICOM file to check the integrity and modality completeness of the data, and convert the DICOM file into the NIfTI format, which is an image format commonly used in medical image analysis. At the same time, the first cranial brain medical image obtained by a computer tomography scanner is subjected to noise removal by a bilateral filtering method, and the second cranial brain medical image obtained by a magnetic resonance imaging device is subjected to noise removal by a non-local mean filtering method. The cranial brain medical image group is then subjected to grayscale value normalization processing, and the grayscale value range is adjusted so that the image input has a consistent grayscale distribution, which is convenient for feature extraction; and the normalized cranial brain medical image group is automatically segmented to identify and extract the lesion area in the cranial brain medical image group; wherein the cranial brain medical image group includes the first cranial brain medical image obtained by a computer tomography scanner and the second cranial brain medical image obtained by a magnetic resonance imaging device;

[0051] The feature extraction unit is used to process the cranial medical image group in advance to generate a gray level co-occurrence matrix, and extract a multimodal feature set in the white matter of the patient's lesion area according to the lesion area, wherein the multimodal feature set includes relevant fluid penetration features and relevant distinguishing features in the white matter of the brain, and the relevant fluid penetration features include fluid accumulation volume and the total volume of the lesion area ; The relevant distinguishing features include the density of the corresponding pixels in the lesion area , the total area of ​​the lesion , the total length of the lesion area boundary , the gray level Hdj of the corresponding pixel in the gray level co-occurrence matrix (GLCM), and the average T1 signal intensity of the lesion area , the average T1 signal intensity of normal brain tissue and lesion diffusion .

[0052] In this embodiment, a computed tomography scanner (CT) and a magnetic resonance imaging device (MRI) are used simultaneously to perform a cranial brain scan on the patient, combining the advantages of the two images (such as CT is good at density analysis, and MRI is good at soft tissue contrast), ensuring the comprehensiveness and diversity of the image data, and applying bilateral filtering to CT images to smooth noise while retaining boundary characteristics. Non-local mean filtering is applied to MRI images to effectively reduce random noise and retain key brain tissue details, adjust the gray value range of the image group, so that all image inputs have a consistent gray distribution, which is helpful for subsequent feature extraction and model analysis, and avoids deviations caused by grayscale differences. Automatic segmentation technology is used to accurately identify and extract the lesion area from the normalized image, reducing the error and time of manual intervention.

[0053] Automatic extraction of lesion areas lays a solid foundation for subsequent feature analysis and evaluation. The system not only extracts the fluid infiltration characteristics of the lesion area, but also extracts distinguishing features, extracts the fluid accumulation volume and the total volume of the lesion area, and provides an accurate quantitative basis for judging the degree of interstitial fluid accumulation in the white matter area. Providing basic data on interstitial fluid infiltration-related features can identify the risk of interstitial brain edema early and guide clinical intervention.

[0054] Combining the high-density contrast characteristics of CT images with the soft tissue sensitivity of MRI images, multi-dimensional features are extracted to avoid the limitations of single-modality images.

[0055] Through the design and implementation of the image data acquisition module, the system can efficiently and high-quality complete the process from image acquisition to lesion feature extraction, providing important support for the intelligent diagnosis of craniocerebral injury, thereby improving clinical diagnosis and treatment efficiency and patient prognosis.

[0056] Example 3

[0057] Please refer to Figure 1 ,Specifically: the preliminary judgment module includes a non-specific image performance analysis unit and a preliminary judgment unit;

[0058] The non-specific image performance analysis unit is used to analyze the abnormal accumulation of interstitial fluid infiltrating into the white matter area of ​​the brain according to the relevant fluid infiltration characteristics to obtain the water content ratio Shb. The water content ratio Shb is obtained by the following formula:

[0059]

[0060] In the formula, Expressed as the liquid accumulation volume, It is expressed as the total volume of the lesion area;

[0061] Interstitial brain edema and its indistinguishable types of injury usually present with an increase in water content;

[0062] The above liquid accumulation volume It refers to the volume of fluid accumulated in the lesion area due to edema or abnormal osmotic pressure, which is one of the important characteristics of interstitial cerebral edema.

[0063] The total volume of the lesion area Refers to the three-dimensional spatial volume of the lesion area, including fluid accumulation, diseased tissue, etc. It is an important indicator to characterize the spread of the lesion;

[0064] Combined with the cranial medical images of the lesion area, the average density of the lesion area is analyzed and obtained. , the average density The specific method is as follows:

[0065]

[0066] In the formula, It is represented as the density of the i-th pixel in the lesion area, N is represented as the total number of pixels in the lesion area, i=1, 2, 3, ..., N.

[0067] The density of the corresponding pixels in the above lesion area Refers to the grayscale value of each pixel in the lesion area, reflecting the physical properties of the tissue;

[0068] By dividing the water content ratio Shb and the average density The permeability anomaly coefficient Syxs is calculated and obtained after dimensionless processing. The permeability anomaly coefficient Syxs is obtained by the following formula:

[0069] In the formula, and are weight values, represents the first correction constant, where and The specific value is set by the user according to the situation, ranging from: 0< ≤1,0< ≤1.

[0070] The preliminary judgment unit is used to extract a group of normal human cranial medical images from the hospital, and based on the group of normal human cranial medical images, set a standard threshold H, and compare and analyze it with the abnormal permeability coefficient Syxs to diagnose whether there is a risk of interstitial brain edema in the current patient's cranial brain. The specific diagnosis content is as follows:

[0071] If the abnormal permeability coefficient Syxs exceeds the standard threshold value H, it is diagnosed that the patient has a risk of interstitial cerebral edema in the brain, and an identification instruction is sent to the type diagnosis module;

[0072] If the abnormal permeability coefficient Syxs does not exceed the standard threshold value H, it is diagnosed that there is no risk of interstitial cerebral edema in the current patient's brain. At this time, no identification instruction will be issued to the type diagnosis module, and the analysis process of the risk of interstitial cerebral edema in the patient's brain will be ended.

[0073] In this embodiment, the water content ratio is calculated by the ratio of the volume of fluid accumulation to the total volume of the lesion, which can directly quantify the degree of fluid accumulation in the white matter area of ​​the brain, and provide an accurate indicator for judging the abnormal permeability of interstitial fluid. This indicator fully reflects the characteristics of interstitial brain edema, that is, the water content is significantly increased. Combined with the CT image of the lesion area, the average density of the lesion area is extracted, and the abnormality of fluid accumulation is further verified by the reduction of the density value. Density analysis is combined with water content analysis to reflect the damage characteristics of the white matter area in more aspects. The permeability abnormality coefficient Syxs can more scientifically evaluate the abnormal permeability of liquid in the white matter area by quantifying a variety of key features uniformly, and provide an intuitive basis for the diagnosis of interstitial brain edema. Based on the standard threshold set by the normal brain medical imaging group, a clear reference benchmark is provided for whether there is a risk of interstitial brain edema, which enhances the scientificity and objectivity of the diagnostic results. By comparing with normal data, it is ensured that the diagnostic process has a reliable control basis. The system automatically compares with the standard threshold, quickly draws a diagnostic conclusion, and improves diagnostic efficiency. If the diagnosis shows a risk of interstitial cerebral edema, the system will automatically trigger the type diagnosis module for further analysis to achieve intelligent linkage between modules. If the risk does not exceed the threshold, the diagnosis process is automatically terminated to optimize the diagnosis process. The data is dimensionlessly processed to eliminate the influence of different feature dimensions, making the diagnosis results more standardized. The standard threshold is set based on normal human data to ensure that the diagnosis has a reliable reference. The initial judgment module can quickly screen for early risks of interstitial cerebral edema and provide clinicians with a basis for timely intervention. For patients at risk, the system can quickly trigger further type diagnosis to help with accurate classification.

[0074] Example 4

[0075] Please refer to Figure 1 ,Specifically: the type diagnosis module includes an image feature analysis unit and a type recognition unit;

[0076] The image feature analysis unit is used to receive the recognition instruction, analyze the boundary characteristics of the image in the cranial medical image group according to the relevant distinguishing features in the multimodal feature set, so as to calculate and obtain the boundary clarity Bqcd and the texture contrast Wdbd respectively. The boundary clarity Bqcd and the texture contrast Wdbd are respectively obtained by the following formulas: In the formula, It is expressed as the total area of ​​the lesion region; It is expressed as the total length of the boundary of the lesion area; It is expressed as pi, which regulates the deviation of boundary regularity from the ideal circle;

[0077] The total area of ​​the above lesion areas It refers to the area of ​​the lesion region on a two-dimensional plane slice and is an important parameter of the morphological characteristics of the lesion.

[0078] The total length of the lesion area boundary It refers to the total length of all boundary pixels in the lesion area on a two-dimensional plane and is an important feature to characterize the regularity of the boundary.

[0079] In the formula, It is expressed as the joint probability of gray levels between adjacent pixels in the gray-level co-occurrence matrix (GLCM); Represented as the gray level of the j-th pixel, It is represented as the gray level of the j+1th pixel, and j represents each pixel in the gray level co-occurrence matrix; Represents the square of the difference between gray levels, emphasizing that pixels with larger grayscale differences contribute more to the contrast; grayscale refers to the gray value level of image pixels.

[0080] Texture contrast is an indicator that measures the degree of grayscale change in image texture. It describes the difference between gray levels in an image. The larger the value, the more obvious the grayscale difference and the higher the texture complexity. The smaller the value, the lower the grayscale difference and the more uniform the texture. It is used to evaluate the texture characteristics of lesions, such as distinguishing interstitial cerebral edema (uniform texture, low contrast) and cerebral contusion (complex texture, high contrast).

[0081] The gray-level co-occurrence matrix is ​​a two-dimensional matrix that describes the grayscale distribution of image pixels. The acquisition steps are as follows: Image preprocessing: First, convert the image into a grayscale image and bin the grayscale values ​​(for example, divide the grayscale into 16 levels to simplify the calculation). Select the direction and distance of adjacent pixels (such as horizontal, vertical or diagonal direction). Common values ​​for the direction are: 0° for the horizontal direction, 45° for the diagonal direction, and 90° for the vertical direction. Count the number of times the grayscale of adjacent pixel pairs appears in the selected direction and distance, and then normalize it to obtain the joint probability. ;

[0082] In this embodiment, the regularity and clarity of the boundary of the lesion area are accurately described by calculating the boundary clarity Bqcd, and the edge detection algorithm is used to count the boundary pixel length to ensure the high accuracy of the boundary clarity calculation. By calculating the texture contrast Wdbd, the texture complexity and grayscale difference of the lesion area are deeply evaluated, the texture characteristics are calculated based on the grayscale co-occurrence matrix, and the significant characteristics of the lesion texture are reflected by the joint probability of the pixel grayscale distribution. The boundary clarity and texture contrast are used together to more comprehensively distinguish the lesion type and reduce the limitations of single feature analysis. Using boundary clarity and texture contrast as the core indicators for determining the type of injury can help doctors determine the nature and extension range of the injury type, so as to formulate targeted treatment plans.

[0083] Example 5

[0084] Please refer to Figure 1 Specifically, the type recognition unit is used to obtain the locking coefficient Sdxs by associating the boundary definition Bqcd and the texture contrast Wdbd obtained in the image feature analysis unit and performing dimensionless processing. The locking coefficient Sdxs is obtained by the following formula: In the formula, Expressed as signal intensity ratio, , and are weight values, among which, , and The specific value is set by the user according to the situation, ranging from: 0< ≤1,0< ≤1,0< ≤1; where the signal strength ratio Obtained by the following formula: In the formula, It is expressed as the average T1 signal intensity of the lesion area. It is expressed as the average T1 signal intensity of normal brain tissue.

[0085] The average T1 signal intensity of the above lesion area It refers to the mean signal intensity of all pixels in the lesion area in the T1-weighted image, reflecting the recovery process of the longitudinal magnetization of hydrogen atoms (T1 relaxation time); the average T1 signal intensity of normal brain tissue It refers to the average T1 signal intensity of normal brain tissue (such as adjacent white matter or gray matter areas), which serves as a reference value for comparing the signal intensity of the lesion area.

[0086] The division threshold E is preset, and the locking coefficient Sdxs is compared with the division threshold E to diagnose the type of injury in the current patient's craniocerebral region again. The specific contents are as follows:

[0087] If the locking coefficient Sdxs exceeds the division threshold E, it will be diagnosed that the current patient's intracranial injury type is not interstitial cerebral edema;

[0088] If the locking coefficient Sdxs does not exceed the division threshold E, the current patient's intracranial injury type will be diagnosed as interstitial cerebral edema.

[0089] In this embodiment, the calculation of the locking coefficient Sdxs combines the boundary clarity, texture contrast and signal intensity ratio, quantifies the geometric characteristics, texture characteristics and signal characteristics of the lesion area, and comprehensively characterizes the characteristics of the lesion from multiple dimensions. Interstitial cerebral edema: fuzzy boundaries, uniform texture, and signal intensity ratio close to normal tissue. Non-interstitial cerebral edema (such as cerebral contusion): clear boundaries, complex textures, and obvious differences in signal intensity. Through the dynamic calculation of the locking coefficient, the classification results are ensured to be more accurate, and misjudgments and missed judgments are reduced. The division threshold is pre-set to convert the complex multi-feature analysis results into a simple binary classification problem (whether it is interstitial cerebral edema), ensuring that the diagnosis process has a unified standard and high repeatability. Signal intensity ratio directly quantifies the signal difference of the lesion through image characteristics. Boundary clarity, texture contrast and signal intensity ratio are automatically extracted by the algorithm, and the classification results are automatically output in combination with the locking coefficient formula, which reduces the workload of manual intervention by doctors as much as possible. The system quickly outputs the diagnosis results, significantly improves the diagnosis efficiency, and is particularly suitable for clinical rapid decision-making scenarios. Through the design of the type recognition unit, the injury type classification based on multidimensional features is realized, which provides strong support for the intelligent diagnosis of craniocerebral injury, effectively assists doctors in formulating personalized treatment plans, and optimizes the patient's clinical prognosis.

[0090] Example 6

[0091] Please refer to Figure 1 ,Specifically: the craniocerebral injury assessment module includes a modeling unit and an evaluation output unit;

[0092] The modeling unit is used to construct an initial model using convolutional neural network technology, and use a multimodal feature set as the input of the initial model for training and testing, and use the trained initial model as a state recognition model to obtain feature information in the state recognition model respectively, and use the acquired feature information to train and test the state recognition model, and combined with the current patient's craniocerebral injury type, use the trained state recognition model as a craniocerebral injury degree diagnosis model.

[0093] It should be noted that the core of the entire process of the above-mentioned modeling unit is to build an initial model through convolutional neural network technology (CNN), and then perform state recognition and injury degree diagnosis based on the features extracted by the model. This process includes multimodal feature input, state recognition and training, and is ultimately applied to craniocerebral injury diagnosis. The structure of the initial model and craniocerebral injury degree diagnosis model is specifically analyzed below:

[0094] 1. The initial model is the basis of the entire system. Its structure is mainly based on the convolutional neural network (CNN) and takes a multimodal feature set as input. The structure of the initial model can be divided into the following key layers: input layer, convolution layer, fully connected layer, and output layer; Among them, input layer: The input data is a multimodal feature set. The size of image-related data is usually adjusted to a consistent size (for example, 256×256 pixels, 3-channel RGB images), while other features (non-image-related data) are encoded in numerical form.

[0095] Convolutional layers: Multiple convolutional layers are used to extract spatial features from a multimodal feature set. These convolutional layers automatically learn local patterns, textures, and other features in the image. In medical images, convolutional layers can help identify important structures such as brain tissue, blood vessels, and lesions. Each convolutional layer is usually followed by a pooling layer (such as a max pooling layer) to reduce the spatial size, dimensionality, and retain the most important features.

[0096] Fully connected layer: After extracting deep features, multiple fully connected layers are connected to integrate the features and output a representation for state recognition. These layers perform nonlinear mapping on the image to better handle complex medical features. Output layer: The output layer performs numerical prediction and evaluation;

[0097] The goal of the initial model is to learn to extract useful features from the input data by training the CNN model for subsequent state recognition and damage diagnosis.

[0098] 2. For the state recognition model, its task is to classify or predict the patient's state through the features extracted from the initial CNN model. The state recognition model retrains the initial model to obtain more specific state features (such as injury type, severity of the disease, etc.). The structure of the state recognition model includes: feature extraction layer, state classification layer and output layer. Among them, the feature extraction layer: extracts the output features of the fully connected layer from the initial model. These features are the activation values ​​of the convolutional layer and contain the deep semantic information of the image. At the same time, the feature extraction layer will further perform dimensionality reduction (such as using principal component analysis PCA) to ensure that the extracted features can better represent the state changes.

[0099] State classification layer: One or more fully connected layers are used to input the extracted features and perform nonlinear transformations to determine the patient's current state, such as the severity of the disease.

[0100] The output layer will again predict the numerical result of the damage (such as the extent of the damage).

[0101] 3. For the craniocerebral injury degree diagnosis model, the main goal of the craniocerebral injury degree diagnosis model is to evaluate and predict the specific degree of the patient's craniocerebral injury based on the trained state recognition model. This part is the final output of the system and determines the diagnosis result. The structure of the craniocerebral injury degree diagnosis model includes an input layer and an output layer; among them, the input layer: the input layer receives the features obtained from the state recognition model; the output layer: the output layer will give a diagnosis result, including the type of injury, the severity of the injury, and treatment recommendations.

[0102] In summary, the initial model uses convolutional neural networks (CNNs) to extract features of relevant medical images and other modal data to establish a basic model for state recognition. The state recognition model uses the features extracted by the initial model to further analyze and classify the patient's health status and identify the type of injury; the craniocerebral injury degree diagnosis model evaluates and diagnoses the patient's craniocerebral injury based on the results of the state recognition model, and gives the final injury degree and treatment recommendations.

[0103] The evaluation output unit is used to fit and output the injury assessment index Spzs according to the craniocerebral injury degree diagnosis model. The injury assessment index Spzs is obtained by the following formula: Where Syxs is the permeability anomaly coefficient, Expressed as the lesion diffusion in the relevant distinguishing feature, It is expressed as the dynamic adjustment factor of the locking coefficient, V represents the second correction constant; the above-mentioned lesion diffusion Obtained by the following formula:

[0104]

[0105] In the formula, represents the total area of ​​the lesion region, Total white matter area It refers to the area of ​​cerebral white matter on a two-dimensional plane slice, and is an important parameter for evaluating the ratio of the lesion area to normal brain tissue. Obtained by the following formula:

[0106]

[0107] Where, e represents the base of the natural logarithm, and its approximate value is 2.71828; the dynamic adjustment factor of the locking coefficient is The purpose is to: directly use the locking coefficient to weight the damage assessment parameters may lead to discontinuity, while the smooth change of the exponential function ensures the gradual change of the result and further avoids abrupt weight adjustment.

[0108] By presetting an assessment threshold K, it is compared and analyzed with the injury assessment index Spzs to comprehensively assess the degree of craniocerebral injury of the current patient. If the injury assessment index Spzs exceeds the assessment threshold K, it is judged that the craniocerebral injury of the current patient is in a dangerous state. At this time, the doctor will be notified immediately to perform artificial intervention on the current patient, transfer the patient to the intensive care unit (ICU), and monitor the patient's vital signs in real time. The vital signs monitoring includes blood pressure, heart rate, respiratory rate and intracranial pressure (ICP); if the injury assessment index Spzs does not exceed the assessment threshold K, it is judged that the current patient's craniocerebral injury is not in a dangerous state, indicating that the patient's craniocerebral injury is relatively mild, which may be localized edema or slight infiltration abnormality, and the lesion has not yet reached a life-threatening level. At this time, the patient's vital signs will be monitored regularly, and a CT / MRI review plan will be adopted within 24 hours to observe changes in lesions and avoid worsening of the disease.

[0109] In this embodiment, a state recognition model is constructed and trained, and a multimodal feature set is used as input, including the permeability anomaly coefficient Syxs, the lesion diffusion and dynamic adjustment factor of the locking coefficient , further realizing multi-dimensional modeling of craniocerebral injury degree, the deep feature extraction capability of convolutional neural network ensures the high adaptability of the model to complex data, can fully explore the potential correlation between injury features, and improve the accuracy of model evaluation. The injury assessment index integrates multiple key features and outputs a quantitative injury assessment index Spzs, which provides an intuitive and objective indicator for the determination of injury degree.

[0110] By comparing the injury assessment index Spzs with the preset threshold K, it is intelligently determined whether the patient is in a dangerous state. This automatic judgment mechanism reduces the doctor's judgment burden and reduces the possibility of misdiagnosis or missed diagnosis. During the modeling process, the convolutional neural network improves the model's recognition ability and diagnostic accuracy for complex features through training and testing. The trained craniocerebral injury degree diagnosis model can reliably distinguish different injury characteristics, further reducing the errors that are prone to occur in traditional manual analysis.

[0111] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A medical image-based diagnostic aid system for craniocerebral injury, characterized by: It includes image data acquisition module, preliminary judgment module, type diagnosis module and craniocerebral injury assessment module; The image data acquisition module is used to acquire a group of cranial and brain medical images of the patient, and to identify the lesion area in the group of cranial and brain medical images through image preprocessing, and to obtain a multimodal feature set in the white matter of the patient's lesion area based on the lesion area; The preliminary judgment module is used to preliminarily analyze the accumulation degree of interstitial fluid in the white matter of the brain according to the multimodal feature set to obtain the abnormal permeability coefficient Syxs, and diagnose whether there is a risk of interstitial cerebral edema in the current patient's brain based on the abnormal permeability coefficient Syxs, and if so, issue a recognition instruction; The type diagnosis module is used to extract relevant distinguishing features in the multimodal feature set after receiving the recognition instruction, and obtain the locking coefficient Sdxs based on the relevant distinguishing features to diagnose the type of injury in the current patient's brain again; The type diagnosis module includes an image feature analysis unit and a type recognition unit; The image feature analysis unit is used to receive the recognition instruction, analyze the boundary characteristics of the image in the cranial medical image group according to the relevant distinguishing features in the multimodal feature set, so as to calculate and obtain the boundary clarity Bqcd and the texture contrast Wdbd respectively. The boundary clarity Bqcd and the texture contrast Wdbd are respectively obtained by the following formulas: In the formula, It is expressed as the total area of ​​the lesion region; It is expressed as the total length of the boundary of the lesion area; It is expressed as pi; In the formula, It is expressed as the joint probability of gray levels between adjacent pixels in the gray-level co-occurrence matrix; Represented as the gray level of the j-th pixel, It is represented as the gray level of the j+1th pixel, and j represents each pixel in the gray level co-occurrence matrix; The type recognition unit is used to obtain a locking coefficient Sdxs by correlating the boundary definition Bqcd and the texture contrast Wdbd obtained in the image feature analysis unit and performing dimensionless processing. The locking coefficient Sdxs is obtained by the following formula: In the formula, Expressed as signal intensity ratio, , and are weight values, among which, , and The specific value is set by the user according to the situation; Among them, the signal strength ratio Obtained by the following formula: In the formula, It is expressed as the average T1 signal intensity of the lesion area. It is expressed as the average T1 signal intensity of normal brain tissue; The division threshold E is preset, and the locking coefficient Sdxs is compared with the division threshold E to diagnose the type of injury in the current patient's craniocerebral region again. The specific contents are as follows: If the locking coefficient Sdxs exceeds the division threshold E, it will be diagnosed that the current patient's intracranial injury type is not interstitial cerebral edema; If the locking coefficient Sdxs does not exceed the division threshold E, the current patient's intracranial injury type is diagnosed as interstitial cerebral edema; The craniocerebral injury assessment module is used to construct a craniocerebral injury degree diagnosis model using convolutional neural network technology, and input the locking coefficient Sdxs and the permeability abnormality coefficient Syxs into the craniocerebral injury degree diagnosis model. After dimensionless processing, the injury assessment index Spzs is fitted and output, and displayed in a visual form.

2. The system for diagnosing craniocerebral injury based on medical images according to claim 1 is characterized in that: The image data acquisition module includes an image acquisition unit, an image preprocessing unit and a feature extraction unit; The image acquisition unit is used to perform a cranial brain scan on the patient using a computer tomography scanner and a magnetic resonance imaging device in advance to acquire a cranial brain medical image set of the patient; The image preprocessing unit is used to import the cranial brain medical image group into a DICOM file, and convert the DICOM file into a NIfTI format, and at the same time remove noise from the No. 1 cranial brain medical image obtained by a computer tomography scanner through a bilateral filtering method, and remove noise from the No. 2 cranial brain medical image obtained by a magnetic resonance imaging device through a non-local mean filtering method, and then normalize the grayscale value of the cranial brain medical image group, adjust the grayscale value range, and automatically segment the normalized cranial brain medical image group to identify and extract the lesion area in the cranial brain medical image group; The cranial brain medical image group includes a first cranial brain medical image acquired by a computer tomography scanner and a second cranial brain medical image acquired by a magnetic resonance imaging device; The feature extraction unit is used to process the cranial medical image group in advance to generate a gray level co-occurrence matrix, and extract a multimodal feature set in the white matter of the patient's lesion area according to the lesion area, wherein the multimodal feature set includes relevant fluid penetration features and relevant distinguishing features in the white matter of the brain, and the relevant fluid penetration features include fluid accumulation volume and the total volume of the lesion area ; The relevant distinguishing features include the density of the corresponding pixels in the lesion area , the total area of ​​the lesion , the total length of the lesion area boundary , the gray level Hdj of the corresponding pixel in the gray level co-occurrence matrix, and the average T1 signal intensity of the lesion area , the average T1 signal intensity of normal brain tissue and lesion diffusion .

3. The system for diagnosing craniocerebral injury based on medical images according to claim 2 is characterized in that: The preliminary judgment module includes a non-specific image performance analysis unit and a preliminary judgment unit; The non-specific image performance analysis unit is used to analyze the abnormal accumulation of interstitial fluid infiltrating into the white matter area of ​​the brain according to the relevant fluid infiltration characteristics to obtain the water content ratio Shb. The water content ratio Shb is obtained by the following formula: In the formula, Expressed as the liquid accumulation volume, It is expressed as the total volume of the lesion area; Combined with the cranial medical images of the lesion area, the average density of the lesion area is analyzed and obtained. , the average density The specific method is as follows: In the formula, It is represented as the density of the i-th pixel in the lesion area, N is represented as the total number of pixels in the lesion area, i=1, 2, 3, ..., N.

4. The system for diagnosing craniocerebral injury based on medical images according to claim 3 is characterized in that: By dividing the water content ratio Shb and the average density The permeability anomaly coefficient Syxs is calculated and obtained after dimensionless processing. The permeability anomaly coefficient Syxs is obtained by the following formula: In the formula, and are weight values, represents the first correction constant, where and The specific value is set by the user according to the situation.

5. The system for diagnosing craniocerebral injury based on medical images according to claim 4 is characterized in that: The preliminary judgment unit is used to extract a group of normal human cranial medical images from the hospital, and based on the group of normal human cranial medical images, set a standard threshold H, and compare and analyze it with the abnormal permeability coefficient Syxs to diagnose whether there is a risk of interstitial brain edema in the current patient's cranial brain. The specific diagnosis content is as follows: If the abnormal permeability coefficient Syxs exceeds the standard threshold value H, it is diagnosed that the patient has a risk of interstitial cerebral edema in the brain, and an identification instruction is sent to the type diagnosis module; If the abnormal permeability coefficient Syxs does not exceed the standard threshold value H, it is diagnosed that there is no risk of interstitial cerebral edema in the current patient's brain. At this time, no identification instruction will be issued to the type diagnosis module, and the analysis process of the risk of interstitial cerebral edema in the patient's brain will be ended.

6. The system for diagnosing craniocerebral injury based on medical images according to claim 1 is characterized in that: The craniocerebral injury assessment module includes a modeling unit and an evaluation output unit; The modeling unit is used to construct an initial model using convolutional neural network technology, and use a multimodal feature set as the input of the initial model for training and testing, and use the trained initial model as a state recognition model to obtain feature information in the state recognition model respectively, and use the acquired feature information to train and test the state recognition model, and combined with the current patient's craniocerebral injury type, use the trained state recognition model as a craniocerebral injury degree diagnosis model.

7. The system for diagnosing craniocerebral injury based on medical images according to claim 6 is characterized in that: The evaluation output unit is used to fit and output the injury assessment index Spzs according to the craniocerebral injury degree diagnosis model. The injury assessment index Spzs is obtained by the following formula: Where Syxs is the permeability anomaly coefficient, Expressed as the lesion diffusion in the relevant distinguishing feature, It is expressed as the dynamic adjustment factor of the locking coefficient, and V represents the second correction constant; Among them, the dynamic adjustment factor of the locking coefficient Obtained by the following formula: In the formula, e represents the base of natural logarithm; By presetting an assessment threshold K and comparing it with the injury assessment index Spzs, a comprehensive assessment is made of the degree of craniocerebral injury of the current patient. If the injury assessment index Spzs exceeds the assessment threshold K, it is determined that the craniocerebral injury of the current patient is in a dangerous state. At this time, the doctor will be notified immediately to perform artificial intervention on the current patient, transfer the patient to the intensive care unit, and monitor the patient's vital signs in real time. The vital signs monitoring includes blood pressure, heart rate, respiratory rate and intracranial pressure. If the injury assessment index Spzs does not exceed the assessment threshold K, it is determined that the craniocerebral injury of the current patient is not in a dangerous state. At this time, the patient's vital signs will be monitored regularly, and a CT and MRI review plan will be adopted within 24 hours to observe changes in lesions.

Images