Severe patient acute lung injury assessment system and method based on machine learning

Abstract

The invention discloses a critical patient acute lung injury assessment system and method based on machine learning, and relates to the field of acute lung injury assessment, and the method comprises the steps: obtaining chest multi-modal image data of a critical patient, and carrying out the standardization processing to generate a standard data set; dividing the lung area into structural blocks and extracting structural indexes, and selecting compensation missing data through cross-modal space mapping or candidate blocks; identifying a modal conflict area and a structure mutation area to construct a scoring key area; after disturbance screening effective areas are applied, a mean value or weighted fusion strategy is selected according to a consistency index; calculating a lung injury score in combination with the dynamic weight and performing iterative optimization; according to the method, the recognition accuracy of the acute lung injury key area can be improved, and the accurate quantitative evaluation of the lung injury degree is realized.

Description

Technical Field

[0001] The present invention relates to the field of acute lung injury assessment, and specifically to a machine learning-based acute lung injury assessment system and method for critically ill patients. Background Art

[0002] Acute lung injury (ALI) is a common respiratory complication in critically ill patients, characterized by rapid onset, severity, and rapid progression. In severe cases, it can develop into acute respiratory distress syndrome (ARDS), significantly increasing the patient's mortality rate. Early identification of ALI and objective assessment of injury severity are crucial for guiding clinical intervention and improving prognosis. Imaging, as a non-invasive method for assessing lung structure, plays a key role in the diagnosis and assessment of severe ALI.

[0003] However, traditional imaging assessment methods rely primarily on manual film reading or empirical rules for lesion identification and scoring, which are subject to high subjectivity, low efficiency, and inconsistent quantitative standards. These methods struggle to meet the demands for high-precision, structured assessments of lung injury in complex clinical scenarios. In recent years, machine learning methods have been widely used in medical image processing. Leveraging the models' strengths in multidimensional feature extraction, complex pattern recognition, and data fusion, they can automatically identify and assess structural indicators in multimodal data environments. However, these methods still face the following deficiencies: A lack of detailed delineation of lung imaging structural regions and modeling of local structural mutations prevents accurate identification of scoring-sensitive areas; and In actual multimodal imaging data, structural information is often lost due to imaging occlusion, resolution mismatch, and other factors, for which existing solutions fail to provide effective compensation mechanisms.

[0004] In view of the above problems, the existing technology is in urgent need of improvement. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a machine learning-based acute lung injury assessment system and method for critically ill patients.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention discloses a method for assessing acute lung injury in critically ill patients based on machine learning, comprising the following steps:

[0008] Acquiring multimodal chest imaging data of critically ill patients, and performing standardization processing on the multimodal chest imaging data to generate a multimodal standard data set;

[0009] The lung area is divided into several structural blocks, and structural indicators are extracted from the structural blocks based on the standard data set. If there is a lack of structural indicators in any modality, spatial mapping compensation is preferentially performed on the same structural block in the other modality. If compensation is still not possible, the best candidate block is selected from the structural blocks in the current modality, and its structural indicators are extracted for compensation.

[0010] Identifying modal conflict regions and structural mutation regions in the structural block based on the structural indicators, and merging the modal conflict regions and structural mutation regions to construct a scoring key region;

[0011] Applying a preset disturbance to the structural indicators of the key scoring area and calculating the disturbance impact value, determining whether the disturbance impact value is greater than a preset impact threshold, and if so, retaining the key scoring area; otherwise, eliminating the key scoring area;

[0012] A consistency index is calculated based on the structural indicators of the same scoring key area in all modalities, and it is determined whether the consistency index is greater than a preset consistency threshold. If so, the structural indicators of the same scoring key area are averaged and fused. Otherwise, a weighted fusion is performed based on the credibility of each modality.

[0013] Calculating a lung injury score for the critically ill patient based on the fused structural indicators and the dynamic weights of the structural indicators in key scoring areas;

[0014] Determine whether the key scoring area meets the preset integrity condition. If so, output the lung injury score; otherwise, reconstruct the key scoring area and calculate the lung injury score until the preset number of iterations is reached. If the integrity condition is still not met, output the average lung injury score of all iterations.

[0015] In a second aspect, the present invention discloses a system for assessing acute lung injury in critically ill patients based on machine learning, comprising:

[0016] A data acquisition module is used to acquire multimodal chest imaging data of critically ill patients and perform standardization processing on the multimodal chest imaging data to generate a multimodal standard data set;

[0017] a data processing module, configured to divide the lung region into a plurality of structural blocks, and extract structural indicators within the structural blocks based on the standard data set; if structural indicators are missing in any modality, spatial mapping compensation is preferentially performed on the same structural blocks in other modalities; if compensation is still not possible, the optimal candidate block is selected from the structural blocks in the current modality, and its structural indicators are extracted for compensation;

[0018] A region extraction module is used to identify the modal conflict region and the structural mutation region in the structural block based on the structural indicators, and merge the modal conflict region and the structural mutation region to construct a scoring key region;

[0019] A region optimization module is used to apply a preset disturbance to the structural indicators of the key scoring region and calculate the disturbance impact value, and determine whether the disturbance impact value is greater than a preset impact threshold. If so, the key scoring region is retained; otherwise, the key scoring region is eliminated;

[0020] The feature fusion module is used to calculate the consistency index based on the structural indicators of the same scoring key area in all modalities, and determine whether the consistency index is greater than a preset consistency threshold. If so, the structural indicators of the same scoring key area are averaged and fused. Otherwise, weighted fusion is performed based on the credibility of each modality.

[0021] an injury assessment module, configured to calculate a lung injury score for the critically ill patient based on the fused structural indicators and the dynamic weights of the structural indicators in key scoring areas;

[0022] The evaluation and optimization module is used to determine whether the key scoring area meets the preset integrity conditions. If so, the lung injury score is output; otherwise, the key scoring area is reconstructed and the lung injury score is calculated until the preset number of iterations is reached. If the integrity conditions are still not met, the average lung injury score of all iterations is output.

[0023] The beneficial effects of the present invention are:

[0024] 1. By introducing a modal consistency index and a credibility weighting mechanism, data consistency is evaluated based on the correlation coefficient of the structural indicators between the modes. If the consistency is high, mean fusion is performed; if the consistency is poor, weighted fusion is performed based on the modal credibility. This effectively avoids the evaluation errors introduced by traditional simple averaging methods in modal conflict situations, thereby improving the robustness and accuracy of the evaluation results.

[0025] 2. When multimodal images lack structural indicators, a cross-modal spatial mapping compensation strategy is prioritized. If cross-modal compensation fails, intra-modal compensation is performed by constructing a block feature prediction model and selecting the optimal candidate block based on feature similarity. This dual compensation mechanism effectively reduces the impact of missing data on lung injury assessment results while ensuring the rationality of structural block features.

[0026] 3. We integrate modal conflict detection and structural mutation identification to construct key scoring regions, and further introduce perturbation sensitivity analysis. By applying positive and negative perturbations to structural indicators, we assess the impact of regions on the score, retaining only regions that are sensitive to score changes. This mechanism improves the assessment model's ability to identify lesion-sensitive regions, reduces interference from redundant blocks, and enhances the accuracy and interpretability of the assessment.

[0027] 4. By combining the dual constraints of area proportion and anatomical key points and adopting a dynamic expansion mechanism, the scoring area range can be adaptively adjusted to ensure the comprehensiveness and accuracy of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 This is an overall block diagram of a method according to an embodiment of the present invention;

[0030] Figure 2 Flowchart for extracting structural indicators in the method of embodiment 1 of the present invention;

[0031] Figure 3 1 is a flow chart of the lung injury assessment method in Example 1 of the present invention;

[0032] Figure 4 This is an overall block diagram of the system according to the second embodiment of the present invention. DETAILED DESCRIPTION

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0034] Application Overview: In existing technologies, the assessment of acute lung injury mainly relies on manual interpretation of single-modality images, making it difficult to effectively integrate multimodal imaging data. Traditional methods use a simple averaging strategy for multimodal fusion, ignoring spatial differences and credibility differences between modalities, resulting in unstable assessment results. When data is missing in a certain modality, existing technologies lack an effective compensation mechanism, affecting the integrity of the assessment. In addition, the weight distribution in the scoring model is mostly fixed, which cannot adapt to different lesion distributions and changes in modal quality, resulting in inaccurate identification of sensitive areas.

[0035] To address these issues, researchers skilled in the art have identified the need for a spatial mapping compensation mechanism for multimodal data to address the lack of structural indicators. By analyzing the relationship between intermodal feature conflicts and structural mutations, they proposed constructing key scoring regions as the core of the evaluation. To address the shortcomings of traditional fusion strategies, they designed a region screening mechanism based on perturbation sensitivity, combined with dynamic weight adjustment to improve evaluation stability. They also iteratively optimized the coverage of key regions to ensure the integrity of the scoring results.

[0036] Example 1:

[0037] like Figure 1-3 As shown, a method for assessing acute lung injury in critically ill patients based on machine learning comprises the following steps: obtaining chest multimodal imaging data of critically ill patients, and standardizing the chest multimodal imaging data to generate a multimodal standard data set; dividing the lung area into several structural blocks, and extracting structural indicators in the structural blocks based on the standard data set; if there is a lack of structural indicators in any modality, spatial mapping compensation is performed on the same structural blocks in other modalities first; if compensation is still not possible, selecting the best candidate block in the structural blocks in the current modality, and extracting its structural indicators for compensation; identifying the modal conflict area and the structural mutation area in the structural block based on the structural indicators, merging the modal conflict area and the structural mutation area to construct a scoring key area; applying a preset perturbation to the structural indicators of the scoring key area and calculating the perturbation effect. The method comprises the following steps: first, determining whether the disturbance impact value is greater than a preset impact threshold, and then retaining the key scoring area if so, and removing the key scoring area if so; calculating a consistency index based on the structural indicators of the same key scoring area in all modalities, and determining whether the consistency index is greater than a preset consistency threshold, and then performing mean fusion on the structural indicators of the same key scoring area, and then performing weighted fusion based on the credibility of each modality; calculating the lung injury score of the critically ill patient based on the fused structural indicators and the dynamic weights of the structural indicators in the key scoring area; determining whether the key scoring area meets the preset integrity condition, and then outputting the lung injury score if so; and then reconstructing the key scoring area and calculating the lung injury score until the preset number of iterations is reached. If the integrity condition is still not met, then outputting the average value of the lung injury score in all iterations.

[0038] Normalization refers to the unification of the resolution, grayscale values, and spatial location of images from different modalities. This can be achieved through image resizing, histogram equalization, and alignment of anatomical landmarks to ensure comparability across multiple modalities. Structural segmentation divides the lungs into grid cells based on bronchial tree bifurcations. This can be achieved using rectangular grid segmentation, providing a spatial basis for local feature analysis. Spatial mapping compensation replaces missing data by finding structural indicators at the same location in other modalities. This can be achieved through coordinate transformation and feature matching to address the incompleteness of single-modal data. Optimal candidate block selection searches for neighboring regions within the current modality with the highest feature similarity to the missing block. This can be achieved using cosine similarity calculation and prediction models to ensure the rationality of data compensation. Key scoring region construction combines blocks with modality conflicts and structural abrupt changes to form a core evaluation region. This can be achieved through variance analysis and Euclidean distance thresholding to focus on sensitive areas. Dynamic weight calculation adjusts indicator weights based on spatial variation and consistency deviation. This can be achieved through normalization and weighted summation to enhance the adaptability of the scoring model.

[0039] Specifically, the method first standardizes multimodal data through image resolution adjustment, grayscale normalization, and anatomical landmark alignment. After dividing the lungs into grid blocks aligned with bronchial bifurcations, texture, density, and edge features are extracted from each block. When block data is missing in one modality, alternative data is preferentially obtained from the corresponding location in other modalities. If cross-modal compensation is not possible, the optimal candidate block is selected within the current modality through spatial distance screening and feature prediction. Intermodal variance is calculated to identify conflicting regions, and key assessment regions are determined by combining feature mutation detection of adjacent blocks within a single modality. After applying positive and negative perturbations to key regions, valid regions are selected based on the change in score. During the multimodal data fusion stage, a mean or credibility weighting strategy is adopted based on the correlation coefficient. Finally, the weights are dynamically adjusted based on the spatial variation and consistency deviation, and the lung injury score is output after iteratively optimizing the coverage of key regions.

[0040] Traditional methods use fixed weights to fuse multimodal data, but are unable to handle spatial feature differences and modal conflicts. This solution significantly improves assessment stability by constructing key scoring regions to focus on sensitive areas, combining perturbation analysis and dynamic weight adjustment. Existing technologies use simple interpolation to handle missing data, while this solution compensates for this through cross-modal mapping and candidate block prediction, improving data integrity. Traditional scoring models rely on fixed anatomical partitions, while this solution iteratively optimizes key region coverage to enhance adaptability to changes in lesion distribution.

[0041] Through the above technical solutions, this application effectively integrates the spatial and feature information of multimodal imaging data to solve the problems of data loss compensation and modality conflict. Through perturbation impact analysis and dynamic weight adjustment, the accuracy and robustness of lung injury scoring are improved. The iterative optimization mechanism ensures that all predefined anatomical landmarks are covered in key areas to avoid assessment omissions. This method provides reliable technical support for the early identification and quantitative assessment of acute lung injury in critically ill patients.

[0042] The present application further proposes that chest multimodal imaging data includes CT scan images, X-ray images and ultrasound images; standardization processing includes image resolution unification, grayscale value normalization and spatial alignment; the specific process of standardization processing includes: unifying the resolution of different modal images to a preset benchmark resolution; achieving grayscale value normalization through grayscale histogram stretching; and performing spatial alignment using the sternal angle and tracheal bifurcation as anatomical reference points.

[0043] Among them, chest multimodal imaging data refers to a collection of lung imaging data acquired through different imaging devices. Specifically, it can be acquired using CT scanners, X-ray machines, and ultrasound equipment. The limitations of a single imaging method are compensated by covering imaging modalities with different tissue characteristics. Image resolution unification refers to adjusting the pixel sizes of different modalities to the same physical scale. Specifically, interpolation algorithms or downsampling methods can be used to adjust the image size to eliminate the interference of resolution differences between modalities on subsequent structural index extraction. Grayscale value normalization refers to eliminating the grayscale response differences between different imaging devices. Specifically, histogram matching or linear stretching methods can be used to map grayscale values ​​to a unified interval to achieve comparability across modal data. Spatial alignment refers to mapping multimodal images to the same anatomical coordinate system. Specifically, a rigid registration algorithm based on the sternal angle and tracheal bifurcation point can be used to eliminate the spatial offset caused by differences in patient posture by aligning anatomical landmarks.

[0044] Specifically, after acquiring CT, X-ray, and ultrasound images, the resolution of the CT image is first adjusted to a preset baseline resolution (e.g., 0.5 mm / pixel). The X-ray image is adjusted to the same resolution using a bicubic interpolation algorithm, while the ultrasound image is spatially calibrated based on the physical parameters of the scanning probe. Subsequently, the Hounsfield units of the CT image are linearly transformed, the grayscale range of the X-ray image is mapped to the 0-255 range using a histogram stretch, and the grayscale values ​​of the ultrasound image are normalized using a contrast enhancement algorithm. Finally, in the spatial registration stage, the three-dimensional coordinates of the sternal angle and the tracheal bifurcation point in the CT image are detected and used as reference points. An affine transformation is then used to align the two-dimensional projective coordinate system of the X-ray image and the three-dimensional volumetric coordinate system of the ultrasound image to the anatomical space of the CT image, forming a multimodal dataset with a unified geometric reference.

[0045] Traditional multimodal image processing often involves simple resizing or grayscale adjustment without establishing anatomically based spatial registration, making it difficult to accurately align lesion locations between modalities. This solution addresses the spatial misalignment issue of multimodal images by implementing rigid registration using anatomical landmarks. Furthermore, through a combination of resolution unification and grayscale normalization, it eliminates the impact of device characteristics on structural index extraction.

[0046] Through the above technical solution, this application achieves the geometric and grayscale spatial consistency of multimodal imaging data, provides standardized input for subsequent structural block division and indicator extraction, avoids feature matching errors caused by inconsistent resolution, grayscale distribution and spatial coordinates between modalities, significantly improves the accuracy of cross-modal data fusion, and supports the reliability of subsequent lung injury scoring.

[0047] The present application further proposes that the division of structural blocks is based on the anatomical structural features of the lungs, and the lung area is divided into multiple rectangular grid units, and the segmentation boundaries of the rectangular grid units are aligned with the bifurcation points of the bronchial tree; the structural indicators include texture eigenvalues, density distribution values ​​and edge clarity; the texture eigenvalues ​​are obtained by calculating the contrast of the pixel grayscale values ​​in the structural block; the density distribution value is obtained by statistically calculating the mean of the CT values ​​in the structural block; the edge clarity is quantified by the Canny operator response intensity; the spatial mapping compensation includes spatially aligning the missing structural blocks in the modality with the known structural blocks in other modalities, and using the latter's structural indicators as replacement data; the process of selecting the optimal candidate block is: calculating the physical space distance between all structural blocks in the current modality image and the structural blocks with missing structural indicators; setting A distance threshold is used to screen out a set of candidate blocks whose physical space distance is less than the distance threshold; if the candidate block set is empty, the distance threshold range is expanded and re-screened until at least one candidate block is found; the physical space distance is calculated based on the image coordinate system; the spatial position coordinates and structural indicators of all complete structural blocks in the current modality image are extracted; with the spatial position coordinates as input and the vector of structural indicators as output, a block feature prediction model is established through supervised learning; the spatial position coordinates of the structural blocks with missing structural indicators are input into the trained block feature prediction model, and the predicted structural indicators are output; the feature similarity between the candidate blocks and the predicted structural indicators is calculated; the structural block with the largest feature similarity is selected as the optimal candidate block; the feature similarity is obtained by calculating the cosine similarity between the vectors of structural indicators.

[0048] Among them, the alignment of the segmentation boundary of the rectangular grid unit with the bifurcation point of the bronchial tree refers to setting the boundary of the grid division according to the anatomical landmark points of the bronchial branches. Specifically, image registration technology can be used to align the grid segmentation line with the coordinates of the bronchial bifurcation point, thereby ensuring that the structural block division matches the physiological structure and avoiding cross-anatomical region segmentation. Among them, the contrast calculation of the texture feature value refers to analyzing the distribution differences of pixel grayscale values ​​in local areas. Specifically, the gray level co-occurrence matrix algorithm can be used to extract texture contrast parameters to characterize the homogeneity or degree of lesions of lung tissue. Among them, the position alignment of spatial mapping compensation refers to using the spatial registration relationship of multimodal images to map the coordinates of the missing region to the corresponding position of other modalities. Specifically, the affine transformation algorithm can be used to achieve cross-modal coordinate conversion, thereby using the complete data of other modalities to fill the missing information of the current modality. Among them, the supervised learning of the block feature prediction model refers to using the spatial coordinates of the complete structural block as input features and the structural indicator vector as output label. Specifically, the random forest or neural network algorithm can be used to train the model to achieve spatial position-based structural indicator prediction and provide alternative data for the missing blocks.

[0049] Specifically, based on the standardized multimodal imaging data, the lungs are first divided into rectangular grid units according to the anatomical landmarks of the bifurcation points of the bronchial tree, and each unit corresponds to an independent structural block. For each structural block, three types of indicators, texture feature values, density distribution values, and edge clarity, are extracted from different modal images. When there are missing indicators for a specific structural block of a certain modality, the spatial coordinates of the block are mapped to the corresponding positions of other modalities through a spatial mapping compensation mechanism to obtain alternative indicator data. If other modalities cannot provide compensation data, the adjacent candidate blocks are screened by physical space distance, and the prediction indicators are generated by combining the pre-trained block feature prediction model. Finally, the candidate block closest to the prediction result is selected as the compensation source based on cosine similarity. This process effectively solves the problem of missing data caused by imaging conditions, while ensuring the reliability of the compensation data through anatomical alignment and spatial correlation constraints.

[0050] Traditional methods typically use interpolation or simple neighbor filling when data is missing, without considering the spatial mapping relationship between multiple modalities and anatomical structural constraints, which can easily introduce compensation errors. This solution achieves multi-level compensation for missing data through cross-modal spatial alignment compensation and machine learning-based block feature prediction, avoiding the impact of missing single-modality data on the overall assessment. In addition, combined with the anatomical alignment of bronchial bifurcation points, the division of structural blocks is more in line with actual clinical needs.

[0051] Through the above technical solutions, this application can effectively address the problem of local information loss in multimodal imaging data. By combining spatial mapping compensation with machine learning prediction, the accuracy of missing data compensation is improved, avoiding assessment bias caused by incomplete data. At the same time, the anatomical block division and structural index extraction method enhances the sensitivity of local lesion detection capabilities, providing more reliable basic data for subsequent lung injury scoring.

[0052] This application further proposes a method for identifying modal conflict areas and structural mutation areas in structural blocks based on structural indicators. The process of identifying modal conflict areas includes: calculating the variance of structural indicators of the same structural block between different modes, and if the variance is greater than the conflict judgment threshold, it is determined to be a modal conflict area; the process of identifying structural mutation areas includes: calculating the Euclidean distance difference between the structural indicators between the structural block and the adjacent structural blocks within a single mode, and if the difference exceeds a preset mutation threshold, it is marked as a structural mutation area.

[0053] Among them, the modal conflict area refers to the area that is marked because the structural indicators between different modal images are significantly different. Specifically, variance calculation can be used to quantify the difference between modalities. The variance reflects the degree of fluctuation of the structural indicators of the same structural block under different modalities. The conflict judgment threshold can be an empirical value set in advance based on the distribution of multimodal data, for example, it is set to 0.5, which is used to screen out modal inconsistency areas that need attention. Among them, the structural mutation area refers to the area where the structural indicators between adjacent structural blocks in the same modality image mutate. Specifically, Euclidean distance difference calculation can be used to capture local abnormalities. The Euclidean distance difference is achieved by comparing the difference in the structural indicator vectors of the current block and the adjacent blocks. The preset mutation threshold can be set based on the statistical results of normal lung image data, for example, it is set to twice the average difference in structural indicators between adjacent blocks, which is used to identify abnormal structural changes.

[0054] Specifically, in the process of identifying modal conflict areas, for the same structural block, its structural indicators under CT, X-ray and ultrasound modalities, such as texture feature values, density distribution values, etc., are extracted, and the variance of these indicators between different modalities is calculated respectively. If the variance exceeds the conflict judgment threshold, it means that there is significant inconsistency in the multimodal data in this area and it needs to be marked as a modal conflict area. In the process of identifying structural mutation areas, for the image data of a certain modality, each structural block is traversed and the Euclidean distance difference between its structural indicator vector and that of the adjacent blocks is calculated. If the difference exceeds the preset mutation threshold, it is determined that there is a local structural mutation in this area. By merging the modal conflict area and the structural mutation area, a key scoring area covering multimodal inconsistency and local abnormalities can be constructed, providing a data basis for subsequent refined scoring.

[0055] Traditional methods often rely solely on a single modality or fail to clearly distinguish between inter-modal differences and local structural anomalies, leading to inaccurate scoring region selection. This method, by quantifying inter-modal variance and local mutations within a single modality, can dynamically identify conflicting regions in multimodal data and abnormal regions within the image. The combined scoring key regions reflect both cross-modal inconsistencies and capture local structural changes within a single modality, effectively improving the coverage and accuracy of scoring-sensitive regions.

[0056] Through the above technical solution, this application solves the problem of biased scoring region selection caused by the ineffective quantification of inter-modality differences and the inaccurate identification of local structural abnormalities in the existing technology. By jointly determining the modal conflict region and the structural mutation region, key regions with significant impact on lung injury scores can be dynamically screened, avoiding scoring errors introduced by inter-modality conflicts or unaddressed local mutations, thereby improving the reliability and sensitivity of lung injury assessment results.

[0057] The present application further proposes to calculate the average value of each type of structural indicator in all key scoring areas, and to weight the average values ​​of different types of structural indicators according to preset weights to obtain a benchmark scoring value; to apply a preset fixed proportion of positive adjustment and negative adjustment to each type of structural indicator in the key scoring area, and to calculate the benchmark scoring value based on the adjusted structural indicators to obtain positive scoring values ​​and negative scoring values ​​respectively; the calculation formula for the disturbance impact value is: the disturbance impact value is equal to the sum of the absolute value of the difference between the positive scoring value and the benchmark scoring value plus the absolute value of the difference between the negative scoring value and the benchmark scoring value divided by two.

[0058] Among them, the benchmark score value refers to the reference score value formed by weighting the average values ​​of different types of structural indicators. Specifically, this can be achieved by setting the importance weights of different structural indicators, such as setting the weight of the texture feature value to 0.4, the density distribution value to 0.3, and the edge clarity to 0.3. This benchmark value is used to quantify the basic state of the key scoring area when it is undisturbed. Among them, the preset fixed-ratio positive adjustment and negative adjustment refer to the artificially controllable numerical increase or decrease of the structural indicator, which can be achieved specifically by percentage adjustment, such as setting it to a 5% fluctuation. By applying bidirectional perturbations, the sensitivity of the key scoring area to changes in indicators can be tested. Among them, the perturbation impact value refers to quantifying the stability of the key area by calculating the average deviation of the score values ​​before and after the disturbance, which can be achieved specifically by the absolute value averaging method. This indicator is used to screen out areas that have a significant impact on the overall scoring results, thereby optimizing the selection of key scoring areas.

[0059] Specifically, after completing the preliminary construction of the key scoring areas, its effectiveness needs to be verified through perturbation testing. First, various structural indicators are extracted from all key scoring areas and their average values ​​are calculated, and then weighted and summed according to the preset weights to form a benchmark score value. Subsequently, each type of structural indicator is adjusted positively and negatively. For example, the density distribution value is increased by 5% and the score is recalculated to obtain a positive score value; the density distribution value is then reduced by 5% to obtain a negative score value. Finally, the perturbation impact value is determined by calculating the average deviation of the two perturbation score values ​​from the baseline value. If the value is higher than the preset threshold, it indicates that the area is sensitive to indicator changes and needs to be retained in the key scoring area; otherwise, it is eliminated. Through this dynamic screening mechanism, redundant areas with low contribution to the overall score can be eliminated, thereby improving the reliability of the evaluation results.

[0060] Traditional methods often rely on fixed rules or static thresholds when constructing scoring regions, failing to consider the potential impact of indicator perturbations on scoring results. For example, some existing solutions directly use manually defined anatomical partitions as scoring regions, ignoring differences in lesion distribution and indicator sensitivity across patients. However, this solution, through bidirectional perturbation testing and impact value calculation, achieves dynamic optimization and screening of key scoring regions, effectively avoiding scoring bias caused by inappropriate region selection.

[0061] Through the above technical solution, this application solves the problem of lack of dynamic adaptability in the selection of scoring regions in existing lung injury assessment methods. By quantifying the impact of disturbances on the scoring results, it can automatically identify and retain key areas sensitive to lung injury scores, while eliminating areas with poor stability. This mechanism significantly improves the anti-interference ability and accuracy of the scoring results, especially when the patient's lung lesions are complexly distributed or there is local noise in the imaging data, which can effectively reduce the assessment error.

[0062] This application further proposes a method of combining all modes in pairs and calculating the correlation coefficient when calculating the consistency index. Specifically, it includes combining all modes in pairs, calculating the correlation coefficient between the structural indicators for each combination, taking the average of the correlation coefficients to obtain the consistency index. The correlation coefficient is determined by the ratio of the covariance and the standard deviation product.

[0063] Among them, the combination of all modalities in pairs refers to pairing different imaging modalities such as CT, X-ray, and ultrasound in pairs, such as CT and X-ray combination, CT and ultrasound combination, and X-ray and ultrasound combination. This step is used to fully cover all possible correlations between multimodal data to ensure the comprehensiveness of consistency assessment. The correlation coefficient refers to a numerical value that reflects the degree of linear correlation between the structural indicators of two modalities in the same structural block. Specifically, it can be calculated using the Pearson correlation coefficient formula, which is obtained by dividing the covariance by the product of the standard deviations of the structural indicators of each modality. This value can quantify the consistency of the changing trends of the structural indicators between modalities. The ratio of the covariance to the product of the standard deviation refers to calculating the covariance of the structural indicators of the two modalities, calculating their respective standard deviations and multiplying them, and dividing the covariance by the product value. It is used to eliminate the influence of dimensional differences on the correlation assessment and realize standardized correlation measurement. The consistency index refers to a comprehensive consistency measurement value obtained by averaging the correlation coefficients of all pairwise modal combinations. For example, the three correlation coefficients of the three pairwise combinations of three modalities are added together and averaged. This index can comprehensively reflect the overall consistency level of multimodal data and provide a quantitative basis for the selection of subsequent fusion strategies.

[0064] Specifically, after obtaining the structural indicators of each modality, the modalities are first paired to form combinations, such as the combination of CT and X-ray, CT and ultrasound, and X-ray and ultrasound. For each combination, the corresponding structural indicators are extracted in the same scoring key area, such as texture feature values, density distribution values, etc. The covariance between each pair of modal structural indicators is calculated, and the standard deviation of the structural indicators of each modality is calculated separately. The covariance is divided by the product of two standard deviations to obtain the correlation coefficient of the combination. This process is repeated until all pairwise combinations are calculated, and the arithmetic average of all correlation coefficients is taken to obtain the final consistency index. If the consistency index exceeds the preset threshold, it means that the structural indicators between the multiple modalities are highly consistent, and the mean fusion method is adopted; if it is lower than the threshold, it indicates that there are significant differences between the modalities, and weighted fusion is performed according to the credibility of each modality. For example, when the correlation coefficient between CT and X-ray is 0.85, that between CT and ultrasound is 0.72, and that between X-ray and ultrasound is 0.65, the consistency index is (0.85+0.72+0.65) / 3=0.74. Assuming the preset consistency threshold is 0.7, mean fusion is performed.

[0065] Traditional multimodal fusion methods typically only calculate the correlation coefficient of a single modality combination or simply average the data across all modalities, failing to systematically evaluate the interrelationships between all modalities. For example, existing techniques may only calculate the correlation between CT and X-rays, while ignoring the combination of CT and ultrasound, resulting in ineffective identification of data deviations between some modalities. By exhaustively enumerating all pairwise modality combinations and calculating the average correlation coefficient, this method can more comprehensively capture the consistency and differences in multimodal data, avoiding the evaluation bias caused by calculating a single combination, thereby improving the scientific nature of the fusion strategy selection.

[0066] Through the above technical solution, the present application can effectively identify local inconsistent areas in multimodal data. For example, when the ultrasound modality has significant differences in structural indicators compared to CT due to imaging quality, the correlation coefficient calculation can accurately reflect this difference, thereby triggering a weighted fusion strategy based on credibility. This approach improves adaptability to modality quality fluctuations while ensuring the stability of the evaluation results, solving the error accumulation problem caused by simple average fusion in the existing technology. It is particularly suitable for complex clinical scenarios where some modalities have poor imaging quality.

[0067] This application further proposes to calculate the absolute difference between the structural indicators under each mode and the manually labeled reference structural indicators, average all absolute differences under the same mode to obtain the average deviation value, take the inverse of the average deviation value and normalize it as the modal credibility weight.

[0068] Manually annotated reference structural indices refer to clinically significant structural feature values ​​manually labeled by professional physicians based on imaging data. This can be achieved by averaging the values ​​of two annotations, providing a baseline for the true value of the structural indices. The absolute difference refers to the absolute difference between the algorithm-derived structural indices and the manually annotated values ​​within the same structural block. This can be achieved by calculating the absolute values ​​point by point, taking the sum of the absolute values, and then calculating the difference. This measures the degree of deviation between modality data and the true value. The average deviation refers to the arithmetic average of the absolute differences across all structural blocks within the same modality. This can be achieved by summing the differences and dividing by the total number of blocks, reflecting the overall deviation of the modality from the true value. The reciprocal operation refers to the mathematical reciprocal of the average deviation value. This can be achieved through a numerical conversion formula, converting the deviation value into a positive confidence indicator. Normalization involves adjusting the reciprocal results of each modality to a weighted ratio that sums to 1. This can be achieved through linear scaling to ensure comparability of confidence weights across modalities.

[0069] Specifically, after obtaining multimodal structural indices, the algorithm-derived structural indices are first compared block by block with manually annotated reference values. For example, for a structural block in the CT modality, the algorithm-calculated density distribution value is 85HU, while the manually annotated value is 80HU, resulting in an absolute difference of 5. By averaging the absolute differences across all blocks in that modality, assuming an average deviation of 3.2, its reciprocal is approximately 0.3125. If the average deviation for the ultrasound modality is 4.0, its reciprocal is 0.25. Adding the reciprocal values ​​of 0.3125 and 0.25 for the two modalities yields a total of 0.5625. After normalization, the confidence weights for the CT modality are 0.3125 / 0.5625≈0.556, and for the ultrasound modality are 0.25 / 0.5625≈0.444. Therefore, during weighted fusion, the more reliable CT modality data is given a higher weight, thereby reducing the impact of the modality with larger deviations on the final score.

[0070] Traditional methods typically perform multimodal data fusion based on preset fixed weights or simple averaging strategies, such as assigning fixed weights of 0.6 and 0.4 to CT and X-ray modalities, respectively. However, in real-world scenarios, the imaging quality of different modalities may fluctuate due to factors such as equipment differences and patient positioning, and fixed weights cannot dynamically adapt to data deviations. This solution, by quantifying the deviation between modality data and the true value of manual annotations, transforms the credibility calculation into a mathematical derivation process, allowing the weight assignment to dynamically adjust with actual data quality, avoiding the spread of scoring errors caused by distortion of single modality data.

[0071] Through the above technical solution, this application solves the problem of unreasonable weight distribution caused by quality differences when fusion of multimodal data. For example, when the structural indicators of some areas of ultrasound images deviate significantly due to lung gas interference, their credibility weight is automatically reduced, while the CT modality receives a higher weight due to its higher data stability, thereby significantly improving the clinical consistency of the fusion results. At the same time, the objective calculation process based on manually annotated reference values ​​reduces the reliance on empirical parameter settings and enhances the robustness of the evaluation system.

[0072] The present application further proposes that the calculation process of the dynamic weight includes: based on each structural block in the scoring key area, the absolute value of the difference between the structural indicators after fusion between adjacent structural blocks is calculated to obtain the spatial variation; all spatial variations in the scoring key area are normalized to obtain the variation amplitude factor; in the same scoring key area, for the structural indicators before fusion under different modes, the standard deviation is calculated to obtain the consistency deviation value; the consistency deviation is normalized to obtain the consistency adjustment factor; the variation amplitude factor and the consistency adjustment factor are weighted according to the set ratio to obtain the dynamic weight of each scoring key area.

[0073] Among them, the spatial variation refers to the degree of difference in structural indicators after fusion between adjacent structural blocks. Specifically, it can be achieved by calculating the absolute difference in structural indicators of adjacent blocks, which is used to reflect the amplitude of local structural changes in the key scoring area. Normalization processing refers to converting numerical values ​​of different dimensions or ranges into a unified ratio. Specifically, the minimum-maximum normalization method can be used to map the spatial variation to the range of 0 to 1 to eliminate the influence of dimensional differences on weight calculation. The consistency deviation value refers to the degree of discreteness of structural indicators in different modalities of the same key scoring area. Specifically, it can be achieved by calculating the standard deviation, which is used to measure the credibility difference of multimodal data in this area. The dynamic weight refers to the contribution of the key scoring area to the final lung injury score. Specifically, it can be obtained by linearly weighting the variation amplitude factor and the consistency adjustment factor. For example, the ratio is set to 6:4 or 7:3 to achieve a balance between spatial heterogeneity and modal consistency.

[0074] Specifically, when calculating the dynamic weight, first traverse all the structural blocks in the key scoring area, and calculate the absolute value of the difference between the fusion structural indicators of each block and its adjacent blocks in turn to form a set of spatial variation. The set is then normalized so that the normalized value corresponding to the maximum variation is 1 and the minimum is 0. Next, for the same key scoring area, the original structural indicators before fusion are extracted from each modality, and their standard deviations are calculated and normalized to obtain a consistency adjustment factor that reflects the degree of deviation between modalities. Finally, the normalized variation amplitude factor and the consistency adjustment factor are weighted and summed according to a preset ratio, for example, the variation amplitude factor accounts for 60% and the consistency adjustment factor accounts for 40%, to generate a dynamic weight value. This weight value will directly participate in the weighted calculation process of the lung injury score.

[0075] Traditional methods use a fixed weighting strategy, which is unable to adapt to structural heterogeneity and modal credibility fluctuations in different key scoring regions. However, this solution introduces a dual adjustment mechanism of spatial variation and consistency deviation. It can dynamically adjust weights based on the degree of structural mutation within a region and the level of inter-modal consistency. For example, higher weights are assigned to areas with significant structural differences and high modal consistency, thereby improving the objectivity and adaptability of scoring.

[0076] Through the above technical solution, this application solves the problem that static weights cannot reflect changes in local structural features and differences in modal credibility, and realizes dynamic optimization of the weights of key scoring areas. In the presence of uneven distribution of lesions or fluctuations in the quality of some modal data, the solution can automatically adjust the contribution of each area in the score. For example, the weight of areas with blurred edges but high modal consistency will be reduced, while the weight of areas with obvious structural mutations and consistent multimodal data will be increased, thereby significantly improving the accuracy of lung injury scores and the credibility of clinical assessment results.

[0077] The present application further proposes preset integrity conditions, including that the total area of ​​the scoring key area accounts for a proportion of the lung area that is not less than a set lower limit and covers all predefined anatomical key points; the process of reconstructing the scoring key area includes selecting several structural blocks whose spatial positions are continuous with the edges of the scoring key area, and adding them to the current scoring key area to form an extended scoring key area.

[0078] The preset completeness condition refers to the minimum coverage range and key anatomical location coverage requirements that the key scoring area must meet. This can be achieved by setting an area ratio threshold and a predefined anatomical coordinate list to ensure that the scoring area is sufficiently representative and covers key lesion areas. The process of reconstructing the key scoring area refers to supplementing the area through neighborhood expansion when the completeness condition is not met. Specifically, an edge continuity detection algorithm can be used to screen adjacent structural blocks, and regional expansion can be achieved through iterative addition to solve the problem of insufficient coverage of the initial scoring area.

[0079] Specifically, after completing the screening of the key scoring areas, the system automatically calculates the proportion of its total area to the lung area and compares it with the preset lower limit. At the same time, check whether the predefined anatomical key points are all contained in the key scoring area. If the above conditions are not met at the same time, starting from the edge of the current area, adjacent structural blocks that are continuous with its physical space are selected for merging. The expanded area is verified for integrity conditions again until the requirements are met or the preset upper limit of the number of iterations is reached. When the conditions are still not met at the end of the iteration, the average value of the scoring results generated in the historical iterations is used as the final output.

[0080] In some embodiments, predefined anatomical key points may include landmark locations such as bronchial bifurcations and pulmonary lobe junctions, with a lower limit set at, for example, 30% of the total lung area. When expanding the scoring key region, a spatial adjacency algorithm may be used to calculate the distance between the candidate block and the edge of the current region, with the block with the smallest distance being preferentially selected for merging.

[0081] Traditional methods rely solely on a single area threshold to assess the completeness of the scoring region, failing to consider the coverage requirements for key anatomical locations, which can easily lead to the omission of important lesions. However, this solution, by combining the dual constraints of area proportion and anatomical key points and employing a dynamic expansion mechanism, can adaptively adjust the scoring region to ensure the comprehensiveness and accuracy of the assessment results.

[0082] Through the above technical solution, this application can effectively avoid evaluation bias caused by incomplete coverage of key scoring areas, especially when there are lesions in key anatomical areas such as the edge of the lung lobe or the bronchial bifurcation. By forcibly covering predefined key points, the clinical applicability of the lung injury score and the credibility of the evaluation results are significantly improved.

[0083] Example 2:

[0084] like Figure 4 As shown in the figure, the machine learning-based acute lung injury assessment system for critically ill patients includes:

[0085] A data acquisition module is used to acquire multimodal chest imaging data of critically ill patients and perform standardization processing on the multimodal chest imaging data to generate a multimodal standard data set;

[0086] a data processing module, configured to divide the lung region into a plurality of structural blocks, and extract structural indicators within the structural blocks based on the standard data set; if structural indicators are missing in any modality, spatial mapping compensation is preferentially performed on the same structural blocks in other modalities; if compensation is still not possible, the optimal candidate block is selected from the structural blocks in the current modality, and its structural indicators are extracted for compensation;

[0087] A region extraction module is used to identify the modal conflict region and the structural mutation region in the structural block based on the structural indicators, and merge the modal conflict region and the structural mutation region to construct a scoring key region;

[0088] A region optimization module is used to apply a preset disturbance to the structural indicators of the key scoring region and calculate the disturbance impact value, and determine whether the disturbance impact value is greater than a preset impact threshold. If so, the key scoring region is retained; otherwise, the key scoring region is eliminated;

[0089] The feature fusion module is used to calculate the consistency index based on the structural indicators of the same scoring key area in all modalities, and determine whether the consistency index is greater than a preset consistency threshold. If so, the structural indicators of the same scoring key area are averaged and fused. Otherwise, weighted fusion is performed based on the credibility of each modality.

[0090] an injury assessment module, configured to calculate a lung injury score for the critically ill patient based on the fused structural indicators and the dynamic weights of the structural indicators in key scoring areas;

[0091] The evaluation and optimization module is used to determine whether the key scoring area meets the preset integrity conditions. If so, the lung injury score is output; otherwise, the key scoring area is reconstructed and the lung injury score is calculated until the preset number of iterations is reached. If the integrity conditions are still not met, the average lung injury score of all iterations is output.

[0092] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

[0093] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0094] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for assessing acute lung injury in critically ill patients based on machine learning, characterized by: The following steps are involved: Acquiring multimodal chest imaging data of critically ill patients, and performing standardization processing on the multimodal chest imaging data to generate a multimodal standard data set; Dividing the lung region into a plurality of structural blocks, and extracting structural indicators within the structural blocks based on the standard data set; If there is a lack of structural indicators in any mode, spatial mapping compensation is performed on the same structural blocks in other modes first; if compensation is still not possible, the best candidate block is selected from the structural blocks in the current mode and its structural indicators are extracted for compensation; Identifying modal conflict regions and structural mutation regions in the structural block based on the structural indicators, and merging the modal conflict regions and structural mutation regions to construct a scoring key region; Applying a preset disturbance to the structural indicators of the key scoring area and calculating the disturbance impact value, determining whether the disturbance impact value is greater than a preset impact threshold, and if so, retaining the key scoring area; otherwise, eliminating the key scoring area; A consistency index is calculated based on the structural indicators of the same scoring key area in all modalities, and it is determined whether the consistency index is greater than a preset consistency threshold. If so, the structural indicators of the same scoring key area are averaged and fused. Otherwise, a weighted fusion is performed based on the credibility of each modality. Calculating a lung injury score for the critically ill patient based on the fused structural indicators and the dynamic weights of the structural indicators in key scoring areas; Determining whether the key scoring area meets a preset integrity condition, and outputting a lung injury score if so; Otherwise, the scoring key area is reconstructed and the lung injury score is calculated until the preset number of iterations is reached. If the integrity condition is still not met, the average lung injury score of all iterations is output.

2. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 1, characterized in that: The chest multimodal image data includes CT scan images, X-ray images and ultrasound images; The standardization process includes image resolution unification, grayscale value normalization and spatial alignment; The specific process of the standardization processing includes: unifying the resolution of different modal images to a preset reference resolution; normalizing the grayscale values ​​by stretching the grayscale histogram; and performing spatial alignment using the sternal angle and tracheal bifurcation as anatomical reference points.

3. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 2, characterized in that: The division of the structural blocks is based on the anatomical structure characteristics of the lungs, and the lung area is divided into a plurality of rectangular grid units, wherein the segmentation boundaries of the rectangular grid units are aligned with the bifurcation points of the bronchial tree; The structural indicators include texture feature values, density distribution values ​​and edge clarity; The texture feature value is obtained by calculating the contrast of the pixel grayscale value in the structure block; The density distribution value is obtained by statistically calculating the mean of the CT values ​​within the structural block; the edge clarity is quantified by the Canny operator response intensity; The spatial mapping compensation includes spatially aligning the missing structural blocks in the modality with the known structural blocks in other modalities and using the structural indicators of the latter as replacement data; The process of selecting the optimal candidate block is as follows: calculating the physical space distance between all structural blocks in the current modality image and the structural blocks with missing structural indicators; Set a distance threshold and filter out a set of candidate blocks whose physical space distance is less than the distance threshold; if the set of candidate blocks is empty, expand the distance threshold range and filter again until at least one candidate block is found; the physical space distance is calculated based on the image coordinate system; Extract the spatial coordinates and structural indicators of all complete structural blocks in the current modality image; use the spatial coordinates as input and the vector of structural indicators as output to establish a block feature prediction model through supervised learning; input the spatial coordinates of the structural blocks with missing structural indicators into the trained block feature prediction model, and output the predicted structural indicators; Calculate the feature similarity between the candidate block and the predicted structural indicators; The structural block with the largest feature similarity is selected as the optimal candidate block; the feature similarity is obtained by calculating the cosine similarity between the vectors of the structural indicators.

4. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 3, characterized in that: The process of identifying the modal conflict area includes: calculating the variance of the structural index of the same structural block between different modes, and determining it as a modal conflict area if the variance is greater than the conflict determination threshold; The identification process of the structural mutation region includes: calculating the Euclidean distance difference between the structural indicators of the structural block and the adjacent structural blocks in a single mode, and marking it as a structural mutation region if the difference exceeds a preset mutation threshold.

5. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 4, characterized in that: The calculation process of the disturbance impact value includes: For each type of structural indicator, the average value is calculated in all key scoring areas, and the average values ​​of different types of structural indicators are weighted and summed according to the preset weights to obtain the benchmark score value; In the key scoring area, a preset fixed ratio of positive and negative adjustments is applied to each type of structural indicator. The benchmark score is calculated based on the adjusted structural indicators to obtain positive and negative score values ​​respectively. The calculation formula for the disturbance impact value is: disturbance impact value = ((|positive score value - benchmark score value|) + (|negative score value - benchmark score value|)) ÷ 2.

6. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 5, characterized in that: The calculation process of the consistency index includes: Combine all modes in pairs and calculate the correlation coefficient between their structural indicators for each combination; average all correlation coefficient values ​​to obtain the consistency index; The correlation coefficient is the ratio of the covariance between the structural indicators to the product of the standard deviation.

7. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 6, characterized in that: The calculation process of the credibility of each modality includes: For the structural index under each mode, the absolute difference between it and the manually annotated reference structural index is calculated. All absolute differences under the same mode are averaged to obtain the average deviation value. The inverse of the average deviation value is taken and normalized as the modal credibility weight.

8. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 7, characterized in that: The calculation process of the dynamic weight includes: Based on each structural block in the scoring key area, the absolute value of the difference between the structural indicators of adjacent structural blocks after fusion is calculated to obtain the spatial variation; all spatial variations in the scoring key area are normalized to obtain the variation amplitude factor; In the same scoring key area, for the structural indicators before fusion under different modalities, the standard deviation is calculated to obtain the consistency deviation value; the consistency deviation is normalized to obtain the consistency adjustment factor; The change amplitude factor and consistency adjustment factor are weighted according to the set ratio to obtain the dynamic weight of each key scoring area.

9. The method for assessing acute lung injury in critically ill patients based on machine learning according to claim 8, characterized in that: The preset completeness conditions include: the total area of ​​the key scoring regions accounts for a proportion of the lung area that is not less than a set lower limit and covers all predefined anatomical key points; The process of reconstructing the key scoring area includes: selecting a number of structural blocks whose spatial positions are continuous with the edge of the key scoring area, and adding them to the current key scoring area to form an extended key scoring area.

10. A machine learning-based acute lung injury assessment system for critically ill patients, characterized by: The method for assessing acute lung injury in critically ill patients based on machine learning according to any one of claims 1 to 9 is used, comprising: A data acquisition module is used to acquire multimodal chest imaging data of critically ill patients and perform standardization processing on the multimodal chest imaging data to generate a multimodal standard data set; a data processing module, configured to divide the lung region into a plurality of structural blocks, and extract structural indicators within the structural blocks based on the standard data set; if structural indicators are missing in any modality, spatial mapping compensation is preferentially performed on the same structural blocks in other modalities; if compensation is still not possible, the optimal candidate block is selected from the structural blocks in the current modality, and its structural indicators are extracted for compensation; A region extraction module is used to identify the modal conflict region and the structural mutation region in the structural block based on the structural indicators, and merge the modal conflict region and the structural mutation region to construct a scoring key region; A region optimization module is used to apply a preset disturbance to the structural indicators of the key scoring region and calculate the disturbance impact value, and determine whether the disturbance impact value is greater than a preset impact threshold. If so, the key scoring region is retained; otherwise, the key scoring region is eliminated; The feature fusion module is used to calculate the consistency index based on the structural indicators of the same scoring key area in all modalities, and determine whether the consistency index is greater than a preset consistency threshold. If so, the structural indicators of the same scoring key area are averaged and fused. Otherwise, weighted fusion is performed based on the credibility of each modality. an injury assessment module, configured to calculate a lung injury score for the critically ill patient based on the fused structural indicators and the dynamic weights of the structural indicators in key scoring areas; The evaluation and optimization module is used to determine whether the key scoring area meets the preset integrity conditions. If so, the lung injury score is output; otherwise, the key scoring area is reconstructed and the lung injury score is calculated until the preset number of iterations is reached. If the integrity conditions are still not met, the average lung injury score of all iterations is output.

Images