Method and system for clinical injury severity assessment based on medical images

Abstract

This invention discloses a method and system for assessing the degree of clinical injury based on medical imaging. The method includes: acquiring image sequences of clinical users; performing auxiliary injury assessment on the image sequences to obtain a first injury assessment result; based on the image sequences, performing user movement recognition and medical image acquisition quality analysis to obtain medical image quality influence parameters; configuring the number of medical image acquisitions and recognition resource coefficients according to the medical image quality influence parameters; acquiring medical images of the user according to the number of medical image acquisitions to obtain a medical image set; using the recognition resource coefficients to identify the medical image set, fusing and processing to obtain a second injury assessment result; and combining the first injury assessment result to obtain a final clinical injury assessment result. By analyzing patient movement characteristics, dynamically adjusting medical image acquisition parameters, and fusing multiple assessment results, the accuracy of clinical injury assessment is improved.

Description

Technical Field

[0001] This invention relates to the field of image processing, specifically to a method and system for assessing the degree of clinical injury based on medical images. Background Technology

[0002] Clinical injury assessment is a crucial step in the medical diagnostic process, directly impacting the formulation of subsequent treatment plans and prognostic evaluations. Currently, medical imaging technologies such as X-rays, CT scans, and MRI have become the primary means of clinical injury assessment. These technologies can visually display damage to the patient's internal tissue structures, providing doctors with objective diagnostic evidence. However, in actual clinical application, patients often involuntarily move during examinations due to pain, discomfort, or anxiety. These movements can lead to blurring, distortion, and artifacts in the acquired medical images, affecting the accuracy of clinical injury assessment and potentially causing misdiagnosis or missed diagnosis, thus impacting subsequent treatment outcomes. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for assessing the degree of clinical injury based on medical images, in order to solve the technical problem mentioned in the background art above, where the quality of medical images is affected by the patient's movement due to pain, resulting in low accuracy of clinical injury assessment.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0005] In a first aspect, the present invention provides a method for assessing the degree of clinical injury based on medical images, comprising: acquiring image sequences of clinical users; performing auxiliary injury assessment on the image sequences to obtain a first injury assessment result; performing user movement recognition and medical image acquisition quality analysis based on the image sequences to obtain medical image quality influence parameters; configuring the number of medical image acquisitions and recognition resource coefficients based on the medical image quality influence parameters; acquiring medical images of users according to the number of medical image acquisitions to obtain a medical image set; using the recognition resource coefficients to identify the medical image set, fusing and processing to obtain a second injury assessment result; and combining the first injury assessment result with the second injury assessment result to obtain a clinical injury assessment result, which is then displayed as auxiliary assessment information.

[0006] Optionally, acquiring image sequences of clinical users and performing auxiliary damage assessment on the image sequences to obtain a first damage assessment result includes: acquiring images of clinical users before medical image acquisition to obtain image sequences; inputting the image sequences into a first damage evaluator trained based on a convolutional neural network, recognizing and outputting a first damage assessment result, wherein the first damage assessment result includes a damage assessment level.

[0007] Optionally, the training steps of the first injury evaluator include: collecting a set of sample image sequences based on historical clinical injury assessment data, and collecting the actual injury assessment level corresponding to each sample image sequence, labeling it as a set of sample injury assessment levels; constructing the network architecture of the first injury evaluator using a convolutional neural network; and using the set of sample image sequences and the set of sample injury assessment levels as training data and test data to perform supervised training and testing on the first injury evaluator until the test is qualified.

[0008] Optionally, based on the image sequence, user movement recognition is performed, and medical image acquisition quality analysis is conducted to obtain medical image quality impact parameters. This includes: collecting a set of sample image sequences based on historical clinical injury assessment data, and collecting the user's movement distance during medical image acquisition under different sample image sequence sets, labeling them as sample movement distance sets; constructing a network architecture for a movement recognizer; using the sample image sequence set and sample movement distance set as training and testing data, supervising and testing the movement recognizer until the test is qualified; inputting the image sequence into the movement recognizer and outputting the user movement distance; and mapping and classifying the medical image quality impact parameters based on the user movement distance.

[0009] Optionally, based on the user's movement distance, mapping and classification are used to obtain medical image quality impact parameters, including: collecting a set of sample user movement distances based on historical data of clinical injury assessment, and collecting the distortion amplitude of medical image acquisition under different sample user movement distances, labeling it as sample medical image quality impact parameters, and obtaining a set of sample medical image quality impact parameters; constructing a mapping relationship between the set of sample user movement distances and the set of sample medical image quality impact parameters to obtain an image quality classification table; inputting the user movement distance into the image quality classification table, and mapping and classifying to obtain medical image quality impact parameters.

[0010] Optionally, based on the medical image quality impact parameter, configure the number of medical image acquisitions and the recognition resource coefficient, and perform medical image acquisitions on the user according to the number of medical image acquisitions to obtain a medical image set, including: obtaining the maximum medical image quality impact parameter in the medical image acquisition; calculating the ratio of the medical image quality impact parameter to the maximum medical image quality impact parameter as an image quality coefficient; obtaining the maximum number of medical image acquisitions, multiplying the image quality coefficient by the maximum number of medical image acquisitions and rounding to obtain the number of medical image acquisitions; using the image quality coefficient as the recognition resource coefficient; and performing medical image acquisitions on the user according to the number of medical image acquisitions to obtain a medical image set.

[0011] Optionally, the medical image set is identified using the identification resource coefficient, and a second damage assessment result is obtained through fusion processing. The second damage assessment result is then combined with the first damage assessment result to obtain a clinical damage assessment result. This includes: constructing a medical image damage recognizer, wherein the medical image damage recognizer includes U medical image damage recognition branches, where U is a positive integer; obtaining the number V of medical image damage recognition branches for medical image damage recognition by multiplying the identification resource coefficient by U and rounding down; randomly selecting V medical image damage recognition branches, inputting multiple medical images from the medical image set respectively, identifying and outputting multiple damage assessment result sets, calculating the average to obtain a second damage assessment result; and calculating a clinical damage assessment result based on the first and second damage assessment results.

[0012] Optionally, the steps of constructing a medical image injury recognizer include: constructing U medical image injury recognition branches respectively; collecting a sample medical image set and a sample injury assessment level set based on clinical injury assessment data over a historical period; performing U random partitions with replacement on the sample medical image set and the sample injury assessment level set to obtain U sets of training data and U sets of test data; and using the U sets of training data and U sets of test data to perform supervised training and testing on the U medical image injury recognition branches respectively until all branches pass the test, thereby obtaining the medical image injury recognizer.

[0013] Secondly, the present invention provides a clinical injury severity assessment system based on medical images, comprising: a first result acquisition module, used to acquire image sequences of clinical users, perform auxiliary injury severity assessment on the image sequences, and obtain a first injury severity assessment result; an influence parameter acquisition module, used to perform user movement recognition and medical image acquisition quality image analysis based on the image sequences, and obtain medical image quality influence parameters; a medical image acquisition module, used to configure the number of medical images acquired and the recognition resource coefficient based on the medical image quality influence parameters, and acquire medical images of users according to the number of medical images acquired, to obtain a medical image set; and an assessment result acquisition module, used to identify the medical image set using the recognition resource coefficient, perform fusion processing to obtain a second injury severity assessment result, and combine the first injury severity assessment result to obtain a clinical injury severity assessment result, which is displayed as auxiliary assessment information.

[0014] The beneficial effects of this invention are:

[0015] By acquiring image sequences from clinical users, an auxiliary assessment of the degree of injury is performed on the image sequences to obtain a preliminary assessment result of the degree of injury, thus enabling preliminary analysis of the image sequences and preliminary assessment of the degree of injury of clinical users. Based on the image sequences, user movement is identified, and an analysis of the impact on medical image acquisition quality is conducted to obtain medical image quality impact parameters. By analyzing the movement of clinical users and its impact on medical image quality, these parameters are quantified into specific medical image quality impact parameters, providing a basis for subsequent adjustments to the image acquisition strategy. Based on the medical image quality impact parameters, the number of medical image acquisitions and recognition resource coefficients are configured. Medical images are then acquired from users according to the number of medical image acquisitions, resulting in a medical image set, thus achieving root... According to the dynamic adjustment strategy based on the movement of clinical users, when the impact on quality is greater, more medical images will be acquired and a higher recognition resource coefficient will be configured. By increasing the sample size and computing resources, the adverse effects of movement are offset, thereby improving the accuracy of the overall assessment. The recognition resource coefficient is used to identify the medical image set, and the fusion processing is used to obtain the second damage degree assessment result. Combined with the first damage degree assessment result, the clinical damage degree assessment result is obtained and displayed as auxiliary assessment information. By integrating and learning the recognition of multiple medical images, the second damage degree assessment result is obtained and weighted and fused with the first damage degree assessment result to form a comprehensive clinical damage degree assessment result, providing an objective basis for doctors' diagnostic decisions.

[0016] Through the above technical solutions, this application not only solves the adverse effects of clinical user movement on medical image quality, but also uses the movement characteristics of clinical users themselves as an evaluation reference, and improves the accuracy and reliability of the evaluation through multi-sample collection and multi-result fusion. Attached Figure Description

[0017] Figure 1 A schematic flowchart illustrating the clinical injury assessment method based on medical imaging provided by this invention;

[0018] Figure 2 This is a schematic diagram of the structure of the clinical injury assessment system based on medical imaging provided by the present invention.

[0019] In the attached diagram, the components represented by each number are as follows:

[0020] First result acquisition module 11, influencing parameter acquisition module 12, medical image acquisition module 13, evaluation result acquisition module 14.

[0021] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1: This application provides a method for assessing the degree of clinical injury based on medical imaging, such as... Figure 1 As shown, the method includes:

[0024] S100: Collects image sequences from clinical users, performs auxiliary assessment of the degree of damage on the image sequences, and obtains the first degree of damage assessment result.

[0025] Specifically, when a clinical user arrives at a medical institution, a visible light camera positioned near medical imaging equipment (such as an X-ray machine or CT scanner) continuously captures images of the injured area, forming an image sequence. During this process, the clinical user may move due to pain, and this movement is captured in the image sequence. By analyzing and processing these continuously captured image sequences, a preliminary assessment of the degree of injury is obtained.

[0026] The initial damage assessment results obtained are used as a preliminary assessment. Although their accuracy is relatively low, they can quickly provide a preliminary judgment on the degree of damage to clinical users, and provide a reference for subsequent medical image (such as X-ray, CT, etc.) acquisition and analysis.

[0027] S200: Based on the image sequence, perform user movement recognition and medical image acquisition quality analysis to obtain medical image quality impact parameters.

[0028] Specifically, clinical users often involuntarily move during medical imaging examinations due to pain or discomfort. These movements can significantly impact the quality of subsequent formal medical image acquisition, leading to blurred, distorted, or artifact-prone images, thereby affecting the diagnostic value of medical images and the accuracy of damage assessment.

[0029] Therefore, the user movement captured in the image sequence is analyzed, and the user's movement distance is quantified. Based on this movement distance, the potential impact on medical image quality during formal medical image acquisition is predicted, and this impact is quantified into a medical image quality impact parameter.

[0030] Medical image quality impact parameters are crucial for determining subsequent medical image acquisition methods and are directly related to the accuracy of the final damage assessment. Obtaining these parameters allows for the effective identification and quantification of the potential impact of user movement on medical image quality, laying the foundation for subsequent adaptive medical image acquisition.

[0031] S300: Based on the medical image quality impact parameters, configure the number of medical images to be acquired and the recognition resource coefficient, and acquire medical images from the user according to the number of medical images to obtain a medical image set.

[0032] Specifically, once the medical image quality impact parameter is obtained, the acquisition strategy is adaptively adjusted based on this parameter. A larger medical image quality impact parameter indicates a more significant impact of user movement on image quality; in this case, the number of medical images acquired should be increased, and the allocated recognition resource coefficient should be raised accordingly. Conversely, a smaller medical image quality impact parameter indicates less user movement and a limited impact on image quality; therefore, fewer medical images should be acquired, and a lower recognition resource coefficient should be allocated.

[0033] Subsequently, based on the configured number of medical images to be acquired, medical images are acquired from clinical users, resulting in a medical image set containing multiple medical images. These medical images may include various types such as X-rays, CT scans, and MRI, which together constitute a comprehensive description of the user's injury status, providing a direct basis for subsequent assessment of the degree of injury.

[0034] S400: It uses the identification resource coefficient to identify the medical image set, and performs fusion processing to obtain the second damage degree assessment result. Combined with the first damage degree assessment result, it processes and obtains the clinical damage degree assessment result, which is displayed as auxiliary assessment information.

[0035] Specifically, firstly, a medical image injury detector with multiple recognition branches is constructed. The number of recognition branches to be activated is determined based on the recognition resource coefficient. Then, multiple medical images from the medical image set are input into these activated recognition branches, resulting in multiple injury severity assessments. The mean of these results is calculated to obtain a second injury severity assessment result, which is based on standard medical images and has relatively high accuracy. Next, the first and second injury severity assessment results are weighted and fused, with the second injury severity assessment result having a larger weight. This fusion strategy considers both the user's subjective pain response and movement (reflected in the first injury severity assessment result) and fully utilizes the diagnostic information from objective medical images (reflected in the second injury severity assessment result), thus obtaining a more comprehensive and accurate clinical injury severity assessment result. Subsequently, the clinical injury severity assessment result is presented to medical personnel as supplementary assessment information, providing a scientific and objective reference for clinical diagnosis and treatment decisions. This supplementary assessment method effectively reduces the workload of medical personnel and improves the accuracy and consistency of injury assessment, especially when the quality of medical images is affected by pain or other factors in clinical users, demonstrating significant advantages.

[0036] Furthermore, image sequences from clinical users are acquired, and the extent of damage is assessed using these image sequences to obtain a primary damage assessment result, including:

[0037] S110: Collect images from clinical users before they undergo medical image acquisition to obtain image sequences;

[0038] S120: Input the image sequence into the first damage evaluator trained on a convolutional neural network, and identify the output to obtain the first damage level assessment result, wherein the first damage level assessment result includes the damage assessment level.

[0039] As an optional implementation, the clinical user is first continuously photographed to obtain an image sequence. Specifically, a visible light camera positioned near medical imaging equipment (such as an X-ray machine or CT scanner) is used to continuously photograph the clinical user, forming an image sequence containing multiple frames. The image sequence records the clinical user's injury status and also records the clinical user's movements during the examination in a sequential manner, such as slight tremors or positional shifts caused by pain.

[0040] The obtained image sequence is then input into a pre-trained first damage estimator. This first damage estimator is a model trained on a convolutional neural network, capable of extracting and analyzing features from the image sequence. After processing, it outputs a first damage assessment result, which is represented by a specific damage assessment level, such as a numerical level of 1-5, or a classification level such as mild, moderate, or severe.

[0041] Furthermore, the training steps for the first damage evaluator include:

[0042] S121: Based on historical clinical injury assessment data, collect a set of sample image sequences and collect the actual injury assessment level corresponding to each sample image sequence, which is then labeled as a set of sample injury assessment levels.

[0043] S122: A convolutional neural network is used to construct the network architecture of the first damage evaluator;

[0044] S123: Using a set of sample image sequences and a set of sample damage assessment levels as training and testing data, supervise the training and testing of the first damage evaluator until it passes the test.

[0045] In a preferred implementation, to train the first injury evaluator, a dataset required for training and testing is first constructed based on historical clinical injury assessment data. Specifically, based on historical clinical injury assessment data, a large number of sample image sequences from clinical users are collected. These image sequences record the injury status and movement of different clinical users during medical image acquisition. Simultaneously, the actual injury assessment level corresponding to each sample image sequence is collected. These levels are assessment results given by professional medical personnel based on complete medical examination results. By matching and labeling the image sequences with the actual injury levels, a set of sample image sequences and a set of sample injury assessment levels are obtained, providing the basic data for training the first injury evaluator.

[0046] Then, a convolutional neural network (CNN) technique is used to construct the network architecture of the first damage estimator. CNNs are suitable for processing image data and can effectively extract spatial and temporal features from images. The network architecture may include multiple convolutional layers, pooling layers, and fully connected layers to extract features related to the degree of damage from the image sequence and ultimately output the damage assessment level. Subsequently, the constructed first damage estimator is trained and tested under supervised supervision using the obtained set of sample image sequences and sample damage assessment levels. During training, the sample data is divided into a training set and a test set. The parameters of the first damage estimator are optimized using the training set, and the performance of the first damage estimator is evaluated using the test set. The training process continues until the first damage estimator reaches the preset performance indicators on the test set, i.e., the test is passed, and the trained first damage estimator is obtained. The criteria for passing the test may include evaluation indicators such as accuracy, precision, and recall reaching predetermined thresholds.

[0047] Through supervised training based on a large amount of historical clinical data, the first damage assessor can learn the complex relationship between image sequences and damage severity, providing reliable technical support for subsequent damage severity assessment.

[0048] Furthermore, based on the image sequence, user movement is identified, and medical image acquisition quality is analyzed to obtain parameters affecting medical image quality, including:

[0049] S210: Based on historical data of clinical injury assessment, collect a set of sample image sequences, and collect the user's movement distance during medical image acquisition under different sample image sequence sets, and label them as a set of sample movement distances;

[0050] S220: Building the network architecture for mobile identifiers;

[0051] S230: Use a set of sample image sequences and a set of sample movement distances as training and testing data to supervise the training and testing of the motion recognizer until it passes the test;

[0052] S240: Input the image sequence into the motion detector and output the user's movement distance;

[0053] S250: Based on the user's movement distance, map and classify to obtain medical image quality impact parameters.

[0054] In a preferred embodiment, firstly, a dataset for training the motion recognizer is constructed based on historical data from clinical injury assessment. Specifically, a large number of sample image sequences are collected from the historical clinical injury assessment data. These sample image sequences record the state changes of different clinical users during medical image acquisition. Simultaneously, user movement distance data corresponding to these sample image sequences is collected, i.e., the displacement of the clinical user during medical image acquisition. This movement distance data can be obtained by measuring with professional motion capture equipment or recorded by position sensors in medical imaging equipment. These movement distance data are labeled as sample movement distance sets, corresponding one-to-one with the corresponding image sequence sets, forming a complete training dataset. Then, the network architecture of the motion recognizer is constructed. This motion recognizer can employ deep learning models, such as convolutional neural networks, recurrent neural networks, or combinations thereof, to extract user movement-related features from the image sequences and ultimately output an estimated value of the movement distance. The design of the network architecture needs to consider the spatiotemporal characteristics of the image sequences to effectively capture changes in user movement over time.

[0055] Subsequently, the constructed motion recognizer was subjected to supervised training and testing using the obtained sample image sequence set and sample movement distance set. During training, the sample image sequence set and sample movement distance set were divided into training set and test set. The motion recognizer's parameters were optimized using the training set, and its performance was evaluated using the test set. The training process continued until the motion recognizer achieved a preset performance indicator on the test set, such as the average error of movement distance prediction being below a certain threshold, indicating that the test was successful. Next, the image sequence of the current clinical user was input into the trained motion recognizer. Through processing and calculation, the user's movement distance was obtained as the output. This movement distance is a prediction of the potential displacement of the current clinical user during medical image acquisition, providing basic data for subsequent medical image quality impact analysis. Afterward, based on the obtained user movement distance, medical image quality impact parameters were obtained through a preset mapping relationship. This parameter quantifies the potential impact of user movement on medical image acquisition quality and can be expressed as an impact coefficient or impact level, providing an important basis for subsequent medical image acquisition strategy configuration.

[0056] Through the above steps, the user's movement can be accurately identified and transformed into quantifiable parameters affecting medical image quality, laying the foundation for mobile-adaptive medical image acquisition and damage assessment.

[0057] Furthermore, based on the user's movement distance, mapping and classification are used to obtain medical image quality impact parameters, including:

[0058] S251: Based on historical data of clinical injury assessment, collect the set of sample user movement distances, and collect the distortion amplitude of medical image acquisition under different sample user movement distances, label it as the sample medical image quality influencing parameters, and obtain the set of sample medical image quality influencing parameters.

[0059] S252: Construct a mapping relationship between the set of sample user movement distances and the set of parameters affecting the quality of sample medical images to obtain an image quality classification table;

[0060] S253: Input the user's movement distance into the image quality classification table, and obtain the medical image quality impact parameters by mapping the classification.

[0061] In a preferred embodiment, firstly, based on historical data from clinical injury assessments, a correlation is established between movement distance and the impact on image quality. Specifically, movement distance data from a large number of sample users is collected to form a sample user movement distance set. Simultaneously, distortion amplitude data for medical image acquisition at these different movement distances is collected. Here, distortion amplitude refers to the quantified value of the degree of quality defects such as blurring, stretching, and ghosting in medical images caused by user movement. These distortion amplitude data are labeled as sample medical image quality impact parameters, forming a sample medical image quality impact parameter set. The data in these two sets correspond one-to-one, reflecting the intrinsic relationship between movement distance and the impact on image quality.

[0062] Next, a mapping relationship is constructed between the set of sample user movement distances and the set of medical image quality impact parameters. This mapping relationship can be established through statistical analysis methods, such as regression analysis and piecewise function fitting, or through machine learning methods, such as decision trees and support vector machines. By establishing this mapping relationship, a structured image quality classification table is obtained, which defines the medical image quality impact parameter values ​​corresponding to different movement distance intervals. Subsequently, the movement distance of the current clinical user is input into the established image quality classification table. By looking up the table, the mapping classification yields the corresponding medical image quality impact parameters for that user, reflecting the potential impact of user movement on medical image quality and providing a precise basis for subsequent medical image acquisition strategy configuration.

[0063] By using historical data and precise mapping, the user's movement distance can be transformed into a parameter affecting medical image quality, realizing the correspondence between movement and image quality impact, and providing a reliable decision-making basis for adaptive medical image acquisition strategies.

[0064] Furthermore, based on the medical image quality impact parameters, the number of medical images acquired and the recognition resource coefficient are configured. Medical images are then acquired from the user according to the acquired number of images, resulting in a medical image set, including:

[0065] S310: Obtain the maximum medical image quality impact parameter during medical image acquisition;

[0066] S320: Calculate the ratio of the medical image quality impact parameter to the maximum medical image quality impact parameter, and use it as the image quality coefficient;

[0067] S330: Obtain the maximum number of medical images acquired by multiplying the image quality coefficient by the maximum number of medical images acquired and rounding down to get the number of medical images acquired.

[0068] S340: Use the image quality coefficient as the identification resource coefficient;

[0069] S350: Collects medical images from the user according to the number of medical images to obtain a medical image set.

[0070] In a preferred embodiment, firstly, the maximum medical image quality impact parameter that may occur during medical image acquisition is obtained. This parameter represents the maximum impact on medical image quality under the most extreme conditions (such as when the user moves most violently), and is a pre-set benchmark value. This value can be obtained through analysis of a large amount of clinical data, or it can be set according to the characteristics of medical imaging equipment and clinical practice experience. Then, the ratio of the current clinical user's medical image quality impact parameter to the obtained maximum medical image quality impact parameter is calculated, and this ratio is used as the image quality coefficient. The image quality coefficient is a value between 0 and 1; the larger the value, the more significant the impact of user movement on medical image quality, and the more compensation measures are needed.

[0071] Subsequently, the preset maximum number of medical images to be acquired is obtained. This maximum number of medical images is the maximum number required to acquire under extreme conditions (such as when the user is moving violently). Then, the obtained image quality coefficient is multiplied by the maximum number of medical images to be acquired, and the result is rounded to obtain the actual number of medical images to be acquired. In this way, the number of medical images acquired can be dynamically adjusted according to the degree to which user movement affects image quality. The more significant the movement, the higher the image quality coefficient, and the more images are calculated, thus compensating for the decrease in image quality caused by movement by increasing the number of acquisitions. This adaptive adjustment mechanism ensures that when the user moves little, only a few medical images need to be acquired to obtain reliable evaluation results, while when the user moves a lot, the number of medical images acquired is increased, and the accuracy of the final evaluation is improved through multiple acquisitions and subsequent fusion processing.

[0072] Next, the obtained image quality coefficient is directly used as the recognition resource coefficient, ensuring that the allocation of recognition resources is proportional to the impact of medical image quality. That is, when significant user movement leads to poor medical image quality, more recognition resources are allocated to compensate. Then, medical images are acquired according to a predetermined number, resulting in a medical image set containing multiple images. These images can be multiple acquisitions of the same site, or acquisitions from different angles or with different parameter settings, collectively forming a multi-faceted description of the user's injury status.

[0073] Through the above steps, the system achieves adaptive configuration of medical image acquisition strategies, which can dynamically adjust the number of medical images acquired and resource allocation according to the user's movement, thereby optimizing the system's resource utilization efficiency while ensuring the accuracy of the assessment.

[0074] Furthermore, resource identification coefficients are used to identify the medical image set, and the images are fused to obtain a second injury severity assessment result. This second result is then combined with the first injury severity assessment result to obtain a clinical injury severity assessment result, including:

[0075] S410: Construct a medical image injury recognizer, wherein the medical image injury recognizer includes U medical image injury recognition branches, where U is a positive integer;

[0076] S420: The number of medical image damage recognition branches V for medical image damage recognition is obtained by multiplying the recognition resource coefficient by U and rounding down.

[0077] S430: Randomly select V medical image damage recognition branches, input multiple medical images in the medical image set respectively, recognize and output multiple damage degree assessment result sets, calculate the mean to obtain the second damage degree assessment result;

[0078] S440: The clinical injury assessment result is calculated based on the results of the first injury assessment and the second injury assessment.

[0079] In a preferred embodiment, firstly, a medical image damage recognizer is constructed, comprising U medical image damage recognition branches, where U is a positive integer. This multi-branch design employs ensemble learning, allowing each branch to be based on different network structures or different training data, thereby enabling multi-angle and multi-mode recognition and analysis of medical images, improving the robustness and accuracy of the final recognition results. Then, based on the obtained recognition resource coefficients, the number of medical image damage recognition branches to be activated is dynamically determined. Specifically, the recognition resource coefficients are multiplied by the total number of recognition branches U and rounded to obtain the actual number of branches V needed for recognition. This adaptive allocation method can rationally allocate computational resources according to the degree of impact of medical image quality, improving system efficiency while ensuring recognition accuracy.

[0080] Subsequently, V branches are randomly selected from U medical image injury recognition branches, and multiple medical images from the medical image set are input into these selected branches for recognition processing. The recognition results of each branch for each medical image are combined to form an injury severity assessment result set, and then the mean of these results is calculated to obtain a second injury severity assessment result. By integrating recognition and result fusion, the recognition bias of a single model can be reduced, and the reliability of the recognition results can be improved, especially when the quality of medical images is affected by user movement. Then, based on the obtained first injury severity assessment result and the obtained second injury severity assessment result, the final clinical injury severity assessment result is calculated through weighted fusion or other fusion algorithms. In this fusion process, different weights are assigned according to the reliability of the two assessment results. Generally, the second injury severity assessment result based on medical images receives a higher weight, while the first injury severity assessment result based on image sequences, as supplementary information, receives a relatively lower weight.

[0081] Through the above steps, intelligent recognition of medical images and fusion processing of multi-source assessment results are achieved. Even when the quality of medical images is affected by the patient's movement due to pain, accurate and reliable assessment results of the degree of injury can still be provided, providing a basis for clinical diagnosis and treatment decisions.

[0082] Furthermore, the steps for constructing a medical image injury detector include:

[0083] S411: Construct U medical image damage recognition branches respectively;

[0084] S412: Collect a set of sample medical images and a set of sample injury assessment levels based on clinical injury assessment data over a historical period.

[0085] S413: Perform U random partitions with replacement on the sample medical image set and the sample damage assessment level set to obtain U sets of training data and U sets of test data;

[0086] S414: Use U sets of training data and U sets of test data to conduct supervised training and testing on U medical image injury recognition branches until all branches pass the test, and obtain the medical image injury recognizer.

[0087] In a preferred embodiment, when constructing the medical image injury recognizer, firstly, U medical image injury recognition branches are constructed. These branches can employ the same or different network architectures, such as deep learning models like convolutional neural networks or residual networks, and each branch possesses complete feature extraction and injury recognition functions. By constructing multiple medical image injury recognition branches, medical images can be analyzed from different angles or using different methods, providing a foundation for ensemble learning. Then, based on clinical injury assessment data from a historical period, a sample medical image set and a sample injury assessment level set are collected. The sample medical image set contains a large number of labeled medical image samples, covering different types of injuries and varying degrees of image quality; the sample injury assessment level set contains the professional injury assessment results for the corresponding medical images, provided by medical experts based on a complete clinical examination.

[0088] Then, the obtained set of sample medical images and sample damage assessment levels are randomly partitioned with replacement U times to obtain U sets of training data and U sets of test data. This random partitioning method with replacement can generate multiple data subsets with certain differences, which helps to improve the generalization ability and robustness of the final ensemble model. Each set of training and test data contains sample medical images and their corresponding damage assessment levels, used for training and performance evaluation of a single recognition branch. Subsequently, the U sets of training and test data are used to supervise the training and testing of the constructed U medical image damage recognition branches. Each recognition branch uses its corresponding training data for parameter optimization and its corresponding test data for performance evaluation. The training process continues until all recognition branches reach the preset performance indicators, i.e., pass the test. The pass criteria may include multiple indicators such as accuracy, precision, and recall reaching predetermined thresholds. Finally, these U trained medical image damage recognition branches are combined to form a complete medical image damage recognizer.

[0089] By employing an ensemble learning strategy, a medical image injury recognizer composed of multiple recognition branches was constructed. Each branch was trained on a different subset of data, exhibiting a degree of diversity and complementarity. This design effectively improves the stability and accuracy of the recognition results, demonstrating significant advantages, especially in complex situations where medical image quality is affected by user movement.

[0090] Example 2, as Figure 2 As shown, based on the same inventive concept as the clinical injury severity assessment method based on medical imaging provided in Embodiment 1, this embodiment of the invention also provides a clinical injury severity assessment system based on medical imaging, including:

[0091] The first result acquisition module 11 is used to acquire image sequences from clinical users, perform auxiliary assessment of the degree of damage on the image sequences, and obtain a first degree of damage assessment result.

[0092] The influencing parameter acquisition module 12 is used to perform user movement recognition and medical image acquisition quality analysis based on the image sequence to obtain medical image quality influencing parameters.

[0093] The medical image acquisition module 13 is used to configure the number of medical images to be acquired and the recognition resource coefficient according to the medical image quality influence parameters, and to acquire medical images of the user according to the number of medical images to be acquired, thereby obtaining a medical image set.

[0094] The assessment result acquisition module 14 is used to identify the medical image set using the identification resource coefficient, perform fusion processing to obtain a second damage degree assessment result, and combine the first damage degree assessment result to obtain a clinical damage degree assessment result, which is then displayed as auxiliary assessment information.

[0095] Furthermore, the first result acquisition module 11 includes the following execution steps:

[0096] Collect images from clinical users before they undergo medical image acquisition to obtain image sequences;

[0097] The image sequence is input into a first damage evaluator trained on a convolutional neural network, and the recognition output obtains a first damage level assessment result, wherein the first damage level assessment result includes a damage assessment level.

[0098] Furthermore, the first result acquisition module 11 also includes the following execution steps:

[0099] Based on historical clinical injury assessment data, a set of sample image sequences was collected, and the actual injury assessment level corresponding to each sample image sequence was collected and labeled as a sample injury assessment level set.

[0100] A convolutional neural network is used to construct the network architecture of the first damage evaluator;

[0101] The first damage evaluator is trained and tested using the set of sample image sequences and the set of sample damage assessment levels as training and testing data, respectively, until it passes the test.

[0102] Furthermore, the parameter acquisition module 12 includes the following execution steps:

[0103] Based on historical data of clinical injury assessment, a set of sample image sequences was collected, and the movement distance of the user during medical image acquisition was collected under different sample image sequence sets and labeled as the sample movement distance set.

[0104] Constructing the network architecture for a mobile identifier;

[0105] The motion recognizer is trained and tested under supervised supervision using the set of sample image sequences and the set of sample movement distances as training and testing data, respectively, until the test is passed.

[0106] The image sequence is input into the motion detector, and the user's movement distance is output.

[0107] Based on the user's movement distance, medical image quality impact parameters are obtained through mapping and classification.

[0108] Furthermore, the parameter acquisition module 12 also includes the following execution steps:

[0109] Based on historical data of clinical injury assessment, a set of sample user movement distances was collected, and the distortion amplitude of medical image acquisition under different sample user movement distances was collected and labeled as the sample medical image quality influencing parameters, thus obtaining a set of sample medical image quality influencing parameters.

[0110] Construct a mapping relationship between the set of sample user movement distances and the set of sample medical image quality influencing parameters to obtain an image quality classification table;

[0111] The user's movement distance is input into the image quality classification table, and the medical image quality impact parameters are obtained by mapping and classification.

[0112] Furthermore, the medical image acquisition module 13 includes the following execution steps:

[0113] Obtain the maximum medical image quality impact parameter during medical image acquisition;

[0114] The ratio of the medical image quality impact parameter to the maximum medical image quality impact parameter is calculated and used as the image quality coefficient.

[0115] To obtain the maximum number of medical images to be acquired, the image quality coefficient is multiplied by the maximum number of medical images to be acquired and then rounded down to obtain the number of medical images to be acquired.

[0116] The image quality coefficient is used as the identification resource coefficient;

[0117] Based on the stated number of medical images to be acquired, medical images are acquired from the user to obtain a medical image set.

[0118] Furthermore, the evaluation result acquisition module 14 includes the following execution steps:

[0119] Construct a medical image injury recognizer, wherein the medical image injury recognizer includes U medical image injury recognition branches, where U is a positive integer;

[0120] The number of medical image damage recognition branches V is obtained by multiplying the recognition resource coefficient by U and rounding it down.

[0121] V medical image injury recognition branches are randomly selected. Multiple medical images in the medical image set are input respectively. Multiple injury degree assessment result sets are obtained by recognition output. The average value is calculated to obtain the second injury degree assessment result.

[0122] Based on the first injury assessment result and the second injury assessment result, the clinical injury assessment result is calculated.

[0123] Furthermore, the evaluation result acquisition module 14 also includes the following execution steps:

[0124] Construct U medical image injury recognition branches respectively;

[0125] Based on clinical injury assessment data from historical periods, a set of sample medical images and a set of sample injury assessment levels were collected.

[0126] The sample medical image set and the sample damage assessment level set are randomly divided with replacement U times to obtain U sets of training data and U sets of test data;

[0127] The U training data and U test data are used to supervise and test the U medical image injury recognition branches until all of them pass the test, thus obtaining the medical image injury recognizer.

[0128] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for assessing the degree of clinical injury based on medical imaging, characterized in that, The method includes: Image sequences of clinical users are acquired, and the degree of damage is assessed by means of the image sequences to obtain a first degree of damage assessment result. The image sequences are obtained by continuously photographing the damaged parts of the clinical users by a visible light camera set near the medical imaging equipment before medical imaging of the clinical users. Based on the image sequence, user movement is identified, and medical image acquisition quality is analyzed to obtain medical image quality influencing parameters, including: Based on historical data of clinical injury assessment, a set of sample image sequences was collected, and the movement distance of the user during medical image acquisition was collected under different sample image sequence sets and labeled as the sample movement distance set. Constructing the network architecture for a mobile identifier; The motion recognizer is trained and tested under supervised supervision using the set of sample image sequences and the set of sample movement distances as training and testing data, respectively, until the test is passed. The image sequence is input into the motion detector, and the user's movement distance is output. Based on the user's movement distance, medical image quality impact parameters are obtained through mapping and classification, including: Based on historical data of clinical injury assessment, a set of sample user movement distances was collected, and the distortion amplitude of medical image acquisition under different sample user movement distances was collected and labeled as the sample medical image quality influencing parameters, thus obtaining a set of sample medical image quality influencing parameters. Construct a mapping relationship between the set of sample user movement distances and the set of sample medical image quality influencing parameters to obtain an image quality classification table; Input the user's movement distance into the image quality classification table, and obtain medical image quality impact parameters by mapping and classification. Based on the medical image quality impact parameters, configure the number of medical images to be acquired and the recognition resource coefficient, and acquire medical images from the user according to the number of medical images to be acquired to obtain a medical image set; The medical image set is identified using the aforementioned identification resource coefficients, and a second damage assessment result is obtained through fusion processing. Combined with the first damage assessment result, a clinical damage assessment result is obtained and displayed as auxiliary assessment information.

2. The clinical injury severity assessment method based on medical imaging according to claim 1, characterized in that, Image sequences from clinical users are acquired, and the extent of damage is assessed using these image sequences to obtain a first damage assessment result, including: Collect images from clinical users before they undergo medical image acquisition to obtain image sequences; The image sequence is input into a first damage evaluator trained on a convolutional neural network, and the recognition output obtains a first damage level assessment result, wherein the first damage level assessment result includes a damage assessment level.

3. The clinical injury severity assessment method based on medical imaging according to claim 2, characterized in that, The training steps for the first damage evaluator include: Based on historical clinical injury assessment data, a set of sample image sequences was collected, and the actual injury assessment level corresponding to each sample image sequence was collected and labeled as a sample injury assessment level set. A convolutional neural network is used to construct the network architecture of the first damage evaluator; The first damage evaluator is trained and tested using the set of sample image sequences and the set of sample damage assessment levels as training and testing data, respectively, until it passes the test.

4. The method for assessing the degree of clinical injury based on medical imaging according to claim 1, characterized in that, Based on the medical image quality impact parameters, the number of medical image acquisitions and the recognition resource coefficient are configured. Medical images are then acquired from the user according to the specified number of acquisitions, resulting in a medical image set, including: Obtain the maximum medical image quality impact parameter during medical image acquisition; The ratio of the medical image quality impact parameter to the maximum medical image quality impact parameter is calculated and used as the image quality coefficient. To obtain the maximum number of medical images to be acquired, the image quality coefficient is multiplied by the maximum number of medical images to be acquired and then rounded down to obtain the number of medical images to be acquired. The image quality coefficient is used as the identification resource coefficient; Based on the stated number of medical images to be acquired, medical images are acquired from the user to obtain a medical image set.

5. The method for assessing the degree of clinical injury based on medical imaging according to claim 1, characterized in that, Using the aforementioned identification resource coefficients, the medical image set is identified, and a second damage assessment result is obtained through fusion processing. Combined with the first damage assessment result, a clinical damage assessment result is obtained, including: Construct a medical image injury recognizer, wherein the medical image injury recognizer includes U medical image injury recognition branches, where U is a positive integer; The number of medical image damage recognition branches V is obtained by multiplying the recognition resource coefficient by U and rounding it down. V medical image injury recognition branches are randomly selected. Multiple medical images in the medical image set are input respectively. Multiple injury degree assessment result sets are obtained by recognition output. The average value is calculated to obtain the second injury degree assessment result. Based on the first injury assessment result and the second injury assessment result, the clinical injury assessment result is calculated.

6. The clinical injury severity assessment method based on medical imaging according to claim 5, characterized in that, The steps involved in building a medical image injury detector include: Construct U medical image injury recognition branches respectively; Based on clinical injury assessment data from historical periods, a set of sample medical images and a set of sample injury assessment levels were collected. The sample medical image set and the sample damage assessment level set are randomly divided with replacement U times to obtain U sets of training data and U sets of test data; The U training data and U test data are used to supervise and test the U medical image injury recognition branches until all of them pass the test, thus obtaining the medical image injury recognizer.

7. A clinical injury severity assessment system based on medical imaging, characterized in that, The system for performing the method as described in any one of claims 1-6 includes: The first result acquisition module is used to acquire image sequences of clinical users, perform auxiliary assessment of the degree of damage on the image sequences, and obtain a first degree of damage assessment result. The image sequences are obtained by continuously photographing the damaged parts of the clinical users through a visible light camera set near the medical imaging equipment before medical imaging acquisition of the clinical users. The influencing parameter acquisition module is used to perform user movement recognition and medical image acquisition quality analysis based on the image sequence to obtain medical image quality influencing parameters, including: Based on historical data of clinical injury assessment, a set of sample image sequences was collected, and the movement distance of the user during medical image acquisition was collected under different sample image sequence sets and labeled as the sample movement distance set. Constructing the network architecture for a mobile identifier; The motion recognizer is trained and tested under supervised supervision using the set of sample image sequences and the set of sample movement distances as training and testing data, respectively, until the test is passed. The image sequence is input into the motion detector, and the user's movement distance is output. Based on the user's movement distance, medical image quality impact parameters are obtained through mapping and classification, including: Based on historical data of clinical injury assessment, a set of sample user movement distances was collected, and the distortion amplitude of medical image acquisition under different sample user movement distances was collected and labeled as the sample medical image quality influencing parameters, thus obtaining a set of sample medical image quality influencing parameters. Construct a mapping relationship between the set of sample user movement distances and the set of sample medical image quality influencing parameters to obtain an image quality classification table; Input the user's movement distance into the image quality classification table, and obtain medical image quality impact parameters by mapping and classification. The medical image acquisition module is used to configure the number of medical images to be acquired and the recognition resource coefficient according to the medical image quality influence parameters, and to acquire medical images from the user according to the number of medical images to obtain a medical image set. The assessment result acquisition module is used to identify the medical image set using the identification resource coefficient, perform fusion processing to obtain a second damage degree assessment result, and combine it with the first damage degree assessment result to obtain a clinical damage degree assessment result, which is then displayed as auxiliary assessment information.

Images