CST damage level detection method and system based on multi-modal data fusion

Abstract

The invention provides a CST damage level detection method and system based on multi-modal data fusion, and relates to the technical field of medical system control, and the method comprises the steps: obtaining a training sample set which comprises a plurality of training samples, carrying out the iterative training of a neural network model according to a feature matrix corresponding to each training sample and a CST damage level, and obtaining a CST damage level detection result; obtaining a trained neural network model; and determining the CST injury level of the to-be-detected individual through the trained neural network model according to the transcranial magnetic stimulation data, the head image data and the physiological data of the to-be-detected individual. According to the method provided by the invention, the CST injury level can be rapidly and accurately detected, the corresponding rehabilitation scheme and the operation scheme can be determined according to the injury level, accurate grading of the CST injury and dynamic matching of rehabilitation and operation strategies are realized, and the accuracy of a detection result can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data processing, and in particular to a CST injury level detection method and system based on multi-modal data fusion. BACKGROUND

[0002] In the field of neurorehabilitation, central nervous system injuries such as cerebral hemorrhage and cerebral infarction often lead to damage to the corticospinal tract (CST), which in turn causes motor dysfunction in patients and seriously affects their quality of life. How to accurately assess the degree of CST injury and develop targeted rehabilitation strategies has always been a problem to be solved, which not only relates to the rehabilitation effect of patients, but also has important significance for reducing medical costs.

[0003] Currently, the evaluation and rehabilitation of CST injury are usually based on imaging evaluation methods, such as detecting parameters such as anisotropy fraction (FA) to judge the structural integrity of CST through diffusion tensor imaging (DTI), obtaining the detection results, and developing corresponding rehabilitation programs according to the detection results.

[0004] However, the existing technology has obvious limitations, i.e., it cannot achieve accurate grading of CST injury and dynamic matching of surgical rehabilitation strategies. Moreover, the evaluation method based on a single imaging is difficult to comprehensively and accurately reflect the injury state of CST, resulting in a large error in the detection results and poor accuracy. SUMMARY

[0005] The embodiments of the present application provide a CST injury level detection method and system based on multi-modal data fusion, which can quickly and accurately detect the CST injury level and determine the corresponding rehabilitation program according to the injury level, achieve accurate grading of CST injury and dynamic matching of rehabilitation strategies, and improve the accuracy of the detection results.

[0006] To achieve the above-mentioned purpose, the embodiments of the present application adopt the following technical solutions: In a first aspect, a CST injury level detection method based on multi-modal data fusion is provided, the method comprising: obtaining a training sample set, the training sample set comprising a plurality of training samples, each training sample comprising transcranial magnetic stimulation data, head image data, physiological data and a CST injury level of a stroke individual, the transcranial magnetic stimulation data comprising a motor evoked potential amplitude ratio and a motor evoked potential latency extension value; the head image data comprising diffusion tensor imaging data and magnetic resonance imaging data, the physiological data comprising a disease duration, an age, a systolic pressure, a GCS score, a muscle strength score, a Fugl-Meyer score and an Ashworth score, the CST injury level comprising complete, compression, partial rupture and complete rupture; determining, according to the head image data of each training sample, a FA value, a fiber bundle volume, a hemorrhage volume, a Euclidean minimum distance between a hemorrhage point and the CST and an edema compression grade corresponding to each training sample; preprocessing the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the edema compression grade, the disease duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score and the Ashworth score of each training sample to obtain a feature matrix corresponding to each training sample; iteratively training a neural network model according to the feature matrix corresponding to each training sample and the CST injury level to obtain a trained neural network model; and determining, according to the transcranial magnetic stimulation data, the head image data and the physiological data of a to-be-detected individual, the CST injury level of the to-be-detected individual by using the trained neural network model.

[0007] The method provided by the present application can comprehensively and deeply mine the feature information related to the CST injury level by obtaining the training sample set comprising the transcranial magnetic stimulation data, the head image data, the physiological data and the CST injury level, and by comprehensively analyzing and processing the multi-modal data. The transcranial magnetic stimulation data reflects the neural electrophysiological activity, the head image data directly presents the brain structure, and the physiological data reflects the overall physical condition of the patient. The fusion of multi-modal data overcomes the limitations of single data source, making the model training more reliable. By using these multi-modal data to construct a feature matrix and train a neural network model, the accurate detection of the CST injury level of the to-be-detected individual is finally realized, the accuracy and comprehensiveness of the detection are improved, and more abundant and reliable basis for subsequent clinical diagnosis and treatment is provided. That is, the method provided by the present application can quickly and accurately detect the CST injury level, determine the corresponding rehabilitation scheme according to the injury level, realize the accurate grading of the CST injury and the dynamic matching of the rehabilitation strategy, and improve the accuracy of the detection result.

[0008] In a possible implementation manner of the first aspect, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, and the edema compression grade corresponding to each training sample are determined according to the head image data of each training sample, and the determination includes: determining the three-dimensional gravity center coordinates of the hemorrhage point according to the magnetic resonance imaging data of each training sample; constructing a CST three-dimensional model according to the diffusion tensor imaging data of each training sample; and determining the Euclidean minimum distance between the three-dimensional gravity center coordinates of the hemorrhage point and the CST three-dimensional model by using a k-d tree space acceleration algorithm, to obtain the Euclidean minimum distance between the hemorrhage point and the CST.

[0009] The method provided by the application determines the three-dimensional gravity center coordinates of the hemorrhage point by using the magnetic resonance imaging data, constructs the CST three-dimensional model by using the diffusion tensor imaging data, and determines the Euclidean minimum distance between the hemorrhage point and the CST by using the k-d tree space acceleration algorithm. The accurate calculation manner can quantify the spatial relationship between the hemorrhage point and the CST, provides an accurate quantitative index for evaluating the affected degree of the CST, and further improves the detection accuracy of the model based on the spatial relationship between the hemorrhage point and the CST as characteristic information, thereby improving the accuracy of the detection result.

[0010] In a possible implementation manner of the first aspect, the edema compression grade includes low, medium or high; and the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, and the edema compression grade corresponding to each training sample are determined according to the head image data of each training sample, and the determination further includes: determining the edema compression grade as low when the edema region and the CST do not overlap; determining the edema compression grade as medium when the edema region and the CST partially overlap and the cross-sectional area is less than or equal to 50%; and determining the edema compression grade as high when the edema region wraps the CST or the cross-sectional area is greater than 50%.

[0011] The method provided by the application defines and divides the edema compression grade, divides the edema region and the CST into three levels of low, medium and high according to the overlapping condition of the edema region and the CST. The grading manner can systematically evaluate the compression degree of the edema on the CST, provides an accurate quantitative index for evaluating the affected degree of the CST from the spatial position relationship and the influence range of the edema and the CST, and further improves the detection accuracy of the model based on the spatial position relationship between the edema and the CST as characteristic information, thereby improving the accuracy of the detection result.

[0012] In a possible implementation of the first aspect, the method further includes: determining a corresponding rehabilitation strategy from a preset rehabilitation strategy database according to the CST damage level of the individual to be detected, the preset rehabilitation strategy database storing a rehabilitation strategy corresponding to each CST damage level; and determining a corresponding surgical strategy from a preset surgical strategy database according to the CST damage level of the individual to be detected, the preset surgical strategy database storing a surgical strategy corresponding to each CST damage level.

[0013] The method provided by the application can realize direct association of the detection result and the rehabilitation treatment, and the preset rehabilitation strategy database and the preset surgical strategy database established based on a large amount of clinical experience and research can provide personalized rehabilitation schemes and surgical strategies for patients with different CST damage levels. In actual application, when the CST damage level of a patient is determined, the corresponding rehabilitation strategy can be quickly matched, the blindness and randomness of the formulation of the rehabilitation scheme and the surgical scheme are avoided, the rehabilitation treatment is more targeted and scientific, and the efficiency and effect of the rehabilitation treatment are improved.

[0014] In a possible implementation of the first aspect, the motion evoked potential amplitude ratio, the motion evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the edema compression grade, the onset duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score and the Ashworth score of each training sample are preprocessed to obtain a feature matrix corresponding to each training sample, including: performing Z-score standardization processing on the motion evoked potential amplitude ratio, the motion evoked potential latency extension value, the FA value, the fiber bundle volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age and the systolic pressure of each training sample to obtain a standardized numerical feature; performing Min-Max normalization processing on the GCS score, the muscle strength score, the Fugl-Meyer score and the Ashworth score of each training sample to obtain a normalized score feature; performing one-hot encoding on the edema compression grade of each training sample to obtain an edema compression grade feature; and splicing the standardized numerical feature, the normalized score feature and the edema compression grade feature of each training sample to obtain the feature matrix corresponding to each training sample.

[0015] The method provided by the application adopts different preprocessing methods for different types of data, performs Z-score standardization processing on numerical value type data, performs Min-Max normalization processing on scoring type data, and performs one-hot encoding on classification data, and then performs feature splicing to obtain a feature matrix. This preprocessing method can make data of different types and magnitudes have a unified scale and format, eliminate the dimensional difference between data, improve the comparability and usability of data. In the model training process, the preprocessed feature matrix can make the neural network model better learn the data features, improve the convergence speed and training efficiency of the model, and at the same time, enhance the generalization ability of the model, reduce the overfitting phenomenon, thereby improving the accuracy and stability of the CST injury level detection.

[0016] In a possible implementation of the first aspect, the Z-score standardization formula is: ; Wherein, is the i-th item in the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age and the systolic pressure; is the average value of the i-th item in the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age and the systolic pressure in the training sample set; is the standard deviation of the i-th item in the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age and the systolic pressure in the training sample set; is the normalized numerical feature of the i-th item in the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age and the systolic pressure; The Min-Max normalization formula is: ; Wherein, is the GCS score, the muscle strength score, the Fugl-Meyer score or the Ashworth score of each training sample, is the minimum value of the GCS score, the muscle strength score, the Fugl-Meyer score or the Ashworth score in the training sample set, is the maximum value of the GCS score, the muscle strength score, the Fugl-Meyer score or the Ashworth score in the training sample set, a normalized score feature of the GCS score, the muscle strength score, the Fugl-Meyer score, or the Ashworth score of each training sample.

[0017] The method provided by the application can provide accurate quantitative basis for data preprocessing through the Z-score standardization formula and the Min-Max normalization formula, and ensure the consistency and standardization of data processing. In the model training process, the preprocessed feature matrix can make the neural network model better learn the data features, improve the convergence speed and training efficiency of the model, enhance the generalization ability of the model, reduce the overfitting phenomenon, and thus improve the accuracy and stability of the CST injury level detection.

[0018] In a possible implementation manner of the first aspect, the neural network model is iteratively trained according to the feature matrix corresponding to each training sample and the CST injury level, to obtain the trained neural network model, including: constructing a target loss function; iteratively training the neural network model according to the feature matrix corresponding to each training sample and the CST injury level based on the target loss function, to obtain the trained neural network model. The target loss function L is: ; is a class weight of the class c, and the class weights of the complete, compressed and partially broken classes are 1, and the class weight of the completely broken class is 2; is a probability that the sample predicted by the neural network model belongs to the class c, is a focusing coefficient, and is equal to 2, and yc is a one-hot encoding corresponding to the class c.

[0019] The method provided by the application iteratively trains the neural network model based on the target loss function. By introducing the class weight and the focusing coefficient, different attention degrees are given to samples of different classes, especially the attention degree to the completely broken class is improved. In the training process, the model can pay more attention to difficult samples and important classes, so that the model pays more attention to distinguishing different CST injury levels in the learning process, and effectively solves the influence of the data class imbalance problem on the model training. The finally trained neural network model can more accurately identify and judge different CST injury levels, and improve the classification accuracy and robustness of the model.

[0020] In a possible implementation manner of the first aspect, the method further includes: displaying, on a preset interface, a confidence degree corresponding to the CST injury level of the to-be-detected individual; and in a case where the confidence degree corresponding to the CST injury level of the to-be-detected individual is less than 70%, displaying, on the preset interface, an alarm information.

[0021] The method provided by the application displays the confidence corresponding to the CST damage level of the individual to be detected at a preset interface, and displays an alarm information when the confidence is less than 70%. The display of the confidence can enable the technician to intuitively understand the reliability of the detection result of the model. When the confidence is low, the alarm information can attract the attention of the technician, prompting the technician to re-examine the detection process or further supplement the inspection, so as to avoid the false judgment caused by the model misjudgment. In this way, the reliability and safety of the detection result are improved.

[0022] In a second aspect, the application provides a CST damage level detection system based on multi-modal data fusion, which comprises: a data acquisition module, configured to acquire a training sample set, the training sample set comprising a plurality of training samples, each training sample comprising transcranial magnetic stimulation data, head image data, physiological data and a CST damage level of a stroke individual, the transcranial magnetic stimulation data comprising a motor evoked potential amplitude ratio and a motor evoked potential latency extension value; the head image data comprising diffusion tensor imaging data and magnetic resonance imaging data, the physiological data comprising a disease duration, an age, a systolic pressure, a GCS score, a muscle strength score, a Fugl-Meyer score and an Ashworth score, and the CST damage level comprising complete, compression, partial rupture and complete rupture; an image processing module, configured to determine, according to the head image data of each training sample, a FA value, a fiber bundle volume, a hemorrhage volume, a Euclidean minimum distance between a hemorrhage point and the CST and an edema compression grade corresponding to each training sample; a feature determination module, configured to preprocess the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the Euclidean minimum distance between the hemorrhage point and the CST, the edema compression grade, the disease duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score and the Ashworth score of each training sample, to obtain a feature matrix corresponding to each training sample; a model training module, configured to iteratively train a neural network model according to the feature matrix corresponding to each training sample and the CST damage level, to obtain a trained neural network model; and a detection module, configured to determine, according to the transcranial magnetic stimulation data, the head image data and the physiological data of an individual to be detected, the CST damage level of the individual to be detected by the trained neural network model.

[0023] In a third aspect, an electronic device is provided, which comprises a memory and one or more processors; the memory is coupled to the processors; and the memory stores computer program codes, which comprise computer instructions, when executed by the processors, causing the electronic device to perform the method in any implementation manner of the first aspect.

[0024] In a fourth aspect, a computer-readable storage medium is provided, including computer instructions, which, when executed on an electronic device, cause the electronic device to perform the method in any implementation manner of the first aspect.

[0025] In a fifth aspect, a computer program product is provided, which, when executed on a computer, causes the computer to perform the method in any implementation manner of the first aspect.

[0026] It can be understood that the beneficial effects that can be achieved by the system of the second aspect, the electronic device of the third aspect, the computer-readable storage medium of the fourth aspect, and the computer program product of the fifth aspect are referable to the beneficial effects in the first aspect and any possible design manner thereof, which will not be described herein again. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 1. Figure 2 A flowchart of a CST injury level detection method based on multi-modal data fusion provided by an embodiment of the present application is shown in FIG. 2. Figure 3 A flowchart of another CST injury level detection method based on multi-modal data fusion provided by an embodiment of the present application is shown in FIG. 3. Figure 4 A structural schematic diagram of a detection system provided by an embodiment of the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings. In the description of the present application, unless otherwise specified, “ / ” represents an “or” relationship between the objects before and after the “ / ”. For example, A / B can represent A or B. In the present application, “or” is only a description of the relationship between the objects, which means that there can be three relationships, for example, A or B, which means that A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. In addition, in the description of the present application, unless otherwise specified, “multiple” means two or more. “At least one (one)” or the like means any combination of these items, including any combination of single (one) or multiple items.

[0029] In addition, in order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, “first”, “second”, etc. are used to distinguish the same items or similar items with basically the same functions and effects. Those skilled in the art can understand that “first”, “second”, etc. do not limit the quantity and execution order, and “first”, “second”, etc. also do not necessarily mean different.

[0030] Meanwhile, in the embodiments of the present application, the words such as "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any implementation or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being superior or superior to other implementation or design solutions. Rather, the words "exemplary" or "for example" are used to present the relevant concept in a specific manner, facilitating understanding.

[0031] In the field of neurorehabilitation, central nervous system injuries such as cerebral hemorrhage and cerebral infarction often lead to damage to the corticospinal tract (CST), which in turn causes motor dysfunction in patients and seriously affects their quality of life. How to accurately assess the degree of CST damage and develop targeted rehabilitation strategies has always been a problem to be solved, which not only relates to the rehabilitation effect of patients, but also has important significance for reducing medical costs.

[0032] At present, the evaluation and rehabilitation of CST damage are usually based on imaging evaluation methods, such as detecting the structural integrity of CST by detecting parameters such as anisotropy fraction (FA) through diffusion tensor imaging (DTI), obtaining the detection results, and developing corresponding rehabilitation programs according to the detection results.

[0033] However, the existing technology has obvious limitations, i.e., it cannot achieve accurate grading of CST damage and dynamic matching of surgical rehabilitation strategies. Moreover, based on a single imaging evaluation method, it is difficult to comprehensively and accurately reflect the damage state of CST, resulting in large errors in the detection results and poor accuracy.

[0034] In view of this, the embodiment of the present application provides a CST injury level detection method based on multi-modal data fusion, which comprises: obtaining a training sample set, the training sample set comprising a plurality of training samples, each training sample comprising transcranial magnetic stimulation data, head image data, physiological data and a CST injury level of a stroke individual, the transcranial magnetic stimulation data comprising a motor evoked potential amplitude ratio and a motor evoked potential latency extension value; the head image data comprising diffusion tensor imaging data and magnetic resonance imaging data, the physiological data comprising an onset duration, an age, a systolic pressure, a GCS score, a muscle strength score, a Fugl-Meyer score and an Ashworth score, and the CST injury level comprising complete, compression, partial rupture and complete rupture; determining, according to the head image data of each training sample, a FA value, a fiber bundle volume, a hemorrhage volume, a Euclidean minimum distance between a hemorrhage point and the CST and an edema compression grade corresponding to each training sample; preprocessing the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the edema compression grade, the onset duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score and the Ashworth score of each training sample to obtain a feature matrix corresponding to each training sample; iteratively training a neural network model according to the feature matrix corresponding to each training sample and the CST injury level to obtain a trained neural network model; and determining, according to the transcranial magnetic stimulation data, the head image data and the physiological data of a to-be-detected individual, the CST injury level of the to-be-detected individual through the trained neural network model.

[0035] The method provided by the embodiment of the present application can comprehensively and deeply mine the feature information related to the CST injury level by obtaining the training sample set comprising the transcranial magnetic stimulation data, the head image data, the physiological data and the CST injury level, and by comprehensively analyzing and processing the multi-modal data. The transcranial magnetic stimulation data reflects the neural electrophysiological activity, the head image data directly presents the brain structure, and the physiological data reflects the overall physical condition of the patient. The fusion of the multi-modal data overcomes the limitations of single data source, making the model training more reliable. By using these multi-modal data to construct a feature matrix and train a neural network model, the accurate detection of the CST injury level of the to-be-detected individual is finally realized, the accuracy and comprehensiveness of the detection are improved, and more abundant and reliable basis for subsequent clinical diagnosis and treatment is provided. That is, the method provided by the present application can quickly and accurately detect the CST injury level, determine the corresponding rehabilitation scheme according to the injury level, realize the accurate grading of the CST injury and the dynamic matching of the rehabilitation strategy, and improve the accuracy of the detection result.

[0036] In some embodiments, the CST injury level detection method based on multi-modal data fusion provided by the embodiments of the present application can be executed by a CST injury level detection system 100 based on multi-modal data fusion (hereinafter referred to as detection system 100).

[0037] As an example, the detection system 100 can be any electronic device 200 with data processing capability, such as a general-purpose computer, a personal computer, a notebook computer, a switch, or a tablet computer, etc., and the specific implementation of the detection system 100 is not limited here.

[0038] Figure 1 The hardware structure schematic diagram of the electronic device provided by the embodiments of the present application is shown. The electronic device 200 includes a processor 210, a memory 220, and a communication interface 230.

[0039] The processor 210 can include one or more processing cores. The processor 210 connects various parts in the electronic device 200 through various interfaces and lines, executes various functions of the electronic device 200 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 220, and calling data stored in the memory 220. Optionally, the processor 210 can be implemented in at least one of the following hardware forms: a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA).

[0040] The memory 220 can include a random access memory (RAM) and can also include a read-only memory (ROM). Optionally, the memory 220 includes a non-transitory computer-readable storage medium. The memory 220 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 220 can include a program storage area. The program storage area can store instructions for implementing an operating system, instructions for implementing at least one function (such as a data acquisition function, a model training function, and a detection function), and instructions for implementing each method embodiment described above.

[0041] The communication interface 230 is configured to communicate with other devices, apparatuses or communication networks, such as data storage devices, image processing apparatuses or Ethernet, Radio Access Network (RAN), wireless local area networks (WLAN) and the like.

[0042] In physical implementation, the above-mentioned devices (e.g., the processor 210, the memory 220 and the communication interface 230) can be devices in the same apparatus (e.g., a notebook computer) respectively. Alternatively, at least two of the devices can be arranged in the same apparatus as different devices in the apparatus, similar to the deployment of devices or devices in a distributed system.

[0043] It can be understood that the structure illustrated in the embodiment does not constitute a specific limitation on the electronic device 200. In other embodiments of the present application, the electronic device 200 can include more or fewer components than those illustrated, or combine certain components, or split certain components, or different arrangement of components. The illustrated components can be implemented in hardware, software or a combination of software and hardware.

[0044] An embodiment of the present application provides a CST injury level detection method based on multi-modal data fusion, which is described below in combination with the accompanying drawings.

[0045] Figure 2 An embodiment of the present application provides a CST injury level detection method based on multi-modal data fusion, which is described below in combination with the accompanying drawings. Figure 1 The method can be executed by the electronic device 200 shown in the figure. The method can include the following steps: S1, obtaining a training sample set, the training sample set including a plurality of training samples.

[0046] Each training sample includes transcranial magnetic stimulation data, head image data, physiological data and CST injury level of a stroke individual, the transcranial magnetic stimulation data including motor evoked potential amplitude ratio and motor evoked potential latency extension value; the head image data including diffusion tensor imaging data and magnetic resonance imaging data, the physiological data including onset duration, age, systolic pressure, GCS score, muscle strength score, Fugl-Meyer score and Ashworth score, and the CST injury level including complete, compression, partial rupture and complete rupture; Specifically, the motor evoked potential amplitude ratio is a ratio of the amplitude of the target muscle electrical signal after transcranial magnetic stimulation to the baseline value, which is used to evaluate the conduction efficiency of the corticospinal tract (CST); a decrease in the value indicates damage to the neural pathway. The motor evoked potential latency extension value is the time delay of the muscle electrical signal after stimulation, and the extension indicates a decrease in the speed of neural signal conduction, which can be caused by myelin damage or axon lesion.

[0047] Diffusion tensor imaging (DTI) data quantifies white matter fiber integrity by water molecule diffusion direction, especially for detecting microscopic structural damage of CST (e.g., reduced fractional anisotropy). Magnetic resonance imaging data (MRI) is used to provide high-resolution images of brain anatomy, which can identify the location, extent, and macroscopic impact on CST (e.g., compression or interruption) of stroke lesions.

[0048] GCS score is a consciousness score used to assess the consciousness disorder of stroke individuals, which is scored by three dimensions of eye opening response, language response and limb movement, with a total score of 3-15, and a lower score indicates a more severe consciousness disorder. Fugl-Meyer score is a standardized scale for quantifying motor function recovery (0-100 points), covering limb flexibility, coordination and reflex, and a higher score indicates better function. Ashworth score is a clinical tool for assessing muscle tone (0-4 levels), and a higher level indicates more severe spasticity (e.g., level 1 is mild resistance, and level 4 is joint stiffness). Muscle strength score is a standardized method for assessing the active contraction strength of muscles, mainly used to judge the functional status of nerves and muscles, and widely used in the fields of rehabilitation medicine, neurology, sports medicine, etc. Through muscle strength score, doctors or therapists can objectively record the strength of patients' muscle strength, judge the progress of disease, treatment effect or rehabilitation progress. Muscle strength score is divided into 0-5 levels It should be noted that each training sample can also include other more kinds and types of scores for evaluating other functions of stroke individuals, and the embodiments of the present application do not make special limitations thereto.

[0049] S2, determining the FA value, fiber volume, hemorrhage volume, Euclidean minimum distance between the hemorrhage point and the CST, and edema compression grade corresponding to each training sample according to the head image data of each training sample.

[0050] In one possible implementation, the above S2 includes: determining the three-dimensional center of gravity coordinates of the hemorrhage point according to the magnetic resonance imaging data of each training sample; constructing a three-dimensional model of the CST according to the diffusion tensor imaging data of each training sample; determining the Euclidean minimum distance between the three-dimensional center of gravity coordinates of the hemorrhage point and the three-dimensional model of the CST by a k-d tree space acceleration algorithm, to obtain the Euclidean minimum distance between the hemorrhage point and the CST.

[0051] The method provided by the embodiment of the application determines the three-dimensional gravity center coordinates of the hemorrhagic point through magnetic resonance imaging data, constructs a CST three-dimensional model by using diffusion tensor imaging data, and determines the Euclidean minimum distance between the hemorrhagic point and the CST by means of a k-d tree space acceleration algorithm. This accurate calculation method can quantify the spatial relationship between the hemorrhagic point and the CST, and provides an accurate quantitative index for evaluating the affected degree of the CST, and then the spatial relationship between the hemorrhagic point and the CST is taken as characteristic information to improve the detection accuracy of the model, thereby improving the accuracy of the detection result.

[0052] In another possible implementation, S2 includes: in the case that the edema region does not overlap with the CST, determining that the edema compression grade is low; in the case that the edema region partially overlaps with the CST and the cross-sectional area is less than or equal to 50%, determining that the edema compression grade is medium; and in the case that the edema region wraps the CST or the cross-sectional area is greater than 50%, determining that the edema compression grade is high.

[0053] The method provided by the embodiment of the application defines and divides the edema compression grade, and divides the edema region and the CST into three levels of low, medium and high according to the overlapping condition of the edema region and the CST. This grading method can systematically evaluate the compression degree of the edema on the CST, and provides an accurate quantitative index for evaluating the affected degree of the CST from the spatial position relationship and the influence range of the edema and the CST, and then the spatial position relationship of the edema and the CST is taken as characteristic information to improve the detection accuracy of the model, thereby improving the accuracy of the detection result.

[0054] S3, pre-processing the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhagic volume, the Euclidean minimum distance between the hemorrhagic point and the CST, the edema compression grade, the onset duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score and the Ashworth score of each training sample to obtain a feature matrix corresponding to each training sample; For example, the motor evoked potential amplitude ratio of the training sample A is 0.75; the motor evoked potential latency extension value is 5.2 ms; the FA value (fractional anisotropy) is 0.45; the fiber bundle volume (CST fiber number) is 8500 mm³; the hemorrhagic volume is 12.5 mL; the minimum Euclidean distance between the hemorrhagic point and the CST is 4.8 mm; and the edema compression grade of the training sample A is medium; the onset duration is 30 days; the age is 62 years old; the systolic pressure is 145 mmHg; in terms of functional evaluation, the Fugl-Meyer score (upper limb motor function) is 58 points (full score 66 points); the Ashworth score (muscle tension grade) is 2 levels (maximum 4 levels); the GCS score is 10 points (full score 15 points); and the muscle strength score (Lovett grade) is 4 levels (maximum 5 levels).

[0055] The feature matrix corresponding to training sample A is: [0.65,1.2,-0.8,0.72,0.52,0.45,0.5,0.5,0.67,0.88,0.5,0,1,0,0.67,0.8].

[0056] Wherein, 0.65 is the standardized amplitude ratio of motor evoked potentials; 1.2 is the standardized latency prolongation of motor evoked potentials; -0.8 is the standardized FA value (fractional anisotropy); 0.72 is the normalized fiber bundle volume (number of CST fibers); 0.52 is the normalized hemorrhage volume; 0.45 is the standardized minimum Euclidean distance between the hemorrhage point and the CST; 0.5 is the standardized duration of illness; 0.5 is the standardized age; 0.67 The normalized systolic blood pressure is 0.88; the normalized Fugl-Meyer score is 58 / 66≈0.88; the normalized Ashworth score is 0.5, grade 2 / 4=0.5; the next three dimensions are 0 (unique heat code: low), 1 (unique heat code: medium), 0 (unique heat code: high); the normalized GCS score is 0.67, 10 / 15≈0.67; the normalized muscle strength score is 0.8, grade 4 / 5=0.8.

[0057] In some embodiments, S3 above includes: performing Z-score standardization on the amplitude ratio of motor evoked potentials, the latency prolongation value of motor evoked potentials, the FA value, the fiber bundle volume, the hemorrhage volume, the minimum Euclidean distance between the hemorrhage point and the CST, the duration of onset, the age, and the systolic blood pressure of each training sample to obtain standardized numerical features; performing Min-Max normalization on the GCS score, muscle strength score, Fugl-Meyer score, and Ashworth score of each training sample to obtain normalized score features; performing one-heat encoding on the edema compression grade of each training sample to obtain edema compression grade features; and concatenating the standardized numerical features, normalized score features, and edema compression grade features of each training sample to obtain the feature matrix corresponding to each training sample.

[0058] Furthermore, the Z-score standardization formula is: ; in, The i-th item among the following: amplitude ratio of motor evoked potentials, latency prolongation of motor evoked potentials, FA value, fiber bundle volume, hemorrhage volume, minimum Euclidean distance between the hemorrhage point and CST, duration of illness, age, and systolic blood pressure; an average value of the i-th item in the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure in the training sample set; a standard deviation of the i-th item in the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure in the training sample set; a normalized numerical feature of the i-th item in the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure; a Min-Max normalization formula is as follows: ; wherein, a GCS score, a muscle strength score, a Fugl-Meyer score, or an Ashworth score of each training sample, a minimum value of the GCS score, the muscle strength score, the Fugl-Meyer score, or the Ashworth score in the training sample set, a maximum value of the GCS score, the muscle strength score, the Fugl-Meyer score, or the Ashworth score in the training sample set, a normalized score feature of the GCS score, the muscle strength score, the Fugl-Meyer score, or the Ashworth score of each training sample.

[0059] The method provided by the embodiment of the application can provide accurate quantitative basis for data preprocessing through the Z-score standardization formula and the Min-Max normalization formula, and ensure the consistency and standardization of data processing. In the model training process, the preprocessed feature matrix can make the neural network model better learn data features, improve the convergence speed and training efficiency of the model, enhance the generalization ability of the model, reduce the overfitting phenomenon, and thus improve the accuracy and stability of CST injury level detection.

[0060] S4. Iteratively training the neural network model according to the feature matrix corresponding to each training sample and the CST injury level, to obtain a trained neural network model.

[0061] In one example, the neural network model adopts a hybrid architecture of CNN+Attention+DNN, the input layer receives 14-dimensional pre-processed features, including 9-dimensional numerical features such as normalized MEP amplitude ratio, FA value, hemorrhage volume, fiber bundle volume, minimum distance, duration of onset, age, systolic blood pressure, normalized GCS score, muscle strength score, Fugl-Meyer score and Ashworth score, and 4-dimensional evaluation features such as 3-dimensional edema grading information; then the local feature relationship is extracted through two layers of 1D-CNN (using 64 and 128 convolutional kernels respectively, ReLU activation), focusing on the synergistic change pattern of physiological indicators such as FA value and MEP delay, converting the input into 9x128-dimensional time sequence features and connecting to 4 heads of self-attention layer, highlighting key feature combinations (such as reinforcing the joint influence of hemorrhage volume and nerve fiber distance on the degree of rupture) through a dynamic weight distribution mechanism; after global average pooling compression into a 128-dimensional vector, it is transmitted into a fully connected network (256→128 neurons, ReLU activation with Dropout regularization) for high-order feature fusion, and finally a four-class probability (complete / pressed / partially ruptured / fully ruptured) is generated by the Softmax output layer combined with Focal Loss (γ=2, the weight of the fully ruptured class is set to 2), the overall Adam optimizer (learning rate 0.001) is used for training, and the early stopping mechanism is used to monitor the performance of the validation set to prevent overfitting.

[0062] In a possible implementation, referring to Figure 3 The S4 includes the following steps: S41, constructing a target loss function.

[0063] S42, based on the target loss function, iteratively training the neural network model according to the feature matrix and the CST damage level corresponding to each training sample, to obtain a trained neural network model.

[0064] The target loss function L is: ; is the class weight of class c, and the class weights of complete, pressed and partially ruptured are 1, and the class weight of fully ruptured is 2; is the probability that the neural network model predicts that the sample belongs to class c, is the focus coefficient, and is equal to 2, and yc is the one-hot encoding corresponding to class c.

[0065] Specifically, the method provided by the application introduces class weights into the target loss function, and sets the class weight of complete fracture to be greater than that of other damage levels, so that the misclassification of complete fracture will bring greater loss, causing the model to be more inclined to correctly identify these high-risk classes during training. Furthermore, to further solve the class imbalance problem, the method provided by the application combines the Focal Loss function on the basis of the weighted cross-entropy, and sets a focus parameter to further reduce the weight of easy-to-classify samples, so that the model pays more attention to difficult-to-classify samples, and the recall rate and specificity of the neural network model can be optimally balanced.

[0066] The method provided by the embodiment of the application trains the neural network model based on the target loss function. By introducing class weights and focus coefficients, different attention degrees are given to samples of different classes, and the attention degree to complete fracture is particularly improved. During the training process, the model can pay more attention to difficult samples and important classes, so that the model pays more attention to distinguishing different CST damage levels during the learning process, and effectively solves the influence of the data class imbalance problem on the model training. Finally, the trained neural network model can more accurately identify and judge different CST damage levels, and improves the classification accuracy and robustness of the model.

[0067] S5, determining the CST damage level of the to-be-detected individual by the trained neural network model based on the transcranial magnetic stimulation data, the head image data and the physiological data of the to-be-detected individual.

[0068] As can be seen from the above S1-S5, the method provided by the application can obtain a training sample set containing transcranial magnetic stimulation data, head image data, physiological data and CST damage levels, and can comprehensively and deeply mine feature information related to the CST damage level by comprehensively analyzing and processing the multi-modal data. The transcranial magnetic stimulation data reflects the neural electrophysiological activity, the head image data directly presents the brain structure, and the physiological data reflects the overall physical condition of the patient. The fusion of multi-modal data overcomes the limitations of single data source, making the model training more reliable. By using these multi-modal data to construct a feature matrix and train a neural network model, the accurate detection of the CST damage level of the to-be-detected individual is finally realized, the accuracy and comprehensiveness of the detection are improved, and more reliable basis for subsequent clinical diagnosis and treatment is provided. That is, the method provided by the application can quickly and accurately detect the CST damage level, and determine the corresponding rehabilitation scheme according to the damage level, realize the accurate grading of the CST damage and the dynamic matching of the rehabilitation strategy, and improve the accuracy of the detection result.

[0069] In some embodiments, the method provided by the embodiments of the present application further comprises: determining a corresponding rehabilitation strategy from a preset rehabilitation strategy database according to the CST damage level of the individual to be detected.

[0070] Further, the preset rehabilitation strategy database stores a corresponding rehabilitation strategy for each CST damage level. Wherein, the corresponding rehabilitation strategy for the CST damage level of complete is non-targeted motor rehabilitation, routine stroke secondary prevention (blood pressure / lipid control), and monthly DTI follow-up monitoring. The corresponding rehabilitation strategy for the CST damage level of compression is low-frequency repetitive transcranial magnetic stimulation combined with task-oriented training, edema elimination drugs, and 3 times of motor evoked potential monitoring per week. The corresponding rehabilitation strategy for the CST damage level of partial rupture is high-frequency rTMS combined with mirror neuron training, white matter repair drugs, and customized brace anti-spasticity. The corresponding rehabilitation strategy for the CST damage level of complete rupture is spinal cord stimulation (SCS) combined with exoskeleton robot training, stem cell transplantation evaluation, and daily Ashworth score dynamic adjustment of anti-spasticity scheme.

[0071] In other embodiments, the method provided by the embodiments of the present application further comprises: determining a corresponding surgical strategy from a preset surgical strategy database according to the CST damage level of the individual to be detected, the preset surgical strategy database storing a corresponding surgical strategy for each CST damage level.

[0072] The method provided by the embodiments of the present application determines a corresponding rehabilitation strategy from a preset rehabilitation strategy database according to the CST damage level of the individual to be detected, and determines a corresponding surgical strategy from a preset surgical strategy database according to the CST damage level of the individual to be detected. The direct correlation between the detection result and the rehabilitation treatment can be realized, the preset rehabilitation strategy database and the preset surgical strategy database established based on a large amount of clinical experience and research can provide personalized rehabilitation schemes and surgical strategies for patients with different CST damage levels. In actual application, after the CST damage level of the patient is determined, the corresponding rehabilitation strategy can be quickly matched, avoiding the blindness and randomness of the formulation of the rehabilitation scheme and the surgical scheme, so that the rehabilitation treatment is more targeted and scientific, and the efficiency and effect of the rehabilitation treatment are improved.

[0073] In a possible implementation manner, the method provided by the embodiments of the present application further comprises: displaying a confidence degree corresponding to the CST damage level of the individual to be detected on a preset interface; and in a case where the confidence degree corresponding to the CST damage level of the individual to be detected is less than 70%, displaying an alarm information on the preset interface.

[0074] The method provided by the embodiment of the present application can display the confidence corresponding to the CST damage level of the individual to be detected on a preset interface, and display an alarm information when the confidence is less than 70%. The display of the confidence can enable the technician to intuitively understand the reliability of the detection result of the model. When the confidence is low, the alarm information can attract the attention of the technician, prompting the technician to re-examine the detection process or further supplement the inspection, so as to avoid the false judgment caused by the model misjudgment. In this way, the reliability and safety of the detection result are improved.

[0075] The above mainly describes the scheme of the embodiment of the present application from the perspective of the method. It can be understood that the detection system 100 comprises at least one of the corresponding hardware structure and software module for implementing each function. Those skilled in the art should easily realize that, in combination with the units and algorithm steps of each example described in the embodiments disclosed in the present application, the embodiment of the present application can be realized in the form of hardware or the combination of hardware and computer software. Whether a certain function is executed by hardware or computer software driven hardware depends on the specific application and design constraints of the technical scheme. The professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiment of the present application.

[0076] The embodiment of the present application can divide the detection system 100 into functional units according to the above method examples. For example, the detection system 100 can be divided into functional units corresponding to each function, or two or more functions can be integrated into one processing unit. The integrated unit can be realized in the form of hardware or software functional unit. It should be noted that the division of units in the embodiment of the present application is illustrative, and is only a logical functional division. In actual implementation, there can be another division method.

[0077] An exemplary, Figure 4A hardware structure schematic diagram of a detection system provided by an embodiment of the present application is shown. The detection system 100 comprises: a data acquisition module 110, configured to acquire a training sample set, the training sample set comprising a plurality of training samples, each training sample comprising transcranial magnetic stimulation data, head image data, physiological data and a CST injury level of a stroke individual, the transcranial magnetic stimulation data comprising a motor evoked potential amplitude ratio and a motor evoked potential latency extension value; the head image data comprising diffusion tensor imaging data and magnetic resonance imaging data, the physiological data comprising an onset duration, an age, a systolic pressure, a GCS score, a muscle strength score, a Fugl-Meyer score and an Ashworth score, and the CST injury level comprising complete, compression, partial rupture and complete rupture; an image processing module 120, configured to determine, according to the head image data of each training sample, a FA value, a fiber bundle volume, a hemorrhage volume, a Euclidean minimum distance between a hemorrhage point and the CST and an edema compression grade corresponding to each training sample; a feature determination module 130, configured to preprocess the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the edema compression grade, the onset duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score and the Ashworth score of each training sample, to obtain a feature matrix corresponding to each training sample; a model training module 140, configured to iteratively train a neural network model according to the feature matrix corresponding to each training sample and the CST injury level, to obtain a trained neural network model; and a detection module 150, configured to determine, according to transcranial magnetic stimulation data, head image data and physiological data of a to-be-detected individual, the CST injury level of the to-be-detected individual by using the trained neural network model.

[0078] It should be understood that the specific description about the optional manners above can refer to the foregoing method embodiments, which will not be described herein again. In addition, the explanation and beneficial effect of any one of the detection systems 100 provided above can refer to the corresponding method embodiments above, which will not be described herein again.

[0079] The embodiment of the present application further provides a computer readable storage medium, which stores at least one computer instruction, and the at least one computer instruction is loaded and executed by a processor to realize the method of each of the above embodiments. The explanation and beneficial effect of any one of the computer readable storage media provided above can refer to the corresponding embodiments above, which will not be described herein again.

[0080] The embodiment of the present application further provides a chip. The chip integrates a control circuit and one or more ports for realizing the functions of the detection system 100 described above. Optionally, the functions supported by the chip can refer to the foregoing, which will not be described herein again.

[0081] Those skilled in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by programs instructing relevant hardware. The programs can be stored in a computer readable storage medium. The storage medium mentioned above can be a read-only memory, a random access memory, etc. The processing unit or processor mentioned above can be a central processing unit, a general-purpose processor, an application specific integrated circuit (ASIC), a microprocessor (digital signal processor, DSP), a field programmable gate array (FPGA) or other programmable logic device, a transistor logic device, a hardware component or any combination thereof.

[0082] The embodiments of the present application also provide a computer program product containing instructions, which, when executed on a computer, cause the computer to perform any of the methods of the above-mentioned embodiments. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the flow or function according to the embodiments of the present application is generated wholly or partially. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that the computer can access or a data storage device such as a server, data center, etc. integrated with one or more available media sets. The available media can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as an SSD), etc.

[0083] It should be noted that the devices for storing computer instructions or computer programs provided by the embodiments of the present application, such as but not limited to the above-mentioned memories, computer readable storage media, communication chips and the like, are all non-transitory. Those skilled in the art should be aware that in one or more examples described above, the functions described by the embodiments of the present application can be implemented by hardware, software, firmware or any combination thereof. When implemented by software, these functions can be stored in a computer readable storage medium or transmitted as one or more instructions or codes on a computer readable storage medium. The computer readable storage medium includes computer storage medium and communication medium, wherein the communication medium includes any medium that facilitates the transfer of computer programs from one place to another. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0084] Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for detecting CST injury level based on multi-modal data fusion, characterized in that, The method comprises: obtaining a training sample set, the training sample set comprising a plurality of training samples, each training sample comprising transcranial magnetic stimulation data, head image data, physiological data and CST injury level of a stroke individual, the transcranial magnetic stimulation data comprising motor evoked potential amplitude ratio and motor evoked potential latency extension value; the head image data comprising diffusion tensor imaging data and magnetic resonance imaging data, the physiological data comprising onset duration, age, systolic pressure, GCS score, muscle strength score, Fugl-Meyer score and Ashworth score, and the CST injury level comprising complete, compression, partial rupture and complete rupture; determining the FA value, fiber bundle volume, hemorrhage volume, Euclidean minimum distance between the hemorrhagic point and the CST and edema compression classification corresponding to each training sample according to the head image data of each training sample; preprocessing the motor evoked potential amplitude ratio, motor evoked potential latency extension value, FA value, fiber bundle volume, hemorrhage volume, Euclidean minimum distance between the hemorrhagic point and the CST, edema compression classification, onset duration, age, systolic pressure, GCS score, muscle strength score, Fugl-Meyer score and Ashworth score of each training sample to obtain the feature matrix corresponding to each training sample; iteratively training a neural network model according to the feature matrix corresponding to each training sample and the CST injury level to obtain a trained neural network model; determining the CST injury level of a to-be-detected individual through the trained neural network model according to the transcranial magnetic stimulation data, head image data and physiological data of the to-be-detected individual.

2. The method of claim 1, wherein, The method comprises: determining the three-dimensional gravity center coordinates of the hemorrhagic point according to the magnetic resonance imaging data of each training sample; constructing a CST three-dimensional model according to the diffusion tensor imaging data of each training sample; determining the Euclidean minimum distance between the three-dimensional gravity center coordinates of the hemorrhagic point and the CST three-dimensional model through a k-d tree space acceleration algorithm to obtain the Euclidean minimum distance between the hemorrhagic point and the CST.

3. The method of claim 2, wherein, The edema compression classification comprises low, medium or high; and the method further comprises: determining the edema compression classification as low when the edema area and the CST do not overlap; determining the edema compression classification as medium when the edema area and the CST partially overlap and the cross-sectional area is less than or equal to 50%; determining the edema compression classification as high when the edema area wraps the CST or the cross-sectional area is greater than 50%.

4. The method of claim 3, wherein, The method further comprises: determine a corresponding rehabilitation strategy from a preset rehabilitation strategy database according to the CST damage level of the individual to be detected, the preset rehabilitation strategy database storing a corresponding rehabilitation strategy for each CST damage level; determine a corresponding surgical strategy from a preset surgical strategy database according to the CST damage level of the individual to be detected, the preset surgical strategy database storing a corresponding surgical strategy for each CST damage level.

5. The method of claim 1, wherein, The method further comprises: performing pretreatment on the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the edema compression grade, the onset duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score, and the Ashworth score of each training sample to obtain a feature matrix corresponding to each training sample, including: performing Z-score standardization processing on the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure of each training sample to obtain standardized numerical features; performing Min-Max normalization processing on the GCS score, the muscle strength score, the Fugl-Meyer score, and the Ashworth score of each training sample to obtain normalized score features; performing one-hot encoding on the edema compression grade of each training sample to obtain edema compression grade features; 6. The method of claim 5, wherein, concatenating the standardized numerical features, the normalized score features, and the edema compression grade features of each training sample to obtain a feature matrix corresponding to each training sample. ； wherein, is the i-th item among the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure; is the average value of the i-th item among the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure in the training sample set; is the standard deviation of the i-th item among the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure in the training sample set; is the standardized numerical feature of the i-th item among the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the onset duration, the age, and the systolic pressure. The Z-score standardization formula is: ； wherein, GCS score, muscle strength score, Fugl-Meyer score, or Ashworth score of each training sample, minimum value of GCS score, muscle strength score, Fugl-Meyer score, or Ashworth score in the training sample set, maximum value of GCS score, muscle strength score, Fugl-Meyer score, or Ashworth score in the training sample set, normalized score feature of GCS score, muscle strength score, Fugl-Meyer score, or Ashworth score of each training sample.

7. The method of claim 6, wherein, The Min-Max normalization formula is: The method further comprises: constructing a target loss function; iteratively training the neural network model according to the feature matrix corresponding to each training sample and the CST damage level based on the target loss function to obtain a trained neural network model; ； is a class weight for class c, and the class weight for complete, under compression, and partially broken is 1, and the class weight for fully broken is 2; is a probability that the neural network model predicts that the sample belongs to class c, is a focus coefficient, and is equal to 2, and yc is a one-hot encoding corresponding to class c.

8. The method of claim 7, wherein, The target loss function L is: The method further comprises: displaying the confidence corresponding to the CST damage level of the individual to be detected on a preset interface; 9.A CST injury level detection system based on multi-modal data fusion, characterized in that, in a case where the confidence corresponding to the CST damage level of the individual to be detected is less than 70%, displaying an alarm information on the preset interface. The system comprises: The data acquisition module is configured to acquire a training sample set, the training sample set including a plurality of training samples, each training sample including transcranial magnetic stimulation data, head image data, physiological data, and a CST injury level of a stroke individual, the transcranial magnetic stimulation data including a motor evoked potential amplitude ratio and a motor evoked potential latency extension value, the head image data including diffusion tensor imaging data and magnetic resonance imaging data, the physiological data including an onset duration, an age, a systolic pressure, a GCS score, a muscle strength score, a Fugl-Meyer score, and an Ashworth score, and the CST injury level including complete, compression, partial rupture, and complete rupture. The image processing module is configured to determine, according to the head image data of each training sample, a corresponding FA value, a fiber bundle volume, a hemorrhage volume, a Euclidean minimum distance between a hemorrhage point and the CST, and an edema compression grade of each training sample. The feature determination module is configured to preprocess the motor evoked potential amplitude ratio, the motor evoked potential latency extension value, the FA value, the fiber bundle volume, the hemorrhage volume, the Euclidean minimum distance between the hemorrhage point and the CST, the edema compression grade, the onset duration, the age, the systolic pressure, the GCS score, the muscle strength score, the Fugl-Meyer score, and the Ashworth score of each training sample to obtain a corresponding feature matrix of each training sample. The model training module is configured to iteratively train a neural network model according to the corresponding feature matrix and the CST injury level of each training sample to obtain a trained neural network model. The detection module is configured to determine a CST injury level of a to-be-detected individual by the trained neural network model according to transcranial magnetic stimulation data, head image data, and physiological data of the to-be-detected individual.

10. An electronic device, comprising: The method comprises: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the method for detecting a CST injury level based on multi-modal data fusion according to any one of claims 1-8.

Images