Image recognition technology-based nervous system disease teaching and training method and system

Abstract

The invention discloses a nervous system disease teaching and training method and system based on an image recognition technology, and particularly relates to the field of nervous system disease teaching, and the method comprises the steps: obtaining a first dynamic image sequence and interaction behavior data of a user, marking a first dynamic focus segmentation map through an image recognition model, and obtaining a first dynamic focus segmentation map; performing feature comparison operation with a second dynamic image sequence generated by simulation so as to iteratively optimize the image recognition model; and obtaining an interactive feedback operation of the user on the second dynamic image sequence generated by simulation, and analyzing a multi-index evaluation result to generate a personalized learning path plan. According to the nervous system disease teaching and training method and system based on the image recognition technology, a first dynamic focus segmentation map is generated in sequence, automatic labeling is achieved, and the time consumed by manual labeling and case screening is greatly shortened; the dynamic change of the focus in time and space is captured in real time from multiple dimensions, so that the capability of analyzing the dynamic change of the disease is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of teaching of nervous system diseases, and more particularly to a method and system for teaching and training nervous system diseases based on image recognition technology. Background Art

[0002] With the increasing complexity and diversity of neurological diseases, medical professionals are facing higher requirements for diagnostic and treatment capabilities. Therefore, efficient teaching and clinical practice ability training for medical students are particularly important. Traditional teaching mainly relies on two-dimensional medical imaging atlases, textbook cases and pathological section observations, lacking dynamic, high-resolution interactive training methods. Innovative technical means are urgently needed to improve teaching and training effects.

[0003] In recent years, neurological disease teaching and training institutions have begun to introduce digital medical imaging databases and virtual simulation platforms, using three-dimensional reconstruction technology to assist teaching. Existing teaching and training platforms usually impart relevant knowledge by explaining theoretical knowledge, displaying static case images, and playing limited imaging materials. Students mainly rely on memorizing text descriptions and observing a small number of two-dimensional images to understand disease characteristics. Teaching resource libraries built based on medical imaging archives complete knowledge transfer through the display of image slices and text annotations of standardized cases.

[0004] However, in actual use, it still has some shortcomings. For example, the core of the teaching system is still based on static image browsing and manual annotation. Manual annotation and case screening consume a lot of time, and it is impossible to conduct real-time analysis and feedback on the dynamic evolution characteristics of neurological diseases, which makes it difficult for students to establish the three-dimensional cognitive ability required for clinical diagnosis. Traditional methods make it difficult to provide personalized teaching based on each student's learning progress and level of understanding, and are unable to accurately locate students' knowledge weaknesses and provide targeted intensive training, which seriously affects the efficiency and quality of teaching and training. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for teaching and training of neurological diseases based on image recognition technology, and solves the problems raised in the above-mentioned background technology through the following scheme.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The teaching and training methods for neurological diseases based on image recognition technology include:

[0008] S1: Acquire multimodal neurological disease images through edge computing devices, obtain the first dynamic image sequence, and obtain user interaction behavior data in neurological disease teaching and training;

[0009] S2: Obtain an image recognition model, pass the first dynamic image sequence through the image recognition model, obtain a first dynamic lesion segmentation atlas corresponding to the first dynamic image sequence, and mark the first dynamic lesion segmentation atlas;

[0010] S3: generating a second dynamic image sequence through image recognition model simulation, and simultaneously extracting a second dynamic lesion segmentation map corresponding to the second dynamic image sequence;

[0011] S4: performing a feature comparison operation on the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, wherein the feature comparison operation is used to iteratively optimize the image recognition model;

[0012] S5: Obtaining interactive feedback data of the user on the second dynamic image sequence simulated and generated by the image recognition model, where the interactive feedback data is a third dynamic lesion segmentation atlas extracted by the user;

[0013] S6: Obtain a training quality assessment model, analyze the user's learning effect through the training quality assessment model, and obtain multi-indicator evaluation results;

[0014] S7: Generate personalized learning path planning based on the multi-indicator evaluation results.

[0015] Preferably, the step S2, obtaining an image recognition model, specifically includes:

[0016] performing a lesion complexity analysis operation on the first dynamic image sequence to obtain an imaging modality corresponding to the first dynamic image sequence, where the imaging modality is a target imaging mode corresponding to the first dynamic image sequence, and the target imaging mode includes magnetic resonance imaging, computed tomography, and positron emission tomography;

[0017] In the preset system operation database, an image recognition model corresponding to the image modality is obtained. The preset system operation database is used to store the corresponding relationship between the image modality and the image recognition model.

[0018] Preferably, the S2, lesion complexity analysis operation, specifically includes:

[0019] The lesion complexity analysis operation is used to obtain a lesion complexity index corresponding to each frame image in the first dynamic image sequence. The lesion complexity index includes lesion volume complexity, shape complexity, boundary clarity, and time series complexity.

[0020] Preferably, the S2 is based on the voxel intensity value I in the lesion area of each frame image in the first dynamic image sequence. lv , voxel physical size V lv , and the binary segmentation result δ of the voxel (i, j, k) in the three-dimensional image ijk , calculate the lesion volume complexity C of each frame imagelv , specifically expressed as:

[0021]

[0022] Among them, σ(I lv ) is the standard deviation of the voxel intensity within the lesion area of each frame, μ(I lv ) represents the mean value of the voxel intensity in the lesion area of each frame, n, m, l represent the voxel identifier in the three-dimensional image, δ ijk The value of is 1 or 0, where 1 represents the lesion and 0 represents the background;

[0023] The box dimension calculation method is used to calculate the shape complexity C of the frame image based on the lesion area of each frame image in the first dynamic image sequence. s , specifically expressed as:

[0024]

[0025] in, It is represented by the exponential operation with a constant e as the base, r represents the side length of the square box used in the box dimension calculation method, N(r) represents the number of boxes with a side length of r required to cover the lesion outline of each frame image, and D1 and D0 represent normalization parameters;

[0026] Based on the lesion boundary of each frame image in the first dynamic image sequence and the point p of the lesion boundary, calculate the boundary clarity C of the frame image e , specifically expressed as:

[0027]

[0028] Among them, N(p) represents the number of pixels in the neighborhood of the boundary point p. Expressed as the image gradient at the boundary point p, Expressed as the L2 norm of the image gradient at the boundary point p, S a and S m Represented as automatic segmentation and manual annotation results, It is a quantitative calculation of the automatic segmentation and manual annotation results, with a value range between 0 and 1;

[0029] Based on the total number of frames T in the first dynamic image sequence f , calculate the time series complexity C of the first dynamic image sequence t , specifically expressed as:

[0030]

[0031] Where tf represents the frame index in the first dynamic image sequence, DTW(Ftf ,F tf+1 ) is the dynamic time warping distance of the lesion features in adjacent frames, λ is the balance coefficient, ΔH tf→tf+1 It is expressed as the rate of change of the Hausdorff distance of the lesion boundary between adjacent frames.

[0032] Preferably, the step S4, performing a feature comparison operation on the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, specifically includes:

[0033] Obtain correlation difference features between the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, wherein the correlation difference features include lesion three-dimensional volume overlap, maximum boundary offset, lesion volume change curve, gray level co-occurrence matrix entropy value and voxel intensity distribution characteristics.

[0034] Preferably, the S5 interactive feedback data includes the number of times the user adjusts the window level of the second dynamic image sequence, the multi-planar reformation operation path, the manual correction trajectory of the lesion boundary, the diagnostic annotation content of the second dynamic image sequence and the user's score of the simulated lesion evolution path.

[0035] Preferably, the multi-index evaluation results in S6 include spatial cognitive ability evaluation indicators, temporal analysis ability evaluation indicators, and diagnostic logic evaluation indicators.

[0036] Preferably, the step S6 of obtaining the multi-index evaluation results specifically includes:

[0037] Based on the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3, the three-dimensional spatial cognitive ability evaluation index AI is calculated. s1 , specifically expressed as:

[0038]

[0039] Wherein, |S3∩S2| represents the volume of the intersection of the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3, and |S3∪S2| represents the volume of the union of the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3;

[0040] Boundary point set based on the second dynamic lesion segmentation atlas and the boundary point set of the third dynamic lesion segmentation atlas Computational boundary cognitive ability evaluation index AI s2 , specifically expressed as:

[0041]

[0042] Wherein, N(ps3) represents the number of lesion boundary points of the third dynamic lesion segmentation atlas, ps3 and ps2 represent the indices of the lesion boundary points of the third dynamic lesion segmentation atlas and the lesion boundary points of the second dynamic lesion segmentation atlas, respectively, and d(ps3,ps2) represents the Euclidean distance between the lesion boundary points of the third dynamic lesion segmentation atlas and the lesion boundary points of the second dynamic lesion segmentation atlas;

[0043] AI-based three-dimensional spatial cognitive ability evaluation index s1 and boundary cognitive ability evaluation indicators AI s2 , computing spatial cognitive ability evaluation index AI s , specifically expressed as:

[0044]

[0045] To achieve the above objectives, the present invention provides the following technical solutions: a neurological disease teaching and training system based on image recognition technology, comprising a system operation database, a system central processing module, and a user information terminal, and implementing the above-mentioned neurological disease teaching and training method based on image recognition technology, including:

[0046] Dynamic image acquisition module: used to acquire multimodal neurological disease images through edge computing devices, obtain the first dynamic image sequence, and obtain user interaction behavior data in neurological disease teaching and training;

[0047] Image recognition and annotation module: used to obtain an image recognition model, pass the first dynamic image sequence through the image recognition model, obtain a first dynamic lesion segmentation map corresponding to the first dynamic image sequence, and annotate the first dynamic lesion segmentation map;

[0048] Simulation generation module: used for simulating and generating a second dynamic image sequence through an image recognition model, and synchronously extracting a second dynamic lesion segmentation atlas corresponding to the second dynamic image sequence;

[0049] Model optimization module: used to perform feature comparison operation on the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, and the feature comparison operation is used to iteratively optimize the image recognition model;

[0050] Interactive data collection module: used to obtain interactive feedback data of the user on the second dynamic image sequence simulated and generated by the image recognition model, the interactive feedback data being the third dynamic lesion segmentation atlas extracted by the user;

[0051] Training quality assessment module: used to obtain a training quality assessment model, analyze the user's learning effect through the training quality assessment model, and obtain multi-indicator evaluation results;

[0052] Learning path planning generation module: used to generate personalized learning path planning based on multi-indicator evaluation results;

[0053] The system operation database contains all data texts of the neurological disease teaching and training system, and collects the information text output by each module in real time. The system central processing module is used for the information text instructions output by each module in the central control method, and the user information terminal is an information output device for receiving the neurological disease teaching and training system.

[0054] Preferably, the training quality assessment module obtains multi-index assessment results, specifically including:

[0055] Calculate the DTW path cost difference between the disease stage node corresponding to the third dynamic lesion segmentation map annotated by the student and the standard stage corresponding to the second dynamic lesion segmentation map;

[0056] The Pearson correlation coefficient of the lesion volume change curve corresponding to the third dynamic lesion segmentation atlas and the pathological curve corresponding to the second dynamic lesion segmentation atlas was used for evaluation.

[0057] Technical effects and advantages of the present invention:

[0058] 1. The present invention matches image recognition models suitable for different image sequences and sequentially generates a first dynamic lesion segmentation atlas to achieve automatic labeling, greatly reducing the time consumed by manual labeling and case screening, improving the accuracy of segmentation and analysis, and more accurately analyzing and providing feedback on the dynamic evolution characteristics of neurological diseases in real time, greatly improving the ability to analyze the dynamic changes of diseases;

[0059] 2. This invention captures the dynamic changes of lesions in time and space from multiple dimensions in real time, providing more accurate real-time analysis and feedback on the dynamic evolution characteristics of neurological diseases, greatly improving the ability to analyze the dynamic changes of diseases;

[0060] 3. The present invention performs feature comparison between the second dynamic lesion segmentation map and the first dynamic lesion segmentation map, iteratively optimizes the image recognition model, obtains user interactive feedback data and converts it into a third dynamic lesion segmentation map, so as to more effectively help students establish the three-dimensional cognitive ability required for clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a diagram of the implementation steps of the neurological disease teaching and training method based on image recognition technology provided in an embodiment of the present application.

[0062] Figure 2 This is a flowchart of a neurological disease teaching and training system based on image recognition technology provided according to an embodiment of the present application.

[0063] Figure 3 This is an architectural diagram of a neurological disease teaching and training system based on image recognition technology provided according to an embodiment of the present application.

[0064] Explanation of the accompanying symbols: 300, processing architecture diagram of the neurological disease teaching and training system based on image recognition technology; 301, system central processing unit; 302, communication bus; 303, system operation database; 304, user information terminal. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0066] The terms used in the following examples of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification of the present application, the singular expressions "a", "an", "said", "above", "the", and "this" are intended to include plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to and includes any or all possible combinations of one or more of the listed items.

[0067] In the following, the terms "first," "second," and "third" are used for descriptive purposes only and should not be understood to imply or suggest relative importance or implicitly indicate the number of the technical features indicated. Therefore, a feature defined as "first," "second," and "third" may explicitly or implicitly include one or more of the features. In the description of the embodiments of this application, unless otherwise specified, "plurality" means two or more.

[0068] As attached Figure 1 The illustrated neurological disease teaching and training method based on image recognition technology includes obtaining a first dynamic image sequence and user interaction behavior data, annotating the first dynamic lesion segmentation map using an image recognition model, and performing a feature comparison operation with a simulated second dynamic image sequence to iteratively optimize the image recognition model; obtaining the user's interactive feedback on the simulated second dynamic image sequence, analyzing multi-index evaluation results, and generating personalized learning path planning. The specific steps are as follows:

[0069] S1: Obtain multimodal neurological disease images through edge computing devices, obtain the first dynamic image sequence, and obtain the user's interactive behavior data in neurological disease teaching and training.

[0070] Specifically, by deploying edge computing devices in medical institutions and connecting them to image acquisition terminals, the acquisition of multimodal neurological disease images can be realized. The types of multimodal neurological disease images include but are not limited to MRI, CT, PET, etc. During the acquisition process, the voxel intensity values and voxel physical sizes in the lesion area are synchronously recorded, and the data are pre-processed by the edge computing device. The pre-processing operations include but are not limited to format unification, noise removal, data compression, etc. The pre-processed data is transmitted to the central server through a secure and encrypted communication protocol; at the same time, data tracking code is embedded in the teaching and training system to obtain user interaction behavior data in real time, including but not limited to operation path, dwell time, labeling behavior, diagnostic annotations, etc.

[0071] It should be noted that the first dynamic image sequence includes multiple neurological disease images in multimodal forms, voxel intensity values and physical sizes in the lesion area corresponding to the multiple neurological disease images, clinical teaching needs and user annotation behavior of the images.

[0072] S2: Obtain an image recognition model, pass the first dynamic image sequence through the image recognition model, obtain a first dynamic lesion segmentation map corresponding to the first dynamic image sequence, and mark the first dynamic lesion segmentation map.

[0073] Specifically, the image recognition model is a pre-trained learning model. By inputting the first dynamic image sequence into the image recognition model, the image recognition model sequentially generates the corresponding first dynamic lesion segmentation map according to the first dynamic image sequence and labels the first dynamic lesion segmentation map.

[0074] In this embodiment, the framework of the image recognition model is constructed as a 3DU-Net model with a spatiotemporal joint attention mechanism, and an LSTM layer is introduced to capture the evolution characteristics of adjacent lesions in the first dynamic image sequence; when preprocessing the first dynamic image sequence, the image recognition model uses adaptive histogram equalization to enhance the contrast of each image, and combines the spatial transformation network to realize multimodal image registration.

[0075] In one possible embodiment, obtaining an image recognition model includes: performing a lesion complexity analysis operation on a first dynamic image sequence to obtain an image modality corresponding to the first dynamic image sequence, where the image modality is a target imaging mode corresponding to the first dynamic image sequence, and the target imaging mode includes magnetic resonance imaging, computed tomography imaging, and positron emission tomography; obtaining an image recognition model corresponding to the image modality in a preset system operation database, where the preset system operation database is used to save the correspondence between the image modality and the image recognition model.

[0076] It should be noted that in the diagnosis of neurological diseases, magnetic resonance imaging in the target imaging mode can clearly show the lesions of soft tissues such as brain tissue and spinal cord, and the image recognition model needs to have stronger feature extraction capabilities to capture subtle changes in soft tissues; computed tomography imaging has a fast imaging speed and is more suitable for the diagnosis of emergency patients with various lesions such as acute bleeding, fractures, and brain calcifications. The image recognition model needs to focus on the identification of high-density areas; positron emission tomography is used for tumor grading, staging, prognosis judgment, and localization of epileptic foci.

[0077] Furthermore, a lesion complexity analysis operation is performed on the first dynamic image sequence, and the lesion complexity analysis operation is used to obtain the lesion complexity index corresponding to each frame image in the first dynamic image sequence, and the lesion complexity index includes lesion volume complexity, shape complexity, boundary clarity, and time series complexity; the lesion complexity index corresponding to each frame image is integrated into a comprehensive lesion complexity score through a weighted summation formula; the initial weight coefficient of each complexity index is defined according to clinical teaching needs, and the weight coefficient is adaptively updated based on the average error rate of the student group.

[0078] Furthermore, based on the voxel intensity value I in the lesion area of each frame image in the first dynamic image sequence, lv , voxel physical size V lv , and the binary segmentation result δ of the voxel (i, j, k) in the three-dimensional image ijk , calculate the lesion volume complexity C of each frame image lv , specifically expressed as:

[0079]

[0080] Among them, σ(I lv ) is the standard deviation of the voxel intensity within the lesion area of each frame, μ(I lv ) represents the mean value of the voxel intensity in the lesion area of each frame, n, m, l represent the voxel identifier in the three-dimensional image, δ ijk The value is 1 or 0, where 1 represents the lesion and 0 represents the background. The box dimension calculation method is used to calculate the shape complexity C of the frame image based on the lesion area of each frame image in the first dynamic image sequence. s , specifically expressed as:

[0081]

[0082] in, It is represented by the exponential operation with constant e as the base, r represents the side length of the square box used in the box dimension calculation method, N(r) represents the number of boxes with side length r required to cover the lesion outline of each frame image, D1 and D0 represent normalization parameters; based on the lesion boundary of each frame image in the first dynamic image sequence and the point p of the lesion boundary, calculate the boundary clarity C of the frame image e , specifically expressed as:

[0083]

[0084] Among them, N(p) represents the number of pixels in the neighborhood of the boundary point p. Expressed as the image gradient at the boundary point p, Expressed as the L2 norm of the image gradient at the boundary point p, S a and S m Represented as automatic segmentation and manual annotation results, It is expressed as the quantitative calculation of the automatic segmentation and manual annotation results, with a value range between 0 and 1; based on the total number of frames T in the first dynamic image sequence f , calculate the time series complexity C of the first dynamic image sequence t , specifically expressed as:

[0085]

[0086] Where tf represents the frame index in the first dynamic image sequence, DTW(F tf ,F tf+1 ) is the dynamic time warping distance of the lesion features in adjacent frames, λ is the balance coefficient, ΔH tf→tf+1 It is expressed as the rate of change of the Hausdorff distance of the lesion boundary between adjacent frames.

[0087] It should be noted that the lesion volume complexity reflects the size of the lesion volume and the complexity of its distribution in three-dimensional space; the standard deviation of the voxel intensity in the lesion area represents the degree of discreteness of the voxel intensity value in the lesion area, reflecting the heterogeneity inside the lesion; the mean of the voxel intensity in the lesion area represents the average level of the voxel intensity value in the lesion area; the dynamic time warping distance of the lesion features of adjacent frames in the time series complexity is used to measure the similarity of the lesion features of adjacent frames; the change rate of the Hausdorff distance of the lesion boundary between adjacent frames is used to measure the degree of change of the lesion boundary in time.

[0088] Specifically, the first dynamic image sequence is input into the image recognition model to extract the gray matter atrophy area of the single-frame image; the lesion evolution characteristics between consecutive frames are modeled through the LSTM layer to capture the diffusion law in the time dimension; the spatial correlation weights of the feature maps of adjacent frames are calculated to suppress irrelevant background interference and enhance the continuity of the lesion boundary; the probability map output by the model is threshold segmented to generate a binary first dynamic lesion segmentation map, and the voxel coordinates are strictly aligned with the original image; after obtaining the first dynamic lesion segmentation map, the segmentation results are aligned with the preset brain area template according to the NeuroNames anatomical standard, and the anatomical structures involved in the lesion are automatically labeled.

[0089] S3: Generate a second dynamic image sequence through image recognition model simulation, and simultaneously extract a second dynamic lesion segmentation map corresponding to the second dynamic image sequence.

[0090] Specifically, the neurological disease teaching and training system is based on the user's clinical teaching needs for trainees, and simulates and generates through an image recognition model to obtain a second dynamic image sequence. The second dynamic image sequence is the simulation data of the natural progression of the disease based on the first dynamic image sequence, and the second dynamic image sequence and the first dynamic image sequence are derivative versions of the same case under different pathological characteristics; at the same time, the second dynamic image sequence is extracted in the same way as in S2, and a second dynamic lesion segmentation map corresponding to the second dynamic image sequence is obtained.

[0091] S4: performing a feature comparison operation on the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, wherein the feature comparison operation is used to iteratively optimize the image recognition model.

[0092] Specifically, the associated difference features in the first dynamic lesion segmentation map and the second dynamic lesion segmentation map are obtained. The associated difference features are key features extracted from the first dynamic lesion segmentation map and the second dynamic lesion segmentation map through feature comparison operations for evaluating the image recognition model. The associated difference features include the three-dimensional volume overlap of the lesion, the maximum boundary offset, the lesion volume change curve, the grayscale co-occurrence matrix entropy value and the voxel intensity distribution characteristics, and the image recognition model is iteratively optimized based on the associated difference features.

[0093] Furthermore, the Dice similarity coefficient is used to calculate the three-dimensional volume overlap of the lesions in the first dynamic lesion segmentation map and the second dynamic lesion segmentation map based on the lesion voxel set in the first dynamic lesion segmentation map and the second dynamic lesion segmentation map; if the three-dimensional volume overlap of the lesions is lower than the preset clinical threshold, it indicates that there is a significant deviation in the static anatomical segmentation between the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, and the weight distribution needs to be optimized first; the maximum boundary offset is quantified by the Hausdorff distance, and if the boundary deviation exceeds 3mm, the gradient sensitivity of the image recognition model is enhanced, and morphological post-processing is introduced to smooth the segmentation results; the dynamic time warping algorithm is used to align the lesion volume change curves of the first dynamic lesion segmentation map and the second dynamic lesion segmentation map to obtain the image recognition model's ability to fit the disease progression law; if the difference in volume change paths between adjacent frames exceeds 15%, it indicates that the LSTM layer is insufficient in modeling time dependence, and the forget gate parameters are adjusted to strengthen long-range dependence.

[0094] Furthermore, based on the associated difference features, a weighted Dice loss is used to impose higher penalty weights on high-complexity areas; the DTW path cost function is used to constrain the evolutionary logical consistency of the second dynamic lesion segmentation map with real cases; the KL divergence is used to force the image recognition model's predicted distribution to be closer to the clinical gold standard annotated by experts; an adaptive momentum optimizer is used to dynamically adjust the learning rate, and the hierarchical weights of the spatial attention layer, LSTM timing module, and feature pyramid network are differentially updated. If the boundary clarity decreases, the learning rate of the edge detection convolution kernel is significantly improved; when the time series complexity exceeds the standard, the parameters of the LSTM hidden state transfer matrix are updated first.

[0095] S5: Obtain interactive feedback data of the user on the second dynamic image sequence simulated and generated by the image recognition model, where the interactive feedback data is a third dynamic lesion segmentation atlas extracted by the user.

[0096] Specifically, a second dynamic image sequence is obtained by simulating the image recognition model after iterative optimization, the interactive feedback data is the user interaction operation content obtained through the user interface module, and the user's annotations on the second dynamic image sequence are converted into a third dynamic lesion segmentation map in real time.

[0097] Furthermore, the interactive feedback data includes the number of times the user adjusts the window level of the second dynamic image sequence, the multi-planar reformat operation path, the manual correction trajectory of the lesion boundary, the diagnostic annotation content of the second dynamic image sequence, and the user's score of the simulated lesion evolution path.

[0098] S6: Obtain a training quality evaluation model, analyze the user's learning effect through the training quality evaluation model, and obtain multi-indicator evaluation results.

[0099] Specifically, the training quality assessment model is a pre-built three-layer neural network architecture that integrates spatial cognition, temporal analysis and diagnostic logic evaluation. The multi-index evaluation results are obtained by passing the third dynamic lesion segmentation map through the training quality assessment model. The multi-index evaluation results include spatial cognition ability evaluation indicators, temporal analysis ability evaluation indicators, and diagnostic logic evaluation indicators.

[0100] In one possible embodiment, obtaining the multi-index evaluation result includes: calculating the three-dimensional spatial cognitive ability evaluation index AI based on the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3. s1 , specifically expressed as:

[0101]

[0102] Among them, |S3∩S2| represents the volume of the intersection of the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3, |S3∪S2| represents the volume of the union of the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3; the boundary point set based on the second dynamic lesion segmentation map and the boundary point set of the third dynamic lesion segmentation atlas Computational boundary cognitive ability evaluation index AI s2 , specifically expressed as:

[0103]

[0104] Among them, N(ps3) represents the number of lesion boundary points of the third dynamic lesion segmentation map, ps3 and ps2 represent the indexes of the lesion boundary points of the third dynamic lesion segmentation map and the lesion boundary points of the second dynamic lesion segmentation map, respectively, and d(ps3,ps2) represents the Euclidean distance between the lesion boundary points of the third dynamic lesion segmentation map and the lesion boundary points of the second dynamic lesion segmentation map; Based on the three-dimensional spatial cognitive ability evaluation index AI s1 and boundary cognitive ability evaluation indicators AI s2 , computing spatial cognitive ability evaluation index AI s , specifically expressed as:

[0105]

[0106] Among them, AI s1 It measures the degree of overlap between the third dynamic lesion segmentation map and the second dynamic lesion segmentation map in three-dimensional space. The closer the value is to 1, the more consistent the user's annotation is with the image recognition model's annotation, and the stronger the user's three-dimensional spatial cognition ability is; AI s2The average minimum distance between the lesion boundary of the third dynamic lesion segmentation map and the lesion boundary of the second dynamic lesion segmentation map is measured. The smaller the value, the higher the degree of consistency between the user's annotation and the model's annotation on the boundary, and the stronger the user's boundary recognition ability.

[0107] In a possible embodiment, obtaining the multi-index evaluation results further includes: using the dynamic time warping algorithm and the Hausdorff distance change rate to quantify the degree to which the trainee masters the law of disease evolution; calculating the DTW path cost difference between the disease stage node corresponding to the third dynamic lesion segmentation map annotated by the trainee and the standard stage corresponding to the second dynamic lesion segmentation map; if the difference rate Δ DTW >20%, the temporal logic deviation was determined; the Pearson correlation coefficient of the lesion volume change curve corresponding to the third dynamic lesion segmentation map and the pathological curve corresponding to the second dynamic lesion segmentation map was evaluated. If the Pearson correlation coefficient ρ<0.6, the trainee's grasp of the lesion volume change trend was low and the correlation with the standard curve was weak.

[0108] In a possible embodiment, obtaining the multi-index evaluation results also includes: encoding the number of window adjustments, the plane reconstruction operation path, and the manual correction trajectory of the lesion boundary in the trainee's interactive feedback data into multiple time series, calculating the deviation of the multiple time series from the preset standard diagnosis number, and if the deviation exceeds 30%, it is determined that the logic is confused; constructing a confusion matrix based on the trainee's diagnostic annotation content on the second dynamic image sequence and the historical misjudgment of the lesion type, and statistically calculating the specificity and sensitivity. If the specificity or sensitivity index is lower than 70%, it is determined that the trainee has deficiencies in the accuracy and reliability of disease diagnosis.

[0109] S7: Generate personalized learning path planning based on the multi-indicator evaluation results.

[0110] As attached Figure 2 The neurological disease teaching and training system based on image recognition technology shown includes a system operation database, a system central processing module and a user information terminal, and also includes: a dynamic image acquisition module, an image recognition and annotation module, a simulation generation module, a model optimization module, an interactive data collection module, a training quality assessment module and a learning path planning generation module.

[0111] Dynamic image acquisition module: used to obtain multimodal images of neurological diseases through edge computing devices, obtain the first dynamic image sequence, and obtain the user's interactive behavior data in neurological disease teaching and training.

[0112] Image recognition and annotation module: used to obtain an image recognition model, pass the first dynamic image sequence through the image recognition model, obtain the first dynamic lesion segmentation map corresponding to the first dynamic image sequence, and annotate the first dynamic lesion segmentation map.

[0113] Simulation generation module: used to simulate and generate a second dynamic image sequence through an image recognition model, and simultaneously extract a second dynamic lesion segmentation map corresponding to the second dynamic image sequence.

[0114] Model optimization module: used to perform feature comparison operation on the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, and the feature comparison operation is used to iteratively optimize the image recognition model.

[0115] Interactive data collection module: used to obtain interactive feedback data of the user on the second dynamic image sequence simulated and generated by the image recognition model. The interactive feedback data is the third dynamic lesion segmentation map extracted by the user.

[0116] Training quality assessment module: used to obtain a training quality assessment model, analyze the user's learning effect through the training quality assessment model, and obtain multi-indicator evaluation results.

[0117] Learning path planning generation module: used to generate personalized learning path planning based on multi-indicator evaluation results.

[0118] The system operation database contains all data texts of the neurological disease teaching and training system, and collects the information text output by each module in real time. The system central processing module is used for the information text instructions output by each module in the central control method, and the user information terminal is an information output device for receiving the neurological disease teaching and training system.

[0119] In this embodiment, the training quality assessment module obtains multi-index assessment results including: quantifying the degree to which the trainee masters the law of disease evolution using the dynamic time warping algorithm and the Hausdorff distance change rate; calculating the DTW path cost difference between the disease staging node corresponding to the third dynamic lesion segmentation map marked by the trainee and the standard staging corresponding to the second dynamic lesion segmentation map; if the difference rate Δ_DTW>20%, determining the temporal logic deviation; evaluating the Pearson correlation coefficient of the lesion volume change curve corresponding to the third dynamic lesion segmentation map and the pathological curve corresponding to the second dynamic lesion segmentation map; if the Pearson correlation coefficient ρ<0.6, the trainee's mastery of the lesion volume change trend is low and the correlation with the standard curve is weak.

[0120] In this embodiment, a processing structure of a nervous system disease teaching and training system based on image recognition technology is also disclosed. Figure 3The electronic device may include: at least one system central processor 301 , at least one communication bus 302 , a user information terminal 304 , and at least one system operation database 303 .

[0121] Among them, the system central processor 301 is the core operation and control unit of the entire neurological disease teaching and training system; it includes one or more processing cores, which connect various parts of the entire system by utilizing various interfaces and lines; by running or executing instructions, programs, code sets or instruction sets stored in the system operation database, and being able to call the data stored therein, thereby executing various functions of the neurological disease teaching and training system, including analyzing and processing neurological disease imaging data obtained based on image recognition technology, matching corresponding image recognition models according to different imaging modalities to identify lesion characteristics, and using the recognition results to generate personalized learning path planning, etc.

[0122] The communication bus 302 is used to implement connection and communication between components.

[0123] Among them, the system operation database 303 is used to store a large amount of data related to the teaching and training of neurological diseases, including AAL atlas, based on a large amount of experimental data and clinical experience, and the corresponding storage of imaging modalities and optimized and trained image recognition models suitable for the modality; when the system central processor performs various functions, it will frequently call these data from the system operation database to complete key operations such as lesion identification, model selection and adaptation, and teaching content generation, thereby realizing precise control and efficient management of neurological disease teaching and training based on image recognition technology.

[0124] The user information terminal 304 is connected to external devices such as a display screen and a camera through a standard wired interface or a wireless interface to provide an interface for users to interact with the system.

[0125] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0126] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A teaching and training method for neurological diseases based on image recognition technology, characterized in that: include: S1: Acquire multimodal neurological disease images through edge computing devices, obtain the first dynamic image sequence, and obtain user interaction behavior data in neurological disease teaching and training; S2: Obtain an image recognition model, pass the first dynamic image sequence through the image recognition model, obtain a first dynamic lesion segmentation atlas corresponding to the first dynamic image sequence, and mark the first dynamic lesion segmentation atlas; S3: generating a second dynamic image sequence through image recognition model simulation, and simultaneously extracting a second dynamic lesion segmentation map corresponding to the second dynamic image sequence; S4: performing a feature comparison operation on the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, wherein the feature comparison operation is used to iteratively optimize the image recognition model; S5: Obtaining interactive feedback data of the user on the second dynamic image sequence simulated and generated by the image recognition model, where the interactive feedback data is a third dynamic lesion segmentation atlas extracted by the user; S6: Obtain a training quality assessment model, analyze the user's learning effect through the training quality assessment model, and obtain multi-indicator evaluation results; S7: Generate personalized learning path planning based on the multi-indicator evaluation results.

2. The method for teaching and training neurological diseases based on image recognition technology according to claim 1, characterized in that: The step S2, obtaining an image recognition model, specifically includes: performing a lesion complexity analysis operation on the first dynamic image sequence to obtain an imaging modality corresponding to the first dynamic image sequence, where the imaging modality is a target imaging mode corresponding to the first dynamic image sequence, and the target imaging mode includes magnetic resonance imaging, computed tomography, and positron emission tomography; In the preset system operation database, an image recognition model corresponding to the image modality is obtained. The preset system operation database is used to store the corresponding relationship between the image modality and the image recognition model.

3. The method for teaching and training neurological diseases based on image recognition technology according to claim 2, characterized in that: The S2, lesion complexity analysis operation, specifically includes: The lesion complexity analysis operation is used to obtain a lesion complexity index corresponding to each frame image in the first dynamic image sequence. The lesion complexity index includes lesion volume complexity, shape complexity, boundary clarity, and time series complexity.

4. The method for teaching and training neurological diseases based on image recognition technology according to claim 3, characterized in that: The S2 is based on the voxel intensity value I in the lesion area of each frame image in the first dynamic image sequence. lv , voxel physical size V lv , and the binary segmentation result δ of the voxel (i, j, k) in the three-dimensional image ijk , calculate the lesion volume complexity C of each frame image lv , specifically expressed as: Among them, σ(I lv ) is the standard deviation of the voxel intensity within the lesion area of each frame, μ(I lv ) represents the mean value of the voxel intensity in the lesion area of each frame, n, m, l represent the voxel identifier in the three-dimensional image, δ ijk The value of is 1 or 0, where 1 represents the lesion and 0 represents the background; The box dimension calculation method is used to calculate the shape complexity C of the frame image based on the lesion area of each frame image in the first dynamic image sequence. s , specifically expressed as: in, It is represented by the exponential operation with a constant e as the base, r represents the side length of the square box used in the box dimension calculation method, N(r) represents the number of boxes with a side length of r required to cover the lesion outline of each frame image, and D1 and D0 represent normalization parameters; Based on the lesion boundary of each frame image in the first dynamic image sequence and the point p of the lesion boundary, calculate the boundary clarity C of the frame image e , specifically expressed as: Among them, N(p) represents the number of pixels in the neighborhood of the boundary point p. Expressed as the image gradient at the boundary point p, Expressed as the L2 norm of the image gradient at the boundary point p, S a and S m Represented as automatic segmentation and manual annotation results, It is a quantitative calculation of the automatic segmentation and manual annotation results, with a value range between 0 and 1; Based on the total number of frames T in the first dynamic image sequence f , calculate the time series complexity C of the first dynamic image sequence t , specifically expressed as: Where tf represents the frame index in the first dynamic image sequence, DTW(F tf ,F tf+1 ) is the dynamic time warping distance of the lesion features in adjacent frames, λ is the balance coefficient, ΔH tf→tf+1 It is expressed as the rate of change of the Hausdorff distance of the lesion boundary between adjacent frames.

5. The method for teaching and training neurological diseases based on image recognition technology according to claim 1, characterized in that: The step S4, performing a feature comparison operation on the first dynamic lesion segmentation atlas and the second dynamic lesion segmentation atlas, specifically includes: Obtain correlation difference features between the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, wherein the correlation difference features include lesion three-dimensional volume overlap, maximum boundary offset, lesion volume change curve, gray level co-occurrence matrix entropy value and voxel intensity distribution characteristics.

6. The method for teaching and training neurological diseases based on image recognition technology according to claim 1, characterized in that: The interactive feedback data in S5 includes the number of window level adjustments made by the user on the second dynamic image sequence, the multi-planar reformation operation path, the manual correction trajectory of the lesion boundary, the diagnostic annotation content of the second dynamic image sequence, and the user's score on the simulated lesion evolution path.

7. The method for teaching and training neurological diseases based on image recognition technology according to claim 1, characterized in that: The S6, multi-index evaluation results include spatial cognitive ability evaluation indicators, temporal analysis ability evaluation indicators, and diagnostic logic evaluation indicators.

8. The method for teaching and training neurological diseases based on image recognition technology according to claim 1, characterized in that: The step S6, obtaining the multi-index evaluation results, specifically includes: Based on the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3, the three-dimensional spatial cognitive ability evaluation index AI is calculated. s1 , specifically expressed as: Wherein, |S3∩S2| represents the volume of the intersection of the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3, and |S3∪S2| represents the volume of the union of the second dynamic lesion segmentation map S2 and the third dynamic lesion segmentation map S3; Boundary point set based on the second dynamic lesion segmentation atlas and the boundary point set of the third dynamic lesion segmentation atlas Computational boundary cognitive ability evaluation index AI s2 , specifically expressed as: Wherein, N(ps3) represents the number of lesion boundary points of the third dynamic lesion segmentation atlas, ps3 and ps2 represent the indices of the lesion boundary points of the third dynamic lesion segmentation atlas and the lesion boundary points of the second dynamic lesion segmentation atlas, respectively, and d(ps3,ps2) represents the Euclidean distance between the lesion boundary points of the third dynamic lesion segmentation atlas and the lesion boundary points of the second dynamic lesion segmentation atlas; AI-based three-dimensional spatial cognitive ability evaluation index s1 and boundary cognitive ability evaluation indicators AI s2 , computing spatial cognitive ability evaluation index AI s , specifically expressed as:

9. A neurological disease teaching and training system based on image recognition technology, comprising a system operation database, a system central processing module, and a user information terminal, wherein the neurological disease teaching and training method based on image recognition technology according to any one of claims 1 to 8 is characterized in that: Also includes: Dynamic image acquisition module: used to acquire multimodal neurological disease images through edge computing devices, obtain the first dynamic image sequence, and obtain user interaction behavior data in neurological disease teaching and training; Image recognition and annotation module: used to obtain an image recognition model, pass the first dynamic image sequence through the image recognition model, obtain a first dynamic lesion segmentation map corresponding to the first dynamic image sequence, and annotate the first dynamic lesion segmentation map; Simulation generation module: used for simulating and generating a second dynamic image sequence through an image recognition model, and synchronously extracting a second dynamic lesion segmentation atlas corresponding to the second dynamic image sequence; Model optimization module: used for performing a feature comparison operation on the first dynamic lesion segmentation map and the second dynamic lesion segmentation map, and the feature comparison operation is used for iteratively optimizing the image recognition model; Interactive data collection module: used to obtain interactive feedback data of the user on the second dynamic image sequence simulated and generated by the image recognition model, the interactive feedback data being the third dynamic lesion segmentation atlas extracted by the user; Training quality assessment module: used to obtain a training quality assessment model, analyze the user's learning effect through the training quality assessment model, and obtain multi-indicator evaluation results; Learning path planning generation module: used to generate personalized learning path planning based on multi-indicator evaluation results; The system operation database contains all data texts of the neurological disease teaching and training system, and collects the information text output by each module in real time. The system central processing module is used for the information text instructions output by each module in the central control method, and the user information terminal is an information output device for receiving the neurological disease teaching and training system.

10. The neurological disease teaching and training system based on image recognition technology according to claim 9, characterized in that: The training quality assessment module obtains multi-index assessment results, specifically including: Calculate the DTW path cost difference between the disease stage node corresponding to the third dynamic lesion segmentation map annotated by the student and the standard stage corresponding to the second dynamic lesion segmentation map; The Pearson correlation coefficient of the lesion volume change curve corresponding to the third dynamic lesion segmentation atlas and the pathological curve corresponding to the second dynamic lesion segmentation atlas was used for evaluation.

Images