Detection method based on reference fetus umbilical cord blood flow ultrasonic image data

Abstract

The invention discloses an ultrasonic image data detection method based on referenced fetal umbilical cord blood flow, which comprises the following steps of: constructing a three-dimensional dynamic image of the fetal umbilical cord blood flow through multi-angle ultrasonic imaging, and accurately capturing the blood flow direction and speed change in combination with a neural network model; and the problems of inaccurate blood flow direction identification and lagging speed distribution analysis caused by two-dimensional imaging fuzziness and large manual judgment error are avoided. Modeling is carried out on the blood flow dynamic characteristics through a preset blood flow anomaly judgment algorithm, the parameters of the blood flow parameter matrix are automatically extracted, and the recognition blind area of complex blood flow anomaly in the prior art is broken through. A risk assessment result is further automatically generated in combination with a fetus health database, real-time early warning is performed when serious abnormity is detected, a complete closed loop from blood flow monitoring to clinical intervention is constructed, and the problem that in the prior art, the umbilical cord blood flow direction, speed distribution, reverse blood flow and blood flow interruption cannot be accurately recognized, and clinical intervention is affected is solved. And early-stage fetal distress early warning is difficult to realize.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image detection, in particular to a fetal umbilical cord blood flow ultrasound image data detection method based on reference. BACKGROUND

[0002] Fetal umbilical cord blood flow ultrasound parameters and serum levels are not only factors affecting the intrauterine development and birth prognosis of congenital patent ductus arteriosus fetuses, but also can improve the diagnostic efficiency and have good diagnostic value when combined detection is performed.

[0003] In order to ensure that the fetal umbilical cord blood flow ultrasound parameters can be accurately obtained, the fetal umbilical cord blood flow ultrasound is checked by using an ultrasonic diagnostic instrument. However, only using the ultrasonic diagnostic instrument cannot accurately identify the direction of the umbilical cord blood flow, the speed distribution, and the reverse blood flow and blood flow interruption, and it is difficult to realize early fetal distress warning. SUMMARY

[0004] The purpose of the present application is to provide a fetal umbilical cord blood flow ultrasound image data detection method based on reference, to solve the technical problem that the prior art cannot accurately identify the direction of the umbilical cord blood flow, the speed distribution, and the reverse blood flow and blood flow interruption, and it is difficult to realize early fetal distress warning.

[0005] The technical solution of the present application is as follows: A fetal umbilical cord blood flow ultrasound image data detection method based on reference, comprising the following steps: Step S1, acquiring multi-angle dynamic image data of fetal umbilical cord blood flow by an ultrasonic device, generating a three-dimensional blood flow image sequence including blood flow direction information; Step S2, based on the three-dimensional blood flow image sequence, modeling the blood flow direction, speed gradient and abnormal fluctuation area by a neural network model, and extracting local speed distribution features to generate a blood flow parameter matrix; Step S3, identifying the reverse blood flow, blood flow interruption and abnormal speed distribution area in the blood flow parameter matrix by a preset blood flow abnormality determination algorithm, and outputting an abnormal event; Step S4, in response to the abnormal event, generating an umbilical cord blood flow abnormality risk assessment report using a fetal physiological parameter database, and triggering a distress warning signal when reverse blood flow or blood flow interruption is detected.

[0006] Further technical solutions are that the step S1 specifically comprises: Step S11, scanning the fetal umbilical cord region at different angles by the ultrasonic device, covering the full view of the umbilical cord trunk and branch blood vessels, recording the dynamic changes of blood flow speed, direction and blood vessel shape at different angles, and forming a plurality of two-dimensional ultrasonic image sequences; Step S12, blood flow motion trajectories are identified for each set of the two-dimensional image sequence using an edge detection algorithm, and blood flow direction information is marked by color coding to generate two-dimensional blood flow images with direction annotations; Step S13, multiple sets of the two-dimensional images are aligned in a spatial coordinate system, and interpolation algorithms are used to fill in the visual angle blind area, while integrating into three-dimensional blood flow image data containing blood flow direction information; Step S14, based on the three-dimensional blood flow image data, a continuous three-dimensional blood flow image sequence is spliced in time sequence.

[0007] Further technical solutions are that the step S12 specifically includes: Step S121, contrast adjustment is performed on each set of two-dimensional image sequence frame by frame, and a standby image is generated by local area brightness difference analysis; Step S122, a multi-scale edge detection algorithm is used to concatenate edge information of the standby image to form a continuous trajectory path of blood flow motion; Step S123, dynamic color coding is assigned according to the motion direction of the continuous trajectory path.

[0008] Further technical solutions are that the step S2 specifically includes: Step S21, based on the three-dimensional blood flow image sequence, the blood flow dynamics in each grid cell are sampled for speed, and the peak value, average value and fluctuation range are recorded; Step S22, the neural network model is used to mark the direction of the blood flow trajectory in the grid cell, and a direction vector set is calculated; Step S23, the gradient of blood flow acceleration / deceleration characteristics is analyzed through the speed difference and time stamp synchronization technology of adjacent grid cells; Step S24, the grid cells with speed mutation or direction disorder are screened, and their coordinates, speed values, direction vectors and gradient data are integrated into a multi-dimensional blood flow parameter matrix.

[0009] Further technical solutions are that the step S22 specifically includes: Step S221, local speed distribution, direction change rate and motion continuity characteristics of blood flow dynamics in the grid cell are extracted frame by frame through a sliding window to form a multi-dimensional feature set; Step S222, the multi-dimensional feature set is input into a pre-trained neural network model in time sequence, and the initial direction classification result of each grid cell is output through a fully connected layer and an activation function; Step S223, according to the classification result, the grid cell is assigned a corresponding direction annotation label, and an annotation confidence value is recorded; Step S224, calculate the direction vector of each grid unit by combining the direction label and the velocity gradient data of adjacent grid units; Step S225, adjust the amplitude range of the direction vector and correct the global reference system of the direction angle according to the velocity distribution range of the three-dimensional blood flow image sequence, and generate the direction vector set.

[0010] A further technical solution is that the step S3 specifically comprises: Step S31, identify the region where the direction vector continuously reverses and the amplitude exceeds the set threshold by comparing the polarity of the direction vector in the blood flow parameter matrix with the preset reference direction, and mark it as a blood flow reverse region; Step S32, perform time series analysis on the velocity values of consecutive frames in the blood flow parameter matrix, and determine it as a blood flow interruption region when the velocity value is continuously below the minimum detection threshold for a set time; Step S33, compare the velocity amplitude in the blood flow parameter matrix with the preset threshold range, and screen out the region where the velocity amplitude exceeds the threshold range and the velocity change rate of the adjacent region is abnormal, and mark it as an abnormal velocity distribution region; Step S34, associate all abnormal regions of the steps S31 to S33 marked by time stamp and spatial coordinates, and generate abnormal events of abnormal type, occurrence time, location and duration.

[0011] A further technical solution is that the step S33 specifically comprises: Step S331, compare the velocity amplitude of each region in the blood flow parameter matrix with the threshold range based on historical data statistics, and record the velocity amplitude data that exceeds the range; Step S332, calculate the velocity difference rate of the velocity amplitude region that exceeds the threshold range and its adjacent region, screen out the region where the difference rate exceeds the preset change threshold, and mark it as an abnormal velocity distribution region.

[0012] A further technical solution is that the step S34 specifically comprises: Step S341, extract the corresponding time stamp data from the abnormal region, and record the specific frame number or time point where it occurs; Step S342, extract the two-dimensional or three-dimensional spatial coordinates of the abnormal region in the blood flow parameter matrix from the same abnormal region, and form a spatial position data set; Step S343, match the spatial coordinates of the abnormal region at the same time point by time stamp synchronization algorithm, and establish a time-space mapping relationship table; Step S344, assign a corresponding abnormal type label to each abnormal region; Step S345, for the same position, the duration of the abnormal area is calculated by the time stamp difference, and is recorded as the event duration; Step S346, the abnormal type, time stamp, spatial coordinates and duration are integrated according to the event logic to generate an abnormal event list including event attributes.

[0013] Further technical solutions are that the step S4 specifically comprises: Step S41, the abnormal type, occurrence time and position information recorded in the abnormal event are matched with historical blood flow data, fetal heart rate, amniotic fluid index and other parameters in the fetal physiological parameter database to extract relevant physiological indicators; Step S42, the physiological indicators matched are quantitatively analyzed by a weighting algorithm to calculate and generate an umbilical cord blood flow abnormality risk score; Step S43, the current blood flow parameter matrix is monitored, when a reverse blood flow or blood flow interruption event is detected, a fetal distress early warning signal is triggered through a preset communication protocol.

[0014] Further technical solutions are that the step S42 specifically comprises: Step S421, according to the clinical relevance of historical events in the fetal physiological parameter database, a dynamic weight coefficient is assigned to each matched physiological indicator; Step S422, each matched physiological indicator is numerically processed, the original value is multiplied by the dynamic weight coefficient to generate a weighted indicator score; Step S423, different dimensions of the weighted score are mapped to a unified score interval through a linear transformation algorithm; Step S424, the weighted score is integrated according to a preset formula to output a risk score reflecting the severity of umbilical cord blood flow abnormalities.

[0015] The beneficial effects of the present application are: By constructing a three-dimensional dynamic image of fetal umbilical cord blood flow through multi-angle ultrasonic imaging, and combining a neural network model to accurately capture blood flow direction and speed changes, the problems of inaccurate blood flow direction recognition and speed distribution analysis lag caused by two-dimensional imaging ambiguity and large artificial judgment errors are avoided. Through a preset blood flow abnormality determination algorithm, the blood flow dynamic characteristics are modeled, the blood flow parameter matrix parameters are automatically extracted, and the recognition blind area of the prior art for complex blood flow abnormalities is broken through. Further, the risk assessment results are automatically generated in combination with the fetal health database, and real-time warning is made when serious abnormalities are detected, a complete closed loop from blood flow monitoring to clinical intervention is constructed, and the technical problems that the prior art cannot accurately recognize the direction of umbilical cord blood flow, speed distribution, reverse blood flow and blood flow interruption, and early fetal distress warning cannot be realized are solved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flow chart of the detection method steps based on the fetal umbilical cord blood flow ultrasound image data provided by the present application is provided. Figure 2 A flow chart of step S1 in the detection method steps based on the fetal umbilical cord blood flow ultrasound image data provided by the present application is provided. Figure 3 A flow chart of step S2 in the detection method steps based on the fetal umbilical cord blood flow ultrasound image data provided by the present application is provided. Figure 4 A flow chart of step S3 in the detection method steps based on the fetal umbilical cord blood flow ultrasound image data provided by the present application is provided. Figure 5 A flow chart of step S4 in the detection method steps based on the fetal umbilical cord blood flow ultrasound image data provided by the present application is provided. DETAILED DESCRIPTION

[0017] In order to better understand the technical content of the present application, specific embodiments are provided below, and the present application is further described in conjunction with the accompanying drawings.

[0018] Reference Figures 1 to 5 , the present application provides a detection method based on fetal umbilical cord blood flow ultrasound image data, comprising the following steps: Step S1, acquiring multi-angle dynamic image data of fetal umbilical cord blood flow through an ultrasound device, generating a three-dimensional blood flow image sequence including blood flow direction information; Step S2, based on the three-dimensional blood flow image sequence, modeling the blood flow direction, velocity gradient and abnormal fluctuation area through a neural network model, and extracting local velocity distribution features to generate a blood flow parameter matrix; Step S3, identifying the blood flow reverse, blood flow interruption and abnormal velocity distribution area in the blood flow parameter matrix through a preset blood flow abnormality determination algorithm, and outputting an abnormal event; Step S4, in response to the abnormal event, generating an umbilical cord blood flow abnormality risk assessment report using a fetal physiological parameter database, and triggering a distress warning signal when reverse blood flow or blood flow interruption is detected.

[0019] It should be noted that the three-dimensional blood flow image sequence can be a set of stereoscopic images generated by dynamically capturing the motion trajectory of fetal umbilical cord blood flow in three-dimensional space through an ultrasound device and splicing in time sequence, which can clearly show the whole process of blood flow direction, velocity and abnormal changes; the blood flow parameter matrix can be a table form arranged by the key data of blood flow such as velocity, direction and fluctuation according to spatial position and time sequence, which is equivalent to recording the "motion state" of each blood flow with a digital table, facilitating computer rapid analysis of abnormal conditions.

[0020] In the embodiment of the present application, the multi-angle dynamic images of fetal umbilical cord blood flow are collected by the ultrasonic device, the blood flow motion trajectories under different viewing angles are spatially aligned and time-synchronized, a three-dimensional dynamic image sequence containing blood flow direction information is constructed, and the spatial distribution and motion direction of the umbilical cord blood flow are stereoscopically presented; on this basis, a neural network model is used to model the blood flow dynamic characteristics, automatically extract the blood flow velocity gradient change, local velocity distribution characteristics and abnormal fluctuation area, form a structured blood flow parameter matrix, and establish a quantitative description of the blood flow state; further, a preset abnormality determination algorithm is used to automatically identify the blood flow reverse, interruption and abnormal velocity distribution area in the parameter matrix, and output an abnormal event record with spatiotemporal positioning information; finally, in combination with the fetal physiological parameter database, the abnormal event is associated with clinical indicators such as fetal heart rate and amniotic fluid index for analysis, a risk assessment report is generated, and when key abnormalities such as blood flow reversal or interruption are detected, an alarm signal is triggered through a preset warning mechanism, thereby constructing a complete technical link covering blood flow monitoring, abnormality identification to clinical response.

[0021] Specifically, the three-dimensional dynamic images of fetal umbilical cord blood flow are constructed by multi-angle ultrasonic imaging, and the blood flow direction and velocity change are accurately captured by combining a neural network model, thereby avoiding the problems of inaccurate blood flow direction recognition and slow velocity distribution analysis caused by two-dimensional imaging ambiguity and large manual judgment errors. The blood flow dynamic characteristics are modeled by a preset blood flow abnormality determination algorithm, and the blood flow parameter matrix parameters are automatically extracted, thereby breaking through the recognition blind area of complex blood flow abnormalities in the prior art. Further, the risk assessment results are automatically generated in combination with the fetal health database, and real-time warnings are given when serious abnormalities are detected, thereby constructing a complete closed loop from blood flow monitoring to clinical intervention, and solving the technical problems that the prior art cannot accurately identify the direction, velocity distribution, reverse blood flow and blood flow interruption of the umbilical cord blood flow, and early fetal distress warning cannot be achieved.

[0022] As a further improvement of step S1, specifically comprising: Step S11, scanning the fetal umbilical cord region at different angles by the ultrasonic device to cover the full view of the umbilical cord trunk and branch vessels, recording the dynamic changes of blood flow velocity, direction and vessel shape under different angles, and forming a plurality of two-dimensional ultrasonic image sequences; Step S12, using an edge detection algorithm to identify the blood flow motion trajectory of each two-dimensional image sequence, and marking the blood flow direction information by color coding to generate two-dimensional blood flow images with direction annotation; Step S13, aligning the plurality of two-dimensional images according to the spatial coordinate system, filling in the viewing angle blind area by using an interpolation algorithm, and integrating into three-dimensional blood flow image data containing blood flow direction information; Step S14, based on the three-dimensional blood flow image data, sequentially splicing to generate a continuous three-dimensional blood flow image sequence.

[0023] In the embodiment of the present application, the umbilical cord region of the fetus is dynamically scanned by multi-angle ultrasonic imaging technology. Specifically, the blood flow dynamic images of the umbilical cord trunk and branch vessels are obtained from different perspectives by adjusting the azimuth and pitch angles of the ultrasonic probe. The vessel diameter, wall thickness, blood flow velocity, and path information under different perspectives are used to spatially align the multi-perspective data by image registration algorithm to construct a blood flow data set containing spatial position correlation. The collected two-dimensional images are analyzed frame by frame, the continuous path features of blood flow motion are extracted by image edge detection and trajectory tracking technology, and the blood flow direction is intuitively marked by color coding method with red representing antegrade and blue representing retrograde, forming a two-dimensional blood flow feature map with direction identification.

[0024] Subsequently, the X-Y-Z axis alignment method in the three-dimensional coordinate system is used to unify the two-dimensional image coordinates under different perspectives to the same reference system, and the image completion algorithm is combined to eliminate the missing areas of the images caused by perspective limitations. By layer-by-layer superimposition of the corrected two-dimensional image information, a three-dimensional volume data model of blood flow direction features is generated, in which each three-dimensional pixel stores blood flow velocity, direction, and timestamp parameters.

[0025] Finally, the three-dimensional data of adjacent frames are matched in the time dimension by the timestamp alignment algorithm to eliminate inter-frame motion artifacts and ensure that the three-dimensional image sequence can completely present the three-dimensional dynamic process of the umbilical cord blood flow over time.

[0026] Through multi-angle imaging and three-dimensional reconstruction technology, the blood flow direction recognition accuracy is improved, the umbilical cord blood vessel network is observed from all directions, and the detection ability of abnormal events such as blood flow reversal and interruption is enhanced through direction labeling technology. Combined with time axis dynamic analysis, a blood flow dynamic monitoring mode with high spatiotemporal resolution is constructed.

[0027] Preferably, the step S12 specifically performs steps comprising: Step S121, adjusting the contrast of each group of two-dimensional image sequences frame by frame, and generating a standby image through local area brightness difference analysis; Step S122, using a multi-scale edge detection algorithm to concatenate the edge information of the standby image to form a continuous trajectory path of blood flow motion; Step S123, assigning dynamic color coding according to the motion direction of the continuous trajectory path.

[0028] In the embodiment of the present application, the contrast of blood flow dynamics and background tissue is enhanced by dynamic contrast optimization of ultrasound images frame by frame, and an enhanced image is generated for subsequent analysis. Multi-level noise reduction is performed using Gaussian filters of different sizes, and hierarchical feature extraction is performed on the enhanced image, that is, the blood flow boundary profile is captured by small-scale edge detection, the blood vessel trunk morphology is identified by medium-scale detection, and the overall motion trend of blood flow is extracted by large-scale analysis. Based on the weighted average of regional confidence, a continuity description model of blood flow dynamic trajectory is constructed. Further, according to the spatial motion direction feature of the trajectory path, the direction angle value is mapped to the hue channel, and the speed amplitude value is mapped to the saturation channel. The continuous frames are color-coded according to the time sequence, such as red for antegrade direction and blue for retrograde direction. By eliminating the color step jump between adjacent frames, the visual expression of blood flow direction feature is finally realized.

[0029] As a further improvement of step S2, specifically comprising: Step S21, based on the three-dimensional blood flow image sequence, the velocity of the blood flow dynamics in each grid cell is sampled, and the peak value, average value and fluctuation range are recorded; Step S22, using a neural network model to label the direction of the blood flow trajectory in the grid cell, and calculating the direction vector set; Step S23, by the velocity difference of adjacent grid cells and the time stamp synchronization technology, the gradient of blood flow acceleration / deceleration characteristics is analyzed; Step S24, screening the grid cells with velocity mutation or direction disorder, and integrating the coordinates, velocity values, direction vectors and gradient data into a multi-dimensional blood flow parameter matrix.

[0030] In the embodiment of the present application, the three-dimensional blood flow image sequence is divided into grid cells by spatial discretization processing method, and the three-dimensional body data is divided into cubic cells according to equal intervals of X-Y-Z axes. The multi-dimensional velocity characteristics of the blood flow dynamics in each independent grid cell are collected. Specifically, the maximum velocity in each cell is calculated by using sliding window algorithm, the multi-frame velocity average is calculated by using time weighted average method, and the velocity stability is quantified by the difference between the maximum value and the minimum value, forming a velocity description system containing time sequence characteristics.

[0031] On this basis, a feature extraction model based on convolutional neural network is used to analyze the directionality of the spatial motion characteristics of the blood flow trajectory in the region. Specifically, the input features include the preprocessed blood flow trajectory image, the local spatial features are extracted by multiple convolutional layers, and the amplitude (indicating the speed size) and the direction angle (indicating the motion direction) of the direction vector are output by the fully connected layer, generating a direction feature set with vector characteristics.

[0032] Further, the velocity gradient between adjacent cells is quantified by spatial derivative calculation method, and the error caused by time asynchronization is eliminated by timestamp alignment algorithm to generate a spatial gradient matrix reflecting the acceleration / deceleration trend of blood flow.

[0033] Finally, the umbilical cord blood flow velocity, direction and dynamic change characteristics are quantitatively described with high precision, and the blood flow state analysis capability is improved. The velocity fluctuation analysis quantifies the stability by standard deviation calculation, the direction stability evaluation detects abnormal deviation by direction angle change rate, and the gradient distribution modeling reveals the blood flow dynamic trend by spatial derivative calculation. The three together constitute a multi-parameter for blood flow anomaly recognition.

[0034] Preferably, step S22 specifically comprises: Step S221, the local velocity distribution, direction change rate and motion continuity features of blood flow dynamics in the grid cell are extracted frame by frame through the sliding window to form a multi-dimensional feature set; Step S222, the multi-dimensional feature set is input into the pre-trained neural network model in time sequence, processed through the full connection layer and the activation function, and the preliminary direction classification result of each grid cell is output; Step S223, according to the classification result, the corresponding direction annotation label is assigned to the grid cell, and the annotation confidence value is recorded; Step S224, the direction vector of each grid cell is calculated in combination with the direction annotation label and the velocity gradient data of the adjacent grid cell; Step S225, according to the velocity distribution range of the three-dimensional blood flow image sequence, the amplitude range of the direction vector is adjusted, and the global reference system of the direction angle is corrected to generate the direction vector set.

[0035] In the implementation of the present application, a fixed-size cubic window is used to slide in the time dimension, and the blood flow velocity distribution, direction change rate and motion continuity features in each window are quantitatively analyzed to construct a multi-dimensional feature set containing time stamp and spatial coordinate label.

[0036] Then the feature set is input into the multi-layer convolutional neural network architecture in time sequence, and the features are extracted through the hierarchical processing mechanism: the first convolutional layer extracts the local velocity gradient feature, the second convolutional layer performs direction feature fusion, and finally the full connection layer outputs the preliminary direction classification result, such as antegrade, retrograde or no obvious direction. The classification result generates corresponding classification confidence, forming a direction annotation system with classification confidence.

[0037] Each grid cell is taken as a node, and an edge weight is established between adjacent cells, and the weight value is calculated by the direction consistency index and the velocity gradient similarity. The amplitude of the direction vector is calculated by the weighted average of the velocity and the confidence, and the direction angle is determined by the majority voting method.

[0038] The Z-score standardization method of the global speed distribution range is used to standardize the amplitude range of the direction vector, that is, the speed value is mapped to the interval [0, 1], and the reference benchmark of the direction angle is unified, for example, the positive direction of the X-axis of the image coordinate system is 0°, and the modulus 360° normalization processing is adopted, so as to finally form a direction vector set containing three-dimensional coordinates, amplitude, direction angle, time stamp and confidence field for each vector, thereby providing a standardized data basis for the quantitative analysis of the blood flow direction feature.

[0039] As a further improvement of step S3, specifically comprising: Step S31, by comparing the polarity of the direction vector in the blood flow parameter matrix with the preset reference direction, identifying the region where the direction vector is continuously reversed and the amplitude exceeds the set threshold, and marking it as a blood flow reverse region; Step S32, performing time series analysis on the speed values of the continuous frames in the blood flow parameter matrix, and when the speed value is continuously lower than the minimum detection threshold for a set time, determining it as a blood flow interruption region; Step S33, comparing the speed amplitude in the blood flow parameter matrix with the preset threshold range, and screening out the region where the speed amplitude exceeds the threshold range and the speed change rate of the adjacent region is abnormal, and marking it as an abnormal speed distribution region; Step S34, associating all the abnormal regions of steps S31 to S33 according to the time stamp and spatial coordinates, and generating an abnormal event of abnormal type, occurrence time, position and duration.

[0040] In the embodiment of the application, based on the direction vector characteristics in the blood flow parameter matrix, the polarity comparison strategy is used to determine the direction angle: the direction vector of each grid element is calculated with the preset X-axis positive direction, when the included angle exceeds 180° and the duration exceeds 3 continuous frames, and the speed amplitude exceeds the 95% quantile based on the historical data statistics, it is marked as a blood flow reverse region.

[0041] Then the sliding time window (5 frame length) is used to calculate the average speed in the window, and compared with the preset minimum detection speed 80%, when the average speed in the window is continuously lower than the critical value and the duration exceeds 2 seconds, it is determined as a blood flow interruption event.

[0042] At the same time, the speed amplitude of the current grid element is compared with the average speed based on the historical data statistics ± 3 times the standard deviation, and the candidate region exceeding the range is screened out; secondly, the speed gradient of the candidate region and the adjacent element is calculated, when the absolute value of the gradient exceeds 2 times the average gradient of the preset speed threshold, and the duration exceeds the set window length, it is finally determined as an abnormal speed distribution region.

[0043] Finally, by mapping the recording time stamp and spatial coordinates of each abnormal region, data correlation is achieved by constructing the time stamp, coordinate X / Y / Z, and abnormal type fields. Further, clustering analysis is performed on abnormal events occurring continuously at the same spatial location, and the event duration is calculated, finally generating an event record system containing abnormal type, occurrence time, spatial location, and duration.

[0044] By storing event information, complete data support is provided for the quantitative analysis of abnormal blood flow states. The abnormal type field uses enumeration coding (reverse blood flow = 1, blood flow interruption = 2, velocity anomaly = 3), the time stamp field uses the ISO 8601 standard time format, the spatial coordinate field contains the voxel unit conversion relationship (1 voxel = 0.1 mm), and the duration field is in milliseconds, ensuring data traceability and clinical interoperability.

[0045] Preferably, step S33 specifically includes: Step S331, comparing the velocity amplitude of each region in the blood flow parameter matrix with the threshold range based on historical data statistics, and recording the velocity amplitude data that exceeds the range; Step S332, calculating the velocity difference rate of the region whose velocity amplitude exceeds the threshold range with the adjacent region, screening out the region whose difference rate exceeds the preset change threshold, and marking it as an abnormal velocity distribution region.

[0046] In the embodiments of the present application, based on the historical blood flow velocity database, i.e. the blood flow dynamic data under normal and abnormal states, the upper and lower threshold boundaries are generated by calculating the dynamic mean of the historical data and combining the standard deviation amplification coefficient (such as 3σ), forming a threshold range with time adaptability.

[0047] The velocity value of each grid cell is compared with the dynamic upper threshold (μ+3σ) and lower threshold (μ-3σ) point by point, and when the velocity amplitude exceeds any threshold range and the duration exceeds the preset window (more than 3 consecutive frames), it is marked as a candidate abnormal region. This process is achieved by a frame-by-frame traversal algorithm, and a candidate region list containing space-time coordinates is output.

[0048] Then, a sliding window neighborhood analysis method is used to calculate the velocity difference of the 8 adjacent grid cells of each candidate region, and the velocity gradient change rate is quantified by finite difference method. Combined with the preset threshold of twice the historical gradient average value, the region whose gradient absolute value exceeds the threshold is selected as a significant velocity mutation candidate point.

[0049] Finally, the candidate region is analyzed for spatial continuity, adjacent grid cells that meet the gradient threshold condition are merged to form an abnormal region with spatial continuity. By constructing a region attribute table (including region ID, boundary coordinates, gradient mean value, etc.), the abnormal velocity distribution marker result is output, wherein each marker region contains the amplitude change rate (quantified by the maximum / minimum difference) and spatial expansion characteristics (evaluated by the area / perimeter ratio) of the velocity jump feature.

[0050] The method eliminates the misjudgment of static threshold to natural blood flow fluctuations through dynamic threshold modeling, enhances the sensitivity to local abnormalities through spatial gradient analysis, and finally ensures the clinical significance of the marker region through spatial continuity verification, forming an abnormal recognition system with rate characteristics and spatial distribution characteristics.

[0051] Preferably, step S34 specifically includes: Step S341, from the abnormal region, extract the corresponding timestamp data, record the specific frame number or time point when it occurs; Step S342, from the same abnormal region, extract its two-dimensional or three-dimensional spatial coordinates in the blood flow parameter matrix to form a spatial position data set; Step S343, through the timestamp synchronization algorithm, match the spatial coordinates of the abnormal region at the same time point to establish a time-space mapping relationship table; Step S344, assign a corresponding abnormal type label to each abnormal region; Step S345, for the abnormal regions occurring continuously at the same location, calculate the duration of the event by the timestamp difference and record it as the event duration; Step S346, integrate the abnormal type, timestamp, spatial coordinates and duration according to the event logic to generate an abnormal event list including event attributes.

[0052] In the implementation of the present application, the time information of the abnormal occurrence moment is obtained, and the accurate spatial position information of the abnormal region in the blood flow parameter matrix is extracted to construct a two-dimensional feature data set containing timestamp and spatial coordinates (X / Y / Z axis coordinate values).

[0053] The timestamp alignment algorithm is used to process data with different time resolutions, associate and map the timestamp of the abnormal event with the corresponding spatial coordinates, and construct a spatial mapping relationship table with time sequence characteristics. The table structure includes the fields: event ID, timestamp, X coordinate, Y coordinate, Z coordinate, and is stored in a relational database, supporting query by time interval or spatial region.

[0054] Each abnormal region is assigned a corresponding type identifier, and an enumeration coding strategy of reverse blood flow=1, blood flow interruption=2, velocity anomaly=3 is adopted to ensure that the correspondence between abnormal type and characteristic parameter is traceable.

[0055] Further, frame difference quantification duration is adopted, and a sliding window average method (with a window length of 5 frames) is combined to eliminate short-term fluctuations, to generate time-dimension quantification parameters.

[0056] Through the event structured integration module, attribute parameters such as abnormal type, timestamp, spatial coordinates and duration are data modeled according to event logic, and an abnormal event data set is constructed in XML or JSON format. Each event record contains fields: event ID, type code, occurrence time, duration (milliseconds), spatial coordinate range (minimum / maximum X / Y / Z coordinates), and through metadata, the version number of the blood flow parameter matrix and the detection device ID are marked, to ensure data traceability. The data set is connected to a risk assessment system through a standardized interface, to provide data support for subsequent clinical decision-making.

[0057] As a further improvement of step S4, specifically comprising: Step S41, the abnormal type, occurrence time and location information recorded in the abnormal event are matched with historical blood flow data, fetal heart rate, amniotic fluid index and other parameters in the fetal physiological parameter database, to extract relevant physiological indicators; Step S42, the physiological indicators matched are quantitatively analyzed through a weighted algorithm, to calculate and generate an umbilical cord blood flow abnormality risk score; Step S43, the current blood flow parameter matrix is monitored, and when a reverse blood flow or blood flow interruption event is detected, a fetal distress early warning signal is triggered through a preset communication protocol.

[0058] In the implementation of the present application, the type, time and spatial location information of the abnormal event are cross-modally matched with historical blood flow characteristics, fetal heart rate change trend (such as deceleration pattern) and amniotic fluid index in the fetal physiological parameter database, to screen spatio-temporal correlation indicators; then a dynamic weight distribution model is used to assign weight coefficients (such as blood flow impedance index weight 0.6, fetal heart rate weight 0.3) to the matched parameters, combined with Z-score standardization processing to generate a feature vector; a weighted summation algorithm is used to generate a risk score of 0-10 to quantify the severity of the abnormality; at the same time, a real-time monitoring system is constructed to scan the blood flow parameter matrix in a sliding window (window length 5 frames), and when reverse blood flow lasting >2 seconds and fetal heart rate decreasing >15 bpm are detected, a hierarchical early warning signal (red warning ≥8 points, yellow warning ≥4 points) is sent to the clinical terminal through the HL7 FHIR communication protocol, and a blockchain log containing event ID, spatial coordinates, weight distribution table and calculation process is recorded, to ensure data traceability and closed-loop management of clinical decision support.

[0059] Preferably, step S42 specifically comprises: Step S421, according to the clinical relevance of historical events in the fetal physiological parameter database, a dynamic weight coefficient is assigned to each matched physiological indicator; Step S422, numerical processing is performed on each matched physiological indicator, the original value is multiplied by the dynamic weight coefficient to generate a weighted indicator score; Step S423, by linear transformation algorithm, the weighted scores of different dimensions are mapped to a unified scoring interval; Step S424, the weighted scores are integrated according to a preset formula, and a risk score reflecting the severity of umbilical cord blood flow abnormalities is output.

[0060] In the implementation of the present application, the correlation strength of historical abnormal events and historical blood flow characteristics, fetal heart rate baseline fluctuation, amniotic fluid index and other physiological indicators is statistically modeled (r>0.7 is determined as strong correlation), and then adaptive weight coefficients are generated (historical blood flow characteristic weight 0.6±0.1, amniotic fluid index weight 0.2±0.05). After Z-score standardization of the original data, difference amplification weighted scores are generated by weight coefficient multiplication operation (historical blood flow characteristic weighted score=1.2×0.6=0.72); the minimum-maximum normalization is adopted to map the multi-parameter weighted scores to a unified scoring space of 0-1, and finally a comprehensive risk score is generated by linear weighted fusion algorithm (such as R=0.264), according to the clinical threshold, the risk level is divided (0-0.3 low risk / 0.3-0.7 medium risk / 0.7-1 high risk), and through 5-fold cross-validation, the area under the curve is ensured to be >0.9, the sensitivity is >90%, and the specificity is >85%, the explainability and traceability of clinical decision are realized.

[0061] The above is only a preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting fetal umbilical cord blood flow based on ultrasound image data, characterized in that, Includes the following steps: Step S1: Acquire multi-angle dynamic image data of fetal umbilical cord blood flow using ultrasound equipment, and generate a three-dimensional blood flow image sequence including blood flow direction information; Step S2: Based on the three-dimensional blood flow image sequence, the blood flow direction, velocity gradient and abnormal fluctuation area are modeled by a neural network model, and local velocity distribution features are extracted to generate a blood flow parameter matrix. Step S3: Using a preset blood flow anomaly detection algorithm, identify blood flow reversal, blood flow interruption, and abnormal velocity distribution areas in the blood flow parameter matrix, and output abnormal events. Step S4: In response to the abnormal event, generate an umbilical cord blood flow abnormality risk assessment report using the fetal physiological parameter database, and trigger a distress warning signal when reverse blood flow or blood flow interruption is detected.

2. The method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 1, characterized in that, Step S1 specifically includes: Step S11: Scan the fetal umbilical cord area from different angles using ultrasound equipment to cover the full view of the main trunk and branch vessels of the umbilical cord, record the dynamic changes in blood flow velocity, direction and vessel morphology at different angles, and form multiple sets of two-dimensional ultrasound image sequences. Step S12: Use an edge detection algorithm to identify the blood flow trajectory for each group of two-dimensional image sequences, and use color coding to mark the blood flow direction information to generate a two-dimensional blood flow image with direction annotation; Step S13: Align the multiple sets of two-dimensional images according to the spatial coordinate system, and use an interpolation algorithm to fill in the blind spots of the view, while integrating them into three-dimensional blood flow image data containing blood flow direction information; Step S14: Based on the three-dimensional blood flow image data, generate a continuous three-dimensional blood flow image sequence by splicing the images in a time sequence.

3. The method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 2, characterized in that, Step S12 specifically includes: Step S121: Adjust the contrast of each set of two-dimensional image sequences frame by frame, and generate backup images by analyzing the brightness difference in local areas; Step S122: Using a multi-scale edge detection algorithm, the edge information of the backup image is concatenated to form a continuous trajectory path of blood flow. Step S123: Assign dynamic color codes frame by frame according to the motion direction of the continuous trajectory path.

4. The method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 1, characterized in that, Step S2 specifically includes: Step S21: Based on the three-dimensional blood flow image sequence, the velocity of blood flow dynamics in each grid cell is sampled, and the peak value, average value and fluctuation range are recorded; Step S22: Use the neural network model to label the direction of the blood flow trajectory within the grid cell and calculate the direction vector set; Step S23: Analyze the gradient of blood flow acceleration / deceleration characteristics using the velocity difference and timestamp synchronization technology of adjacent grid cells; Step S24: Filter the grid cells with abrupt velocity changes or directional disorder, and integrate their coordinates, velocity values, direction vectors and gradient data into a multidimensional blood flow parameter matrix.

5. The method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 4, characterized in that, Step S22 specifically includes: Step S221: Extract the local velocity distribution, direction change rate, and motion continuity features of blood flow dynamics in the grid cell frame by frame through a sliding window to form a multi-dimensional feature set; Step S222: Input the multi-dimensional feature set into the pre-trained neural network model according to the time series, process it through the fully connected layer and activation function, and output the preliminary orientation classification result of each grid cell; Step S223: Based on the classification results, assign corresponding orientation labels to the grid cells and record the label confidence values; Step S224: Combine the orientation labels and velocity gradient data of adjacent grid cells to calculate the orientation vector of each grid cell; Step S225: Based on the velocity distribution range of the three-dimensional blood flow image sequence, adjust the amplitude range of the direction vector and correct the global reference frame of the direction angle to generate the direction vector set.

6. The method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 1, characterized in that, Step S3 specifically includes: Step S31: By comparing the polarity of the direction vector in the blood flow parameter matrix with the preset reference direction, identify the region where the direction vector is continuously reversed and the amplitude exceeds a set threshold, and mark it as a blood flow reverse region; Step S32: Perform time series analysis on the velocity values ​​of consecutive frames in the blood flow parameter matrix. When the velocity value is continuously lower than the minimum detection threshold for a set time, it is determined to be a blood flow interruption area. Step S33: Compare the velocity amplitude in the blood flow parameter matrix with a preset threshold range, filter out the regions where the velocity amplitude exceeds the threshold range and the velocity change rate of adjacent regions is abnormal, and mark them as abnormal velocity distribution regions. Step S34: Associate all marked abnormal regions from steps S31 to S33 with timestamps and spatial coordinates to generate abnormal events with abnormal type, occurrence time, location, and duration.

7. The method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 6, characterized in that, Step S33 specifically includes: Step S331: Compare the velocity amplitude of each region in the blood flow parameter matrix with the threshold range based on historical data statistics, and record the velocity amplitude data that exceeds the range; Step S332: For the velocity amplitude region that exceeds the threshold range, calculate its velocity difference rate with the adjacent region, filter out the region whose difference rate exceeds the preset change threshold, and mark it as an abnormal velocity distribution region.

8. The method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 7, characterized in that, Step S34 specifically includes: Step S341: Extract the corresponding timestamp data from the abnormal area and record the specific frame number or time point in which it occurred; Step S342: Extract the two-dimensional or three-dimensional spatial coordinates of the abnormal region in the blood flow parameter matrix to form a spatial location dataset; Step S343: Using a timestamp synchronization algorithm, match the spatial coordinates of the abnormal areas at the same time point and establish a time-space mapping relationship table; Step S344: Assign a corresponding anomaly type label to each of the anomaly regions; Step S345: For the abnormal areas that occur consecutively at the same location, calculate their duration using the timestamp difference and record it as the event duration; Step S346: Integrate the exception type, timestamp, spatial coordinates, and duration according to the event logic to generate an exception event list that includes event attributes.

9. A method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 1, characterized in that, Step S4 specifically includes: Step S41: Match the abnormal type, occurrence time and location information recorded in the abnormal event with historical blood flow data, fetal heart rate, amniotic fluid index and other parameters in the fetal physiological parameter database, and extract the relevant physiological indicators. Step S42: Quantitatively analyze the matched physiological indicators using a weighted algorithm to calculate and generate an umbilical cord blood flow abnormality risk score; Step S43: Monitor the current blood flow parameter matrix. When reverse blood flow or blood flow interruption event is detected, trigger a fetal distress warning signal through a preset communication protocol.

10. A method for detecting fetal umbilical cord blood flow based on ultrasound image data according to claim 9, characterized in that, Step S42 specifically includes: Step S421: Assign dynamic weight coefficients to each matched physiological indicator based on the clinical correlation of historical events in the fetal physiological parameter database; Step S422: Numericalize each matched physiological indicator by multiplying its original value by the dynamic weight coefficient to generate a weighted indicator score. Step S423: Map the weighted scores of different dimensions to a unified scoring range using a linear transformation algorithm; Step S424: Integrate the weighted scores according to a preset formula and output a risk score that reflects the severity of umbilical cord blood flow abnormalities.

Images