AKI identification method and system combining biomarkers and image features

Abstract

The invention relates to the technical field of acute kidney injury recognition, in particular to an AKI recognition method and system combining biomarkers and image features, and the method comprises the following steps: based on acute kidney injury high-risk groups, analyzing a matching relation between a biomarker concentration trend and a historical mode direction, constructing a dynamic linkage feature, extracting a renal region image rotation plane feature, and constructing an AKI recognition model; and recognizing the structure continuity abnormity, and calculating the offset condition between the image track and the physiological fluctuation. According to the method, a symbol sequence is used for replacing a concentration threshold, direction consistency is recognized as an early response feature, static index limitation is broken through, a rotation axis section construction mode is adopted for image processing, texture, gray scale and density distribution expression under a multi-angle image is unified, the space tracking capability of anatomical features is enhanced, and the method is suitable for being applied to the field of image processing. The linkage offset of the gray track and the physiological parameter fluctuation forms a time period mapping basis, and the overlapping fragment coverage rate is used for judging the cooperative state, so that the time locking judgment between the structure change and the physiological trend is realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of acute kidney injury recognition, in particular to an AKI recognition method and system combining biomarkers and image features. BACKGROUND

[0002] The technical field of acute kidney injury (AKI) recognition belongs to the key direction of the cross-application of medical diagnosis and monitoring, and covers early detection of abnormal kidney function, disease progression assessment and intervention decision support and the like. The field comprehensively uses multidisciplinary technical means such as clinical biochemistry, physiological monitoring, medical imaging, data modeling and intelligent algorithms, and aims to improve the diagnostic accuracy and timeliness of AKI, reduce disease delay and reduce patient mortality.

[0003] Among them, the AKI recognition method combining biomarkers and image features is a diagnostic technology that comprehensively uses biomarker information and medical image data, and is used for early recognition and risk assessment of acute kidney injury. By fusing biochemical indicators (such as serum creatinine, NGAL, etc.) and kidney image manifestations (such as structural changes, perfusion characteristics, etc.), the diagnostic sensitivity and specificity of AKI can be improved, and the method is widely used in clinical diagnosis auxiliary systems or critical patient monitoring devices and the like.

[0004] The prior art often determines the risk of acute kidney injury through a single point concentration, ignores the value of index fluctuation trend in diagnosis, is easy to cause delayed recognition, and mainly uses two-dimensional static sections for image processing. The spatial consistency evaluation path of the structure under multiple angles has not been established. The texture distortion and density overlap caused by the rotation direction cannot be accurately identified, causing judgment deviation of the anatomical structure. The linkage recognition path of physiological parameters and image results is mainly located in the later decision-making process, and lacks pre-alignment and dynamic judgment means, which is easy to cause the problem that the image abnormality and the physiological abnormality are not synchronized. In addition, in the scene where the postoperative state fluctuates frequently and high-frequency monitoring demand is prominent, the existing technology recognition mechanism lags behind, and cannot timely capture the potential high-risk performance in the fluctuation transition section, affecting the intervention timeliness and reducing the pertinence and stability of the diagnostic response. SUMMARY

[0005] In order to solve the technical problems in the prior art that the risk of acute kidney injury is often judged by single point concentration, the value of index fluctuation trend in diagnosis is ignored, delay recognition is easy to occur, image processing is mainly based on two-dimensional static section, the spatial consistency evaluation path under the structure of multiple angles is not established, the texture distortion and density overlap caused by the rotation direction cannot be accurately identified, the anatomical structure judgment deviation is caused, the linkage recognition path of physiological parameters and image results is mainly located in the late decision-making process, the pre-alignment and dynamic judgment means are lacked, the problem that image abnormalities and physiological abnormalities are not synchronized is easy to occur, in addition, in the scene where postoperative state fluctuation occurs frequently and high-frequency monitoring demand is prominent, the recognition mechanism of the prior art is lagging behind, potential high-risk performance in the fluctuation transition section cannot be captured in time, the intervention timeliness is affected, and the technical problems that the diagnosis response is reduced in pertinence and stability are solved, and the AKI recognition method combining biomarkers and image features is provided. The technical scheme is as follows:

[0006] In one aspect, an AKI recognition method combining biomarkers and image features is provided, including the following steps:

[0007] S1: Based on the high-risk population of acute kidney injury, creatinine, cystatin C and IL-18 sequence are analyzed, the change direction is calculated, the lag mode is matched, the consistent trend segment is screened, the physiological trend continuity is judged, and the dynamic trend linkage feature is obtained;

[0008] S2: According to the dynamic trend linkage feature, the kidney portal vein path symmetry axis is optimized, the rotation plane is constructed, the gray scale, texture and density difference are analyzed, the structure consistency is judged, and the rotation plane anatomical feature group is obtained;

[0009] S3: According to the rotation plane anatomical feature group, the gray scale, texture and density changes between image frames are compared, the direction consistency is judged, the fluctuation amplitude is analyzed, the mutation area and deviation trend are identified, and the structure continuity abnormal data is obtained;

[0010] S4: According to the structure continuity abnormal data, the kidney cortex gray scale track is calculated, the moving direction is analyzed, the creatinine, urea nitrogen and body fluid trend are compared, the linkage deviation is judged, the corresponding abnormal segment is screened, and the multi-modal cooperative deviation marker is obtained;

[0011] S5: Based on the multi-modal cooperative deviation marker, the lag segment and image track key node are compared, the time overlap range is screened, the coverage ratio is calculated, whether the recognition condition is met is judged, and the early cooperative abnormal locking result is obtained.

[0012] In another aspect, the dynamic trend linkage feature includes a concentration direction symbol, a trend sequence number, and a lag matching identifier, the rotation plane anatomical feature group includes a symmetric axis gray baseline, a rotation plane texture distribution, and a voxel aggregation degree, the structural continuity abnormality data includes a parameter fluctuation boundary, a structure deflection dimension, and a spatial mutation label, the multi-modal collaborative bias marker includes an image translation path, a physiological fluctuation trajectory, and a linkage difference region, and the early collaborative abnormality locking result includes an abnormality synchronization segment, a cross-alarm section, and early warning state information.

[0013] In another aspect, the acquisition step of the dynamic trend linkage feature is specifically:

[0014] S101: Based on the high-risk population of acute kidney injury, analyze the monitoring sequence of serum creatinine, cystatin C and IL-18, compare the detection results of adjacent time nodes in turn, mark the relative increase or decrease trend, and obtain the direction transfer marker sequence;

[0015] S102: Compare the current time segment in the direction transfer marker sequence with the adjacent segment in the historical monitoring sequence, align the direction markers by time points, judge the consistency, analyze whether the direction in each corresponding relationship maintains the same trend, and count the proportion of time nodes with consistent direction in the total segments to obtain the time consistency proportion;

[0016] S103: Screen the segments with a duration that meets the basic judgment requirement and a proportion that meets the linkage standard in the time consistency proportion, analyze whether the change direction in the segments that meet the conditions remains consistent, output the monitoring time range with common trend direction, and obtain the dynamic trend linkage feature.

[0017] In another aspect, the acquisition step of the rotation plane anatomical feature group is specifically:

[0018] S201: According to the dynamic trend linkage feature, locate the approximate axis distribution of the renal portal vein region in the gray image, analyze the symmetry intensity distribution of the two sides of the renal portal structure in the gray image, optimize the center trend of the symmetric axis, adjust the distribution position of the axis crossing point in the gray gradient change region, and obtain the symmetric path distribution information;

[0019] S202: Call the symmetric path distribution information to construct a rotation plane set with the axis as the reference, analyze the lateral variation amplitude of the gray distribution in each plane, calculate the dynamic distribution sequence of the gray difference on both sides of the axis under multi-angle rotation, judge whether the gray feature maintains a continuous transition state in the rotation process, and obtain the angle gray variation trend;

[0020] S203: According to the angle gray scale change trend, the arrangement direction of the kidney area texture in each plane is compared with the spatial position relationship of the voxel aggregation area in the image, whether the structure distribution in the multi-angle image section exists common characteristics is judged, the spatial combination of gray scale, texture and density is integrated, and the rotation plane anatomical feature group is obtained.

[0021] On the other hand, the structure continuity abnormal data acquisition step is specifically:

[0022] S301: According to the rotation plane anatomical feature group, the gray scale, texture direction and density change trend of the continuous image frame in the kidney area range are analyzed, whether the spatial direction between the time nodes is consistent is compared, whether there exists direction reversal or offset feature is identified, and the joint direction changing trajectory sequence is obtained.

[0023] S302: The joint direction changing trajectory sequence is called to screen the image segments with discontinuous trend on the gray scale, texture and density path, whether the position offset in the corresponding image frame is concentrated in the target area is judged, the corresponding image coordinate section and its position distribution are marked, and the local space disturbance area is obtained.

[0024] S303: According to the local space disturbance area, whether the structure axis exists trajectory abnormality in the target area is judged, whether the offset behavior exceeds the spatial tolerance range is analyzed, and the structure offset distribution range and deviation condition are tested according to the structure offset, and the structure continuity abnormal data is obtained.

[0025] On the other hand, whether the offset behavior exceeds the spatial tolerance range is analyzed by using the formula:

[0026]

[0027] The offset feature value D is calculated dev , whether the structure axis exists trajectory abnormality in the target area is judged, wherein n represents the number of structure control points, represents the actual measured coordinates of the i th structure control point in space, represents the design coordinates of the i th structure control point in space, represents the mean value of the offset of all structure control points, d tol represents the tolerance limit threshold of structure offset.

[0028] On the other hand, the multi-modal collaborative offset marking acquisition step is specifically:

[0029] S401: According to the structure continuity abnormal data, the gray scale gravity center coordinate distribution of the renal cortex area in the continuous image frame is analyzed, the horizontal and vertical translation path of the gravity center position between each image frame is calculated, whether the path forms forward arrangement or appears position jump in the image plane is judged, and the image sequence displacement trend is obtained.

[0030] S402: Based on the image sequence displacement trend, compare the numerical change direction of creatinine, urea nitrogen and body fluid parameters in the same time period, judge whether the continuous trend of parameter sequence deviates from the direction of image trajectory, identify the combination items forming inconsistent trend, and obtain parameter trajectory deviation coefficient;

[0031] S403: Based on the parameter trajectory deviation coefficient, filter the time sequence fragments with linkage abnormality in the deviation coefficient fluctuation section, judge whether the fragments form co-occurrence structure between image space and detection parameters, and identify the spatial coordinates and data intervals of such time periods, and obtain multi-modal collaborative deviation marker.

[0032] On the other hand, the judgment whether the continuous trend of parameter sequence deviates from the direction of image trajectory adopts the formula:

[0033]

[0034] Identify the combination items forming inconsistent trend to obtain parameter trajectory deviation coefficient ΔSY dev , wherein Δ represents the difference between the z-th moment creatinine, urea nitrogen or body fluid parameter and its previous moment, represents the change value of the gray center trajectory in the main axis direction in the z-th frame image, represents the change value of the direction vector at the z-th moment, represents the difference between the z-th physiological parameter value and the average value of the physiological parameters of the previous three consecutive moments, represents the standard deviation of the gray distribution of the z-th frame of rotating planar image in the renal cortex region, n SY is the total number of image frames in the selected image time period.

[0035] On the other hand, the acquisition step of the early collaborative abnormality locking result is specifically:

[0036] S501: Based on the multi-modal collaborative deviation marker, analyze the positioning information of each time period, compare the time label with the time index of the lagging fragments and image key points in the dynamic trend linkage feature, filter the fragment range with intersection in time dimension, and obtain the synchronous overlapping time period index set;

[0037] S502: Call the synchronous overlapping time period index set, calculate the coverage degree of the overlapping fragments in the image key point sequence, judge whether the coverage ratio reaches the linkage recognition reference range, identify the fragment sequence with strong continuity, and mark the fragment boundary index, and obtain the coincidence fragment coverage ratio;

[0038] S503: According to the coincidence fragment coverage ratio, it is judged whether the linkage range between the image track and the physiological index exists concentrated deviation, time nodes showing consistent linkage reaction in continuous fragments are screened, time range and key feature annotation information are integrated, and early collaborative abnormality locking result is obtained.

[0039] In another aspect, an AKI identification system combining biomarkers and image features is provided, which is applied to the AKI identification method combining biomarkers and image features, and includes:

[0040] The trend linkage identification module analyzes creatinine, cystatin C and IL-18 sequence based on the high-risk population of acute kidney injury, calculates the change direction, matches the lag mode, screens the consistent trend fragments, judges the physiological trend continuity, and obtains the dynamic trend linkage feature.

[0041] The symmetrical structure extraction module optimizes the symmetry axis of the renal portal vein path according to the dynamic trend linkage feature, constructs a rotation plane, analyzes the differences in gray, texture and density, judges the structural consistency, and obtains a rotation plane anatomical feature group.

[0042] The image variation detection module compares the gray, texture and density changes between image frames according to the rotation plane anatomical feature group, judges the direction consistency, analyzes the fluctuation amplitude, identifies the mutation area and deviation trend, and obtains the structural continuity abnormal data.

[0043] The linkage deviation calculation module calculates the renal cortex gray track according to the structural continuity abnormal data, analyzes the moving direction, compares the creatinine, urea nitrogen and body fluid trend, judges the linkage deviation, screens the corresponding abnormal fragments, and obtains the multi-modal collaborative deviation marker.

[0044] The abnormal time sequence locking module compares the lag fragments and the image track key nodes based on the multi-modal collaborative deviation marker, screens the time overlap range, calculates the coverage ratio, judges whether the identification condition is met, and obtains the early collaborative abnormality locking result.

[0045] The technical scheme provided by the embodiment of the application has at least the following beneficial effects:

[0046] By replacing the concentration threshold with a symbol sequence, the directional consistency is identified as an early response feature, breaking the limitations of static indicators. Image processing uses a rotating axis section construction method to unify the texture, gray scale and density distribution performance under multiple angle images, enhancing the spatial tracking ability of anatomical features. The linkage offset between the gray scale trajectory and the physiological parameter fluctuation forms the time period mapping basis, and the overlapping segment coverage is used to identify the cooperative state, realizing the time locking judgment between structural changes and physiological trends. This logic improves the response ability of the recognition mechanism to continuous abnormal trends, avoids the dependence on single time point errors, enhances the sensitivity of atypical state recognition, and helps to quickly separate the high-risk evolution stage, thereby supporting early judgment and precise intervention under complex conditions. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0048] Figure 1 It is the main step flow chart of the present application;

[0049] Figure 2 It is the step flow chart of S1 of the present application;

[0050] Figure 3 It is the step flow chart of S2 of the present application;

[0051] Figure 4 It is the step flow chart of S3 of the present application;

[0052] Figure 5 It is the step flow chart of S4 of the present application;

[0053] Figure 6 It is the step flow chart of S5 of the present application;

[0054] Figure 7 It is the system block diagram of the present application. DETAILED DESCRIPTION

[0055] The technical solutions in the present application will be described below in combination with the drawings.

[0056] In the embodiments of the present application, the words such as "example", "for example" are used to represent an example, illustration, or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0057] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized. "Of", "corresponding" and "corresponding" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized.

[0058] In the embodiments of the present application, sometimes the subscript such as W1 is written in the form of non-subscript such as W1, and the meanings expressed are consistent when the distinction is not emphasized.

[0059] In order to make the technical problems, technical schemes and advantages to be solved by the present application more clear, the following will be described in detail in combination with the drawings and specific embodiments.

[0060] The embodiments of the present application provide an AKI identification method combining biomarkers and image features, as shown in Figure 1 The method comprises the following steps:

[0061] S1: Based on a high-risk population of acute kidney injury, analyze the concentration sequences of serum creatinine, cystatin C and IL-18 in continuous monitoring, calculate the rising and falling directions at each time node, compare the direction consistency of the current sequence and the historical lag mode, screen the fragments with consistent characteristics of sustained direction, judge the stability of the continuous trend in physiological response, and obtain the dynamic trend linkage feature;

[0062] S2: According to the dynamic trend linkage feature, optimize the symmetry axis positioning mode of the renal portal vein path in the gray image, adjust the gradient change trajectory of the axis crossing point, construct a rotation scanning plane with the axis as the core, analyze the feature differences of the kidney region image in gray distribution, texture direction and voxel density, judge whether the plane features are consistent, and obtain the rotation plane dissection feature group;

[0063] S3: According to the rotation plane anatomical feature group, compare the change trend of gray average, texture direction and density distribution in time nodes in the continuous image frames, judge whether the dimension change direction is consistent, analyze the fluctuation amplitude, identify the spatial region with discontinuous direction or feature inversion, judge whether the structure axis positioning appears abnormal deviation, if it meets the spatial deviation standard, get the structure continuity abnormal data;

[0064] S4: According to the structure continuity abnormal data, calculate the motion trajectory of the gray center of the renal cortex region in the image sequence, analyze the coherent direction of the trajectory path in the image plane, compare the continuous change trend of creatinine, urea nitrogen and body fluid related parameters in the target period, judge whether it forms linkage deviation with the image trajectory, screen the time period with corresponding abnormalities, get the multi-modal collaborative deviation mark;

[0065] S5: Based on the multi-modal collaborative deviation mark, compare the lag segment in the dynamic trend linkage feature with the key point sequence in the image abnormal trajectory, screen the time range with overlap, calculate the coverage ratio of the overlapping segment, judge whether it exceeds the linkage recognition reference range, if the time overlap is obvious, determine it as the abnormal stage, get the early collaborative abnormal locking result.

[0066] The dynamic trend linkage feature includes concentration direction symbol, trend sequence number, lag matching mark, the rotation plane anatomical feature group includes symmetry axis gray baseline, rotation plane texture distribution, voxel aggregation degree, the structure continuity abnormal data includes parameter fluctuation boundary, structure deflection dimension, spatial mutation label, the multi-modal collaborative deviation mark includes image translation path, physiological fluctuation trajectory, linkage difference area, the early collaborative abnormal locking result includes abnormal synchronous segment, cross warning section, early warning state information.

[0067] The current sequence refers to the sequence of biomarker concentrations collected by the current subject within the target detection period (such as during postoperative continuous monitoring), including time series data of serum creatinine, cystatin C and IL-18, and equal interval between data, forming comparable trend sequence; the historical lag mode refers to the directional symbol template sequence formed by the fluctuation trajectory of biomarker concentration before the occurrence of kidney injury in the archived high-risk cases of similar acute kidney injury; the template sequence is used to establish a lag response standard as a comparison reference; the characteristic fragment refers to the time sequence fragment in the "current sequence" that has continuous consistency with the "historical lag mode", and the matching condition is that the symbol trend is consistent and the length reaches the linkage recognition requirement, which belongs to the potential early abnormal signal fragment; the symmetry axis positioning method refers to the axis extraction method based on the brightness gradient distribution of the renal portal vein passing area in the kidney grayscale image, which has anatomical structure symmetry reference significance. The method locates the center line of the structure through the maximum variation path of the brightness gradient; the axis crossing point refers to the specific pixel or voxel position crossed by the symmetry axis in the two-dimensional or three-dimensional image, which is usually near the center of the renal portal vein and has the maximum gradient change and structural symmetry significance; the plane feature refers to the set of multi-dimensional structural parameters such as average gray level, texture direction, and voxel density of the kidney region image in the rotating section centered on the axis, and each plane generates a set of features to describe anatomical consistency; the dimension change direction refers to the change trend direction of various image feature dimensions such as gray level, texture direction, and voxel density between consecutive image frames, i.e., whether it is rising, falling, or remaining stable, which is used to judge the stability or mutation trend of feature evolution; the key point sequence refers to the set of image nodes related to abnormal deviation marked in the kidney image track, which usually includes index time points of gray level change extreme points, path turning points, and structure mutation regions, which are used for time sequence coincidence judgment with the lag fragment of biomarkers; the voxel density refers to the compactness or local aggregation degree of gray level distribution of voxels in a unit volume in the kidney region in medical images (such as CT or MRI), which is used to reflect the texture consistency and distribution rule of tissue structure in space.

[0068] As shown in Figure 2 , the acquisition steps of the dynamic trend linkage feature are specifically:

[0069] S101: Based on the high-risk population of acute kidney injury, analyze the monitoring sequence of serum creatinine, cystatin C and IL-18 obtained by continuous monitoring, compare the change direction of the detection results of adjacent time nodes in turn, mark the trend of relative increase or decrease, and get the directional transfer label sequence;

[0070] The actual detection value sequence of serum creatinine, cystatin C and interleukin 18 of the patient at continuous time points is obtained, the sampling frequency should ensure that the interval time is within 12 hours, a time sequence is established for each index, the detection results of adjacent time points are compared item by item, it is judged whether the value at the current time point is greater than that at the previous time point, if the value rises, it is recorded as "rise", if it falls, it is recorded as "fall", taking serum creatinine as an example, if the previous two monitoring values are 89 and 96, it is judged as rising, if the subsequent value is 93, it is judged as falling, and the complete direction sequence can be formed in this way, the same method is used to obtain the direction markers of cystatin C and interleukin 18, the duration span of each change segment needs to be calculated, and a minimum change amplitude threshold is set combining the conventional fluctuation range of the index, only when the detection value changes more than the amplitude, it will be considered as a real direction shift, if the value changes insufficiently, the change at the time point will not be counted in the trend analysis, to avoid misjudging normal fluctuation as pathological change, in implementation, the standard deviation of the same type of people in the past month can be selected as the basis for setting the threshold, for example, if the standard deviation is 8 units, the threshold can be set to 1.2 units, that is, when the value between two time points changes less than 1.2 units, it is not treated as a trend change, the duration of the direction also needs to be clearly defined, for example, it is set that the same direction must appear for three consecutive time points to be considered as a trend segment, if only one or two time points change direction, it is not included in the subsequent trend analysis, in practice, if the serum creatinine values are 85, 87, 89, 88, 90 and 92 in turn, the directions are rising, rising, falling, rising and rising, according to the minimum duration requirement, the middle single point falling is excluded, only two consecutive rising trends are retained, and the direction shift sequence of the three indexes is generated.

[0071] S102: Compare the current time slice in the direction shift marker sequence with the adjacent slice in the historical monitoring sequence, align the direction markers according to the time points, judge the consistency, analyze whether the direction in each corresponding relationship maintains the same trend, and count the proportion of time nodes with consistent direction in the total slice to obtain the time consistency proportion;

[0072] Extract the latest time segment as a control standard, compare it with the same length sequence in the historical monitoring record segment by segment, set a sliding window for the historical sequence, align it with the current segment one by one according to the time point, and judge whether the direction marks of the two sequences are consistent at each position. If the same, it is recorded as matching, if different, it is considered as not matching. The proportion of the number of matching positions in the whole segment length is defined as the consistency ratio. The ratio can be divided into three levels. The ratio is less than four as low consistency, four to seven as medium consistency, and more than seven as high consistency. Taking the actual serum creatinine direction sequence as an example, if the current is the rise, rise, fall, fall, rise of the last five time points, and the history is rise, fall, fall, rise, rise, there are three positions consistent between them, which is considered as medium consistency. After processing each marker respectively, the weight average method of the consistency ratio of the three indexes needs to be set, and the comprehensive consistency ratio is calculated according to the weight average method. The weight of each index can be set according to its reference value in the early prediction in the clinic, for example, serum creatinine can be set as five weight, cystatin C three weight, and interleukin 18 two weight. Repeat the operation in all historical segments to complete the complete statistical result of the consistency ratio between the current time segment and all historical comparison segments.

[0073] S103: Screen the segments with time consistency ratio meeting the basic judgment requirement and the ratio meeting the linkage standard, analyze whether the change direction in the segments meeting the conditions is consistent, output the monitoring time range with common trend direction, and obtain the dynamic trend linkage feature;

[0074] Screen the consistency ratio statistical result, select the segments with ratio higher than seven and duration more than six hours as candidate regions, then verify the direction change in the candidate segments, judge whether the direction remains unchanged in the whole duration, if the direction is reversed in the middle, the segment needs to be excluded, if only one time point is reversed and restored to the original trend and maintained for more than two time points, it is considered to meet the direction continuity requirement, which ensures that the identified trend segment is stable. After executing each marker respectively, integrate the start time and end time of the segments meeting the conditions. If more than half of the time segments of multiple indexes overlap, it is considered as a linkage trend segment. For example, if serum creatinine shows a trend segment from 8am to 2pm, and cystatin C shows a trend segment from 10am to 4pm, the overlapping part is four hours, which accounts for 67% of the shorter time, meeting the integration condition. Integrate the segments meeting the overlap requirement in the three indexes into a unified time segment, and output the time segment as the detection result of the dynamic trend linkage.

[0075] As shown in Figure 3 , the acquisition step of the rotation plane anatomical feature group is specifically:

[0076] S201: According to the dynamic trend linkage feature, the approximate axis distribution of the renal portal region in the gray image is located, the symmetric intensity distribution of the two sides of the renal portal structure in the gray image is analyzed, the center trend of the symmetric axis is optimized, the distribution position of the axis crossing point in the gray gradient change region is adjusted, and symmetric path distribution information is obtained;

[0077] The trend linkage interval identified in the current time period of the patient is determined, the abdominal image sequence corresponding to the interval is called, the approximate spatial region of the renal portal vein in the image is located, the gray level change of the renal portal peripheral tissue and the background region is compared through the image gray level distribution feature, the approximate trend direction of the renal portal vein is preliminarily judged, the renal portal region in the image is divided into two symmetric regions, the corresponding gray value set of each region is extracted, the mean and variance of the gray value of each pixel point in each region are calculated, and the gray mean difference and the symmetric degree of the gray distribution between the two regions are calculated. The symmetric degree is defined as the symmetric region, and the error of the gray distribution is less than 10% of the total gray dynamic range, which is considered to be good symmetry. If the error exceeds 30%, it is determined to be asymmetric. The set axis position is adjusted to offset, the two side regions are redefined after each offset, and the gray mean difference is recalculated. The change trend of the mean difference between the current and previous results is compared. If the change gradually decreases, it means that the optimal axis direction is approached. The position of the crossing point is continuously fine-tuned, and the fine-tuning unit is set to 1 pixel length. The symmetric region gray variance ratio is calculated after each adjustment. If the ratio gradually approaches 1, it means that the symmetry degree is enhanced. The minimum error point is selected as the optimal crossing point of the current image, and the symmetric path axis in the image is precisely distributed and positioned to obtain the symmetric path distribution information of the image frame.

[0078] S202: Call the symmetric path distribution information, construct a set of rotation planes based on the axis, analyze the lateral variation amplitude of the gray level distribution in the kidney region in each plane, calculate the dynamic distribution sequence of the gray difference on both sides of the axis under multiple angle rotations, judge whether the gray feature remains in a continuous transition state during rotation, and obtain the angle gray variation trend;

[0079] A set of rotation planes is established with the axis as the center, the rotation planes are set at equal angle intervals with the axis as the reference, the rotation angle interval is set to 5 degrees, and a total of 36 rotation angle planes are generated from 0 degrees to 180 degrees. Each rotation plane intercepts a respective symmetrical kidney region image region on the left and right of the axis. The gray value mean of each pixel row in the transverse direction of each plane is extracted and arranged in sequence along the transverse direction. The gray value mean difference between the left and right sides in each plane is calculated, and a gray value difference sequence is generated in the order of the 36 angles. In the sequence, the change amount of the gray value difference between adjacent angles is compared in sequence. If the change amplitude is less than 5% of the full gray value range of the image, it is considered to be a continuous transition. If the change amplitude exceeds 15% of the image gray value dynamic range, it is considered to be discontinuous. On this basis, the number of angles with gray value mutations is counted. If it exceeds one-third of the total number of angles, it is determined that the image has a discontinuous feature in the current rotation state. The corresponding rotation plane is marked as a non-continuous plane. Otherwise, it is recorded as a continuous plane. The gray value difference change results of all rotation planes are summarized and sorted by angle to obtain the angle gray value change trend.

[0080] S203: According to the angle gray value change trend, the arrangement direction of the kidney region texture in each plane is compared with the spatial position relationship of the voxel aggregation region in the image. It is judged whether the structure distribution in the multi-angle image section has a common feature. The spatial combination of gray value, texture and density is integrated to obtain a rotation plane anatomical feature group.

[0081] According to the angle gray value change trend, the texture structure features inside the kidney region in each rotation plane are extracted to obtain the main direction of the texture arrangement. By traversing the local region texture in each rotation angle plane, the main direction vector of the texture in each region is extracted. The included angle distribution between the texture direction vector and the rotation angle is calculated. If the included angle between the texture direction and the rotation direction in most planes remains within ±15 degrees, it is determined that the direction consistency is strong. Meanwhile, in combination with the positioning information of the voxel aggregation region in the image, the connected region with a gray value higher than the average of the whole image in each plane is extracted as the voxel aggregation region. The center point coordinates are calculated and positioned in the three-dimensional coordinate system. If the voxel aggregation center points in most rotation planes are distributed symmetrically on both sides of the axis or the distribution radius changes less than 5 pixels, it is determined that the spatial aggregation is consistent. On this basis, the gray value features, texture directions and voxel distribution are combined. If more than 70% of the planes have consistent directions, approximate distribution centers and continuous gray value changes, the rotation plane with structural consistency is recorded. All rotation planes that meet the conditions are combined to obtain a rotation plane anatomical feature group.

[0082] As shown in Figure 4 , the steps of obtaining the structural continuity abnormality data are specifically:

[0083] S301: According to the rotation plane anatomical feature group, analyze the gray scale, texture direction and density change trend of the continuous image frames in the kidney area range, compare whether the spatial direction is consistent between time nodes, identify whether there is a direction reversal or offset feature, and obtain a joint direction change trajectory sequence;

[0084] Call the labeled kidney image area in each rotation angle plane, extract the kidney gray mean value, main texture direction and local density aggregation degree of each frame under the corresponding rotation angle in the continuous image frames, by traversing each frame image kidney pixel block, performing horizontal and vertical statistical calculation on the gray distribution, recording the mean value change trajectory in the same area, calculating the angle value corresponding to the main texture direction in each image frame, comparing the change amount of the texture direction angle of the previous and next frames, and extracting the center point position of the highest density area to judge whether the spatial aggregation area moves, the extraction unit of all indicators is set to every 5 frames, in the continuous time nodes, the change direction of gray mean value, texture direction angle and voxel aggregation center position is recorded, whether it changes with time is judged, if the change direction of a certain index is opposite to the previous time point, it is considered as direction reversal, if the previous time point is upward trend and the current direction is downward trend, it is considered as reversal, if the spatial change path of the index deviates from the original direction more than 5 pixels of the original path, it is considered as direction deviation, all indexes with reversal or deviation events will be marked as direction mutation, and the change type and time index value when the event occurs will be recorded, the mutation events of gray, texture and density three types of indexes will be sorted into joint direction change trajectory sequence in time sequence.

[0085] S302: Call the joint direction change trajectory sequence, filter the image segments with discontinuous direction on the gray, texture and density path, judge whether the position offset in the corresponding image frame is concentrated in the target area, mark the corresponding image coordinate section and its position distribution, and obtain the local space disturbance area.

[0086] All image frames marked as having direction reversal or offset are screened, and the corresponding image position coordinates in each frame are extracted. Based on the kidney region mask boundary of the frame, the distribution areas of pixels with abrupt changes in grayscale, texture, and density are marked on the image. The spatial clustering degree of the abruptly changed pixels is statistically analyzed, and the distance from the centroid of the abruptly changed point to the center point of the kidney region is calculated. If it is greater than 30% of the kidney region radius, it is considered an edge abrupt change; otherwise, it is a central clustered abrupt change. If more than 70% of the abruptly changed points in the same image frame are concentrated in the central region, it is determined to be a concentrated feature of the target region perturbation. The image frame number of such images is recorded, and the coordinate range of the abruptly changed points is bounded into the smallest rectangle. The pixel coordinates of its upper left and lower right corners are marked to describe the two-dimensional boundary range of the perturbation region. If two or more consecutive image frames have overlapping perturbation regions, they are recorded as a continuous perturbation interval. The grayscale range, texture direction change range, and density change slope corresponding to the perturbation region in each image frame are extracted as additional data for further analysis. The spatial anomaly coordinate information and additional statistical features of each frame are integrated to obtain the local spatial perturbation region.

[0087] S303: Based on the local spatial disturbance area, determine whether there is an abnormal trajectory of the structural axis within the target area, analyze whether the offset behavior exceeds the spatial tolerance range, and verify the structural offset distribution range and deviation conditions to obtain structural continuity anomaly data.

[0088] To analyze whether the offset behavior exceeds the spatial tolerance range, the formula is:

[0089]

[0090] Calculate the offset eigenvalue D dev The method determines whether there are trajectory anomalies in the structural axis within the target area, where n represents the number of structural control points. This represents the actual measured coordinates of the i-th structural control point in space. This represents the design coordinates of the i-th structural control point in space. d represents the mean offset of all structural control points. tol The tolerance limit threshold represents the structural offset.

[0091] Suppose the number of consecutive frames in the image sequence is n=6. To save image processing time, the symmetrical grayscale axis points after scanning with a rotating plane are extracted. The image grayscale reference coordinates at time point T0 are set as The reference point is calculated from the center of the renal portal vein region in the initial frame, and the value is obtained by extracting the gray-weighted center of the renal cortex region after image segmentation.

[0092] Grayscale coordinate quantization uses a standard range of 0–255. The center point is calculated jointly using the centroid of the contour and the gradient field. The data is as follows:

[0093]

[0094] All refer to time points

[0095] Perform the gray offset item calculation:

[0096]

[0097] Substitute the average value item:

[0098]

[0099] The rotation plane slice fluctuation parameter is calculated as follows:

[0100] The spatial standard deviation of the rotation cross-section texture direction angle gradient change is extracted from each frame, and the main direction difference after the texture direction tensor transformation is normalized to calculate the following offset (unit: mm):

[0101] d1=1.2, d2=0.8, d3=1.0, d4=1.6, d5=2.3, d6=1.5;

[0102] Take the average:

[0103]

[0104] Let the symmetry axis smooth threshold value d sym =1.0, which is the maximum tolerance of the average symmetry variation limit in the medical image literature by extracting the texture gradient direction distribution from the standard renal portal cross-section image sample. Substitute the formula:

[0105] D dev =6.47+(1.40-1.00)=6.47+0.40=6.87;

[0106] The results show that the gray scale change in the structure continuous image deviates greatly from the rotation plane texture density change, and the value is greater than D dev <4.5 control line, indicating that there is a significant deviation in the positioning of the renal portal vein symmetry axis in the continuous frame. This value is directly used as the output basis for structure continuity abnormal data and is involved in subsequent calculations.

[0107] As Figure 5 shown, the steps for obtaining the multi-modal collaborative offset marker are as follows:

[0108] S401: According to the structural continuity abnormal data, the gray gravity center coordinate distribution of the renal cortex region in the continuous image frame is analyzed, the horizontal and vertical translation path of the gravity center position between each image frame is calculated, whether the path forms a forward arrangement or a position jump in the image plane is judged, and the image sequence displacement trend is obtained;

[0109] Firstly, the specific frame sequence number of the abnormal region in the continuous image frame and the mask boundary of the renal cortex region are determined, the image region where the renal cortex is located is intercepted in each frame, the gray weighted average position of all pixel points in the region is calculated, the gray gravity center coordinates corresponding to the frame are obtained, the coordinates are recorded as a position point in the two-dimensional image plane, and the operation is sequentially performed on all image frames in time sequence to form a time sequence coordinate set of the gray gravity center. The distance difference of the gravity center coordinates between adjacent frames on the horizontal and vertical axes is calculated in turn, and the translation direction of each frame gravity center relative to the previous frame is recorded. If the gravity center coordinates of the current frame and the previous frame change to positive in the horizontal direction and the right movement is continuous for three frames, it is judged as forward right movement. If the gravity center suddenly jumps from right to left in a frame, it is recorded as horizontal jump. If the same direction reversal appears in the vertical movement trend, it is also recorded as vertical jump. The jump definition standard is that the translation direction of the current frame is opposite to the direction of the previous two frames, and the single frame moving distance is more than twice the average value of the previous three frames. Taking the actual image data as an example, if the gray gravity center Y coordinates of frame 1 to frame 3 are 200, 204 and 208 respectively, and frame 4 is 196, frame 4 has a direction reversal compared with frame 3 and a movement amplitude of 12 pixels, which is much larger than twice the average value of 4 pixels, which meets the jump standard. Finally, all the jump frames and the continuous translation frames are grouped and recorded to generate the moving path of the image frame gravity center in the image space, and the time start and end frame numbers of each direction consistent segment and jump segment are identified to obtain the image sequence displacement trend.

[0110] S402: Based on the image sequence displacement trend, the numerical change direction of creatinine, urea nitrogen and body fluid parameters in the same time period is compared, whether the continuous trend of the parameter sequence deviates from the direction of the image trajectory is judged, the combination item forming the inconsistent trend is identified, and the parameter trajectory offset coefficient is obtained;

[0111] Whether the continuous trend of the parameter sequence deviates from the direction of the image trajectory is judged by the formula:

[0112]

[0113] The combination item forming the inconsistent trend is identified, and the parameter trajectory offset coefficient ΔSY is obtained dev , wherein, represents the difference between the z-th moment creatinine, urea nitrogen or body fluid parameter and the previous moment, represents the change value of the gray center trajectory in the main axis direction in the z-th frame image, the direction vector change value representing the z-th moment, the difference between the physiological parameter value representing the z-th moment and the average of the physiological parameter values of the previous three consecutive moments, the standard deviation of the gray scale distribution of the z-th frame of the rotating planar image in the renal cortex region, n SY is the total number of image frames in the selected image time period, each frame corresponding to a sampling moment, so z essentially represents the corresponding relationship between the frame and the moment, for example, frame z = 1 represents moment 1;

[0114] Suppose the image sequence analysis frame number is 3 frames, i.e. n SY = 3, each parameter is obtained as follows:

[0115] is the difference in creatinine concentration at the z-th frame corresponding moment compared to the previous moment, the creatinine values detected by the high-frequency biochemical automatic analyzer are 1.0 mg / dL for the first frame, 1.15 mg / dL for the second frame, and 1.3 mg / dL for the third frame, then:

[0116] The first frame difference is 0, the second frame: 1.15-1.0 = 0.15, the third frame: 1.3-1.15 = 0.15;

[0117] is the difference in the projection displacement of the gray scale center of mass main axis direction, the center displacement values obtained by fitting the main axis of the image sequence are: 0 for the first frame, 2 pixels for the second frame, and 3 pixels for the third frame, then:

[0118] The first frame is 0, the second frame is 2-0 = 2, and the third frame is 3-2 = 1;

[0119] is the rotation main axis direction change rate in the image sequence, which is obtained according to the variation ratio of the main axis angle of the renal portal vein, and after normalization, it is as follows:

[0120] The first frame is 0.2, the second frame is 0.3, and the third frame is 0.25;

[0121] is the deviation of the current moment physiological value from the average of the previous three frames, which is calculated by the sequence mean value:

[0122] The first frame has no data, recorded as 0, the second frame: 1.15- average (1.0) = 0.15, the third frame: 1.3- average (1.0, 1.15) = 1.3-1.075 = 0.225;

[0123] is the gray scale standard deviation, which calculates the fluctuation range of the pixel intensity distribution in the renal cortex region, and the result is:

[0124] Frame 1 is 0.30, frame 2 is 0.35, frame 3 is 0.40;

[0125] Calculate each frame item respectively:

[0126] Frame 1:

[0127] Frame 2:

[0128] Frame 3:

[0129] Total calculation:

[0130]

[0131] The results show that the current monitoring of three frames of images and their corresponding creatinine change trend exist moderate amplitude trend deviation, the deviation coefficient is 0.2075, the value represents that the physiological and image linkage trend exists synchronous deviation, which meets the linkage abnormality screening condition, and can be included in the multi-modal deviation time period marker calculation.

[0132] S403: Based on the parameter trajectory deviation coefficient, screen the time sequence segment of linkage abnormality in the deviation coefficient fluctuation section, judge whether the segment constitutes a co-occurrence structure between image space and detection parameters, and identify the spatial coordinates and data interval of the time period, and obtain the multi-modal collaborative deviation marker;

[0133] According to the time sequence analysis of the change curve, the position section of the sudden increase or sudden decrease of the deviation coefficient is identified, the time segment whose coefficient change exceeds the average of the previous section by more than 20% is extracted, and it is determined as a fluctuation section. In the fluctuation section, the image displacement trajectory in the corresponding image frame is cross-compared with the parameter direction consistency, the segment with directional deviation in the image space and parameter fluctuation is screened, if the image gray center jumps and the parameter deviation coefficient is in a high fluctuation state at the same time, the time period is defined as a linkage abnormality segment. Further, the center coordinates of the corresponding frame in the image are extracted, and the motion path formed by the coordinates in the two-dimensional plane is recorded. The frame number corresponding to the mutation point in the path is marked as a spatial abnormal frame. At the same time, the creatinine, urea nitrogen and body fluid values in the parameter sequence corresponding to the frame time point are extracted, which are marked together with the values of the previous and next time points as a parameter abnormality segment. The image coordinate range of all image jump frames and the parameter abnormal value interval are uniformly corresponding, and are identified as a group of multi-modal linkage abnormality segments. The start and end time, image frame number, image coordinate center point and corresponding creatinine parameter change value of each linkage abnormality segment are output, and the multi-modal collaborative deviation marker is obtained.

[0134] As shown in Figure 6 , the steps of obtaining the early collaborative abnormality locking result are specifically:

[0135] S501: Based on the multi-modal collaborative offset label, analyze the positioning information of each time period, compare the time label with the time index of the lagging segment and image key point in the dynamic trend linkage feature, filter the segment range that occurs intersection in time dimension, and obtain the synchronous overlapping time period index set;

[0136] Extract the time period corresponding to each set of linkage abnormal label, record its start and end frame number, read the coordinate index information of each abnormal segment in the image space, sort the time label into a list, and compare it with the time label of the lagging segment in the dynamic trend linkage feature one by one, merge all overlapping time intervals, judge whether each linkage abnormal time period is included in the lagging segment time range, if partially overlapping, record the segment as an overlapping candidate, index extract the image key point time sequence, get the time stamp of all key point events, compare whether there is time point intersection with the preceding candidate overlapping time period, if the image key point and the collaborative offset segment are completely consistent at a certain time frame, record it as accurate overlap, if the image key point is located between the start and end frame of the collaborative offset time period, record it as range overlap, finally, all time segments with any of the above intersection forms are summarized to generate a complete synchronous overlapping time period index set, which contains the matching results of linkage abnormal time period index, lagging trend time period index and image key point time label, and the items in the set are arranged in ascending order by time frame number, to obtain the synchronous overlapping time period index set.

[0137] S502: Call the synchronous overlapping time period index set, calculate the coverage degree of the overlapping segment in the image key point sequence, judge whether the coverage ratio reaches the linkage recognition reference range, recognize the continuous segment sequence with strong continuity, and mark the segment boundary index to obtain the coincidence segment coverage ratio.

[0138] Perform coverage analysis operation on each synchronous segment, extract the time label of all key points from the image key point sequence, compare the start and end time frames of the current overlapping segment, count the number of key points appearing in the segment, divide the number by the time length of the overlapping segment, and calculate the key point coverage ratio. If the coverage ratio exceeds 50%, the overlapping segment is marked as a high coverage segment, and if it is less than 30%, it is marked as a low coverage segment. Further, the continuity of all high coverage segments is detected to determine whether the frame index in the image sequence is continuous. If the start and end frames of adjacent high coverage segments are separated by no more than two frames, they are considered continuous and merged into a longer segment, and the new start and end indexes are recorded. At the same time, the total number of key points in the merged segment and the proportion of the key points in the image key point sequence are counted. If the proportion exceeds 60% of the total number of image key point sequences, it is marked as a key segment. The start and end frame number, segment length, key point number and coverage ratio of all key segments are marked to generate a coincidence segment coverage ratio data table.

[0139] S503: According to the coincidence fragment coverage ratio, it is judged whether the linkage range between the image track and the physiological index exists a concentrated deviation, the time nodes showing consistent linkage reaction in the continuous fragment are screened, and the time range and key feature annotation information are integrated to obtain the early collaborative abnormality locking result;

[0140] The linkage between the image track and the physiological index is judged, the image gravity track, texture direction change sequence and density change sequence of each high-coverage key fragment are read, and the direction change sequence of creatinine, urea nitrogen and body fluid parameters is extracted in the same time range. The dominant direction trend of the image index is paired with the direction trend of each physiological parameter. If the directions of the image and any two parameters are consistent, it is recorded as a consistent linkage point. If the directions of the three parameters are all consistent, it is recorded as a strong consistent linkage point. The proportion of the number of consistent linkage points in each key fragment is calculated. If the proportion exceeds 70% of the total frame number, it is marked as a high-consistent fragment. The distribution of consistent points between continuous time frames is screened in all high-consistent fragments. If more than five consecutive frames appear consistent points, it is marked as a concentrated linkage section. The start and end frame numbers, image space gravity coordinates and corresponding parameter values of the concentrated linkage section are written into the integration table. The center point and feature point position of the image corresponding to the time frame are marked on the key point image. The structured result composed of the image time section range, key point coordinates and parameter value combination of multiple linkage indicators is output. The early collaborative abnormality locking result is obtained.

[0141] As shown in Figure 7 , the AKI recognition system combining biomarkers and image features includes:

[0142] The trend linkage identification module analyzes creatinine, cystatin C and IL-18 sequence based on high-risk population of acute kidney injury, calculates change direction, matches lag mode, screens trend consistent fragments, judges physiological trend continuity, and obtains dynamic trend linkage feature.

[0143] The symmetrical structure extraction module optimizes the symmetry axis of the renal portal vein path, constructs a rotation plane, analyzes the differences in gray, texture and density, judges the structural consistency, and obtains the rotation plane anatomical feature group according to the dynamic trend linkage feature.

[0144] The image variation detection module compares the gray, texture and density changes between image frames, judges the direction consistency, analyzes the fluctuation amplitude, identifies the mutation area and deviation trend, and obtains the structure continuity abnormal data according to the rotation plane anatomical feature group.

[0145] The linkage deviation calculation module calculates the renal cortex gray track, analyzes the moving direction, compares the creatinine, urea nitrogen and body fluid trend, judges the linkage deviation, screens the corresponding abnormal fragments, and obtains the multi-modal collaborative deviation marker according to the structure continuity abnormal data.

[0146] The abnormal timing locking module judges whether the recognition condition is met based on the multi-modal collaborative offset label, compares the lagging segment with the image track key node, screens the time overlap range, calculates the coverage ratio, and obtains an early collaborative abnormality locking result.

[0147] It should be understood that the term "and / or" herein only describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist simultaneously, and B exists alone, wherein A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects, but it can also represent an "and / or" relationship, which can be understood according to the context before and after.

[0148] In the present application, "at least one" means one or more, and "a plurality of" means two or more. "At least one of the following" or the like means any combination of these items, including any combination of single item or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, wherein a, b, and c can be single or multiple.

[0149] It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0150] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0151] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-mentioned devices, apparatuses and units can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0152] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the above-described apparatus embodiments are merely schematic, and for example, the division of the units is merely a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another device, or some features can be omitted or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0153] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0154] In addition, the functional units in each embodiment of the present application can be integrated into a processing unit, or each unit can be physically present separately, or two or more units can be integrated into one unit.

[0155] If the functions are realized in the form of software functional units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0156] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An AKI identification method combining biomarkers and image features, characterized in that, The method includes: S1: Based on high-risk populations for acute kidney injury, we analyzed the sequences of creatinine, cystatin C, and IL-18, calculated the direction of change, matched hysteresis patterns, screened fragments with consistent trends, judged the continuity of physiological trends, and obtained dynamic trend linkage characteristics. S2: Based on the dynamic trend linkage characteristics, optimize the symmetry axis of the renal portal vein path, construct a rotational plane, analyze the differences in grayscale, texture and density, judge the structural consistency, and obtain the anatomical feature group of the rotational plane. S3: Based on the rotating plane anatomical feature group, compare the changes in grayscale, texture and density between image frames, determine the consistency of direction, analyze the fluctuation amplitude, identify abrupt change regions and offset trends, and obtain structural continuity anomaly data; S4: Based on the structural continuity anomaly data, calculate the gray trajectory of the renal cortex, analyze the direction of movement, compare the trends of creatinine, blood urea nitrogen and body fluid, determine the linkage deviation, screen the corresponding abnormal segments, and obtain the multimodal collaborative offset marker. S5: Based on the multimodal collaborative offset marker, compare the lagging segments with key nodes of the image trajectory, filter the time overlap range, calculate the coverage ratio, determine whether the recognition conditions are met, and obtain the early collaborative anomaly locking result.

2. The AKI identification method combining biomarkers and image features according to claim 1, characterized in that, The dynamic trend linkage features include concentration direction symbols, trend sequence numbers, and lag matching identifiers; the rotational plane anatomical feature group includes symmetry axis grayscale baseline, rotational surface texture distribution, and voxel aggregation degree; the structural continuity anomaly data includes parameter fluctuation boundaries, structural deflection dimensions, and spatial mutation labels; the multimodal collaborative migration markers include image translation paths, physiological fluctuation trajectories, and linkage difference regions; and the early collaborative anomaly locking results include anomaly synchronization segments, cross-warning segments, and early warning status information.

3. The AKI identification method combining biomarkers and image features according to claim 1, characterized in that, The specific steps for obtaining the dynamic trend linkage feature are as follows: S101: Based on high-risk populations for acute kidney injury, the serum creatinine, cystatin C and IL-18 monitoring sequences obtained by continuous monitoring were analyzed. The direction of change of the test results at adjacent time points was compared in turn, and the trend of relative increase or decrease was marked to obtain the direction transfer marker sequence. S102: Compare the current time segment in the direction transfer mark sequence with the adjacent segments in the historical monitoring sequence, align the direction marks according to the time points, make a consistency judgment, analyze whether the directions in each pair of corresponding relationships maintain the same trend, and count the proportion of time nodes with consistent directions in the total segments to obtain the time consistency ratio. S103: Filter the segments whose duration meets the basic judgment requirements and whose proportion meets the linkage standard among the time consistency ratios, analyze whether the change direction of the segments that meet the conditions is consistent, output the monitoring time range with common trend direction, and obtain the dynamic trend linkage characteristics.

4. The AKI identification method combining biomarkers and image features according to claim 1, characterized in that, The specific steps for obtaining the rotational plane anatomical feature set are as follows: S201: Based on the dynamic trend linkage characteristics, locate the approximate axis distribution of the renal portal vein region in the grayscale image, analyze the symmetrical intensity distribution of the regions on both sides of the renal hilum structure in the grayscale image, optimize the central direction of the symmetrical axis, adjust the distribution position of the axis crossing point in the grayscale gradient change region, and obtain the symmetrical path distribution information. S202: Call the symmetrical path distribution information to construct a set of rotating planes based on the axis, analyze the lateral variation of the gray distribution of the kidney region in each plane, calculate the dynamic distribution sequence of the gray difference on both sides of the axis under multi-angle rotation, determine whether the gray features maintain a continuous transition state during the rotation process, and obtain the angle gray change trend. S203: Based on the gray-scale change trend of the angle, compare the arrangement direction of the kidney texture in each plane with the spatial position relationship of the voxel aggregation area in the image, determine whether there are common features in the structural distribution in the multi-angle image cross-section, integrate the spatial combination of gray-scale, texture and density, and obtain the rotating plane anatomical feature group.

5. The AKI identification method combining biomarkers and image features according to claim 1, characterized in that, The specific steps for obtaining the structural continuity anomaly data are as follows: S301: Based on the rotating plane anatomical feature group, analyze the grayscale, texture direction and density change trends of continuous image frames within the kidney region, compare whether the spatial orientation between time nodes is consistent, identify whether there are direction reversal or offset features, and obtain a joint change trajectory sequence. S302: Call the joint directional trajectory sequence, filter image segments with discontinuous directions on grayscale, texture and density paths, determine whether the positional offset in the corresponding image frame is concentrated in the target area, mark the corresponding image coordinate segment and its positional distribution, and obtain the local spatial disturbance area. S303: Based on the local spatial disturbance area, determine whether there is an abnormal trajectory of the structural axis within the target area, analyze whether the offset behavior exceeds the spatial tolerance range, and verify the structural offset distribution range and deviation conditions to obtain structural continuity anomaly data.

6. The AKI identification method combining biomarkers and image features according to claim 5, characterized in that, The analysis of whether the offset behavior exceeds the spatial tolerance range is performed using the following formula: Calculate the offset eigenvalue D dev The method determines whether there are trajectory anomalies on the structural axis within the target area, where n represents the number of structural control points. This represents the actual measured coordinates of the i-th structural control point in space. This represents the design coordinates of the i-th structural control point in space. d represents the mean offset of all structural control points. tol The tolerance limit threshold represents the structural offset.

7. The AKI identification method combining biomarkers and image features according to claim 1, characterized in that, The specific steps for obtaining the multimodal cooperative offset marker are as follows: S401: Based on the structural continuity anomaly data, analyze the gray-level centroid coordinate distribution of the renal cortex region in consecutive image frames, calculate the horizontal and vertical translation paths of the centroid position between each image frame, determine whether the paths form a forward arrangement or a position jump in the image plane, and obtain the displacement trend of the image sequence. S402: Based on the displacement trend of the image sequence, compare the direction of change of creatinine, blood urea nitrogen and body fluid parameters in the same time period, determine whether the continuous trend of the parameter sequence deviates from the direction of the image trajectory, identify the combination of inconsistent trends, and obtain the parameter trajectory offset coefficient. S403: Based on the parameter trajectory offset coefficient, filter out time segments with linkage anomalies in the deviation coefficient fluctuation range, determine whether the segment constitutes a co-occurrence structure between the image space and the detection parameters, and identify the spatial coordinates and data intervals of this type of time segment to obtain multimodal collaborative offset markers.

8. The AKI identification method combining biomarkers and image features according to claim 7, characterized in that, The determination of whether the continuous trend of the parameter sequence deviates from the direction of the image trajectory is made using the following formula: Identify combinations of terms with inconsistent trends to obtain the parameter trajectory offset coefficient ΔSY dev , where Δ This represents the difference between creatinine, blood urea nitrogen, or body fluid parameters at time z and the previous time. This represents the change in the grayscale center trajectory along the principal axis in the z-th frame image. This represents the change in the direction vector at time z. This represents the difference between the physiological parameter value at time z and the average value of the physiological parameter at the previous three consecutive times. The standard deviation of the grayscale distribution in the renal cortex region of the z-th frame of the rotated planar image is given by n. SY It represents the total number of image frames within the selected image time period.

9. The AKI identification method combining biomarkers and image features according to claim 1, characterized in that, The specific steps for obtaining the early collaborative anomaly locking results are as follows: S501: Based on the multimodal collaborative offset marker, analyze the positioning information of each time period, compare its time label with the time index of the lagging segment and image key point in the dynamic trend linkage feature, filter the segment range that intersects in the time dimension, and obtain the synchronous overlapping time period index set. S502: Call the synchronous overlapping time period index set, calculate the coverage degree of overlapping segments in the image key point sequence, determine whether the coverage ratio reaches the linkage recognition benchmark range, identify the segment sequence with strong continuity, mark the segment boundary index, and obtain the overlapping segment coverage ratio. S503: Based on the overlapping segment coverage ratio, determine whether there is a concentrated shift in the linkage range between the image trajectory and physiological indicators, filter out time nodes that show consistent linkage response within continuous segments, and integrate time range and key feature annotation information to obtain early collaborative anomaly locking results.

10. An AKI identification system combining biomarkers and image features, said system being used to implement the AKI identification method combining biomarkers and image features as described in any one of claims 1-9, characterized in that, The system includes: The trend linkage identification module is based on high-risk groups of acute kidney injury. It analyzes the sequences of creatinine, cystatin C and IL-18, calculates the direction of change, matches the lag pattern, screens fragments with consistent trends, judges the continuity of physiological trends, and obtains dynamic trend linkage features. The symmetry structure extraction module optimizes the symmetry axis of the renal portal vein path based on the dynamic trend linkage features, constructs a rotation plane, analyzes the differences in grayscale, texture and density, judges the structural consistency, and obtains the rotation plane anatomical feature group. The image variation detection module compares the changes in grayscale, texture and density between image frames based on the rotating plane anatomical feature group, determines the consistency of direction, analyzes the fluctuation amplitude, identifies abrupt change regions and offset trends, and obtains structural continuity anomaly data. The linkage offset calculation module calculates the gray trajectory of the renal cortex based on the structural continuity anomaly data, analyzes the direction of movement, compares the trends of creatinine, blood urea nitrogen and body fluid, judges the linkage deviation, filters the corresponding abnormal segments, and obtains multimodal collaborative offset markers. The abnormal timing locking module, based on the multimodal collaborative offset marker, compares the lagging segments with key nodes of the image trajectory, filters the time overlap range, calculates the coverage ratio, determines whether the recognition conditions are met, and obtains the early collaborative abnormal locking result.

Images