Ultrasonic multi-parameter time series data fusion and labor progress trend evaluation system

Abstract

The application discloses a system for fusion of ultrasonic multi-parameter time series data and assessment of labor progress trend during delivery, and relates to the technical field of digital data processing. The system comprises a coordinate construction module, a time series measurement module, an event grouping module, a time series fusion module and a trend assessment module. The coordinate construction module detects uterine contraction events, defines a labor period with adjacent uterine contraction events and gives an increasing period index to form a labor phase coordinate. The time series measurement module measures multiple labor parameters under the coordinate and deducts a phase offset according to a phase interval. The event grouping module groups milestone events and uterine contraction events co-occurring in a time window into a composite event. The time series fusion module performs bidirectional fusion of time series data according to a phase interval in segments with the composite event as an anchor point to obtain a fusion result with a period index as a progress axis. The trend assessment module outputs a trend state, a direction undetermined state or a parameter divergence state according to the fusion result, realizes deduction of uterine contraction phase offset and backtracking fusion of multi-parameter time series data.

Description

Technical Field

[0001] This invention relates to the field of digital data processing technology, specifically to a system for fusion of multi-parameter time-series data of intrapartum ultrasound and assessment of labor progress trends. Background Technology

[0002] Intrapartum ultrasound measures labor parameters such as fetal head progression angle, fetal head-perineal distance, fetal head-pubic symphysis distance, and midline angle of the brain through transperineal or transabdominal scanning, converting the fetal head descent process into measurable quantities. Existing intrapartum ultrasound-based techniques for assessing labor progress mostly rely on the automatic measurement of single-frame images or individual labor parameters, recording or fitting the results along a physical timeline. While this technology has been applied to the objective acquisition of labor parameters, it remains constrained by several unprocessed factors when assessing the trend of labor progress using time-series data of multiple labor parameters.

[0003] The values ​​of labor parameters are systematically affected by the phases of uterine contractions. The same labor parameter may be larger at the peak of contractions due to the forward thrusting of the fetal head and smaller at the intervals between contractions. If this phased component, fluctuating with contractions, is not separated, it will be included in the calculation of labor progress. To avoid this influence, current methods often only collect measurements at the intervals between contractions, discarding measurements at the peak of contractions, resulting in underutilization of data. Furthermore, if measurements from different phases are used together, the phase-induced bias will be mixed into the slope estimation of the progress trend, leading to an inaccurate representation of progress.

[0004] The representation of labor progress depends on the selection of the progress axis. Existing partograms use physical time as the horizontal axis. However, the frequency of uterine contractions varies among different women, and the number of effective uterine contractions corresponding to the same physical time span also varies. Therefore, the progress rate obtained by using physical time as the axis lacks comparability among different women, making individualized interpretation difficult.

[0005] Information between multiple parameters and between different time periods is not integrated or traced back. Existing solutions either process single frames and single parameters separately or directly map multiple labor parameters to a single conclusion, without integrating multi-parameter time-series data at the event and state levels. Furthermore, the time-series processing proceeds in a single forward direction, and the biased measurements caused by insufficient image quality in the early stages of labor continue to have an impact thereafter. There is a lack of ways to revise existing estimates after obtaining high-confidence evidence.

[0006] The assessment output is limited to a single conclusion. Existing systems provide a definitive determination of the mode of delivery or stage of labor. However, when multiple labor parameters point to inconsistencies or the evidence is insufficient, the output is still compressed into a single conclusion. This fails to express uncertainties and parameter conflicts beyond the trend of progress, and also fails to compensate for the decline in the imaging quality of one labor parameter by other labor parameters.

[0007] Therefore, there is an urgent need for a labor progress trend assessment system that organizes data based on the inherent rhythm of labor contractions, models the phases of contractions, uses the number of effective labor forces as the progress axis, and uses events as anchors to retrospectively fuse multi-parameter time-series data to output multiple states. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides a system for fusing multi-parameter time-series data of intrapartum ultrasound and assessing the progress of labor, comprising a coordinate construction module, a time-series measurement module, an event grouping module, a time-series fusion module, and a trend assessment module.

[0009] The coordinate construction module is used to acquire uterine contraction signals and ultrasound image sequences during labor, detect uterine contraction events from the uterine contraction signals, take the interval defined by two adjacent uterine contraction events as the labor cycle, assign an increasing cycle index to each labor cycle according to the order of occurrence, assign a phase value representing the relative position of the sampling time within the labor cycle, and construct the labor phase coordinates by the cycle index and the phase value.

[0010] The time-series measurement module is used to measure multiple labor parameters of the ultrasound image sequence under the labor phase coordinates, and to perform phase normalization on each of the labor parameters according to the phase interval to which the phase value belongs, so as to obtain phase-normalized time-series data.

[0011] The event grouping module is used to detect labor milestone events in the ultrasound image sequence, assign the labor milestone events to the labor phase coordinates corresponding to their occurrence time, and group the labor milestone events and the uterine contraction events that co-occur within a preset time window into composite events.

[0012] The time-series fusion module is used to perform bidirectional fusion of the phase-normalized time-series data according to the phase interval, using the composite event as the anchor point, through forward recursion and backward propagation, to obtain a fusion result with the cycle index as the axis of labor progress.

[0013] The trend assessment module is used to generate and output the labor progress trend state based on the fusion result and the directional relationship and confidence of each of the time series data under the labor phase coordinates. The labor progress trend state is one of the following: trend state, direction indeterminate state, and parameter divergence state.

[0014] Compared with the prior art, the present invention has at least the following beneficial effects:

[0015] Firstly, because the timing measurement module estimates the phase offset and subtracts it from the measured value under the phase coordinates of labor, the measured values ​​of the peak phase of uterine contractions and the inter-contraction phase can be in the same comparison domain. The estimation of the slope of the progress trend no longer takes into account the offset caused by the uterine contraction phase, and the sampling of the peak phase of uterine contractions and the inter-contraction phase are included in the estimation, so the measurement data are fully utilized.

[0016] Secondly, since the coordinate construction module uses the uterine contraction cycle index instead of physical time as the progress axis, and the progress rate is determined by the number of effective labor forces per unit, mothers with different uterine contraction frequencies are placed on the same progress axis and are comparable; accordingly, labor stagnation can be determined by the fact that the net progress of the fetal head is less than the preset lower limit of progress within a number of consecutive labor cycles (the number of which is preset), and this determination does not change with the frequency of uterine contractions.

[0017] Third, because the time series fusion module uses composite events as anchor points and makes the process equations segmented according to phase intervals, multi-parameter time series data are fused under the same latent state; composite events, as high-confidence evidence, trigger backpropagation and correct the estimates obtained before the anchor point based on low-confidence observations. Early deviations in labor no longer accumulate along the progress axis, unlike the situation where early deviations cannot be corrected when processing in a single direction.

[0018] Fourth, since the trend assessment module uses confidence-weighted process consistency judgment and outputs trend state, direction indeterminate state and parameter divergence state, uncertainties and parameter conflicts beyond the progress trend can be explicitly expressed; when the imaging quality of a certain labor process parameter decreases, the assessment is maintained by extrapolation from other labor process parameters, and the output is no longer limited to a single conclusion. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the architecture of the intrapartum ultrasound multi-parameter time-series data fusion and labor progress trend assessment system provided in an embodiment of the present invention.

[0020] Figure 2 The flowchart illustrates the working steps of the intrapartum ultrasound multi-parameter time-series data fusion and labor progress trend assessment system provided in this embodiment of the invention. Detailed Implementation

[0021] The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0022] Combination Figure 1The system described in this invention processes data starting from the acquisition of intrapartum data. During labor, ultrasound images arrive continuously over physical time, while uterine contractions occur intermittently. If labor parameters obtained from the images are processed directly along the physical time axis, there is no comparability between mothers with different contraction frequencies, and the measurements of contraction phases and intermittent phases are mixed. Therefore, this system uses the intrinsic rhythm of labor—uterine contractions—rather than physical time as the benchmark for organizing data: first, labor is divided into several labor cycles by contraction events, and labor phase coordinates are constructed so that subsequent parameter measurements, event detection, and fusion assessment are all performed under these coordinates. Accordingly, the acquired intrapartum data is incorporated into the labor phase coordinates by the coordinate construction module, and then, under these coordinates, the time-series measurement module obtains continuous labor parameters, and the event grouping module obtains discrete labor events; the two types of data are merged into the time-series fusion module, and fused into a single fetal head descent process under the anchoring of events; the trend assessment module generates a labor progress trend status based on this, and feeds back this status and the detected events to the acquisition stage to adjust the acquisition frequency. Thus, data forms a closed loop between acquisition, measurement, event detection, fusion, evaluation, and feedback. The following explanations of each module follow the data flow sequence.

[0023] Coordinate Construction Module. This module performs coordinate construction, which reorganizes the continuous labor data arriving with physical time into labor phase coordinates with the number of effective uterine contractions as the progress axis, according to the inherent rhythm of labor. This provides a unified and comparable time reference for subsequent modules. Processing begins from the acquisition start. Two signals arrive simultaneously during labor: image sequences obtained from transperineal or transabdominal ultrasound, and uterine contraction signals from uterine contraction pressure sensors or uterine electromyography. To stably identify each uterine contraction from the contraction signals, the envelope energy and first-order difference are calculated. The envelope energy is obtained by rectifying and low-pass filtering the contraction signal, which is a well-known signal processing method in the art. This embodiment does not limit the window length and filtering parameters. A contraction is determined to have started only when the envelope energy exceeds the baseline threshold and its first-order difference exceeds the slope threshold, so that the identification of the start and end of the contraction is not affected by baseline drift. For each detected contraction, its start, peak, and end times are recorded to form a contraction event sequence. The start, peak, and end times of each contraction are recorded as the start time, peak time, and end time of that contraction, respectively. The periodic exponent is a positive integer, and the first... The first labor cycle and the first The next contraction corresponds to the next uterine contraction.

[0024] A labor cycle is defined as the interval between the onset times of two adjacent contractions. An increasing cycle index is assigned sequentially, and the number of effective contractions experienced becomes the progress axis. The phase value at a sampling moment within the labor cycle is determined according to equation (1), which characterizes the corresponding stage of that contraction at that moment:

[0025]

[0026] In formula (1): Sampling time The phase value is not less than zero and is less than one; Sampling time; It is a cyclical index; , The first The onset and peak times of uterine contractions corresponding to each labor cycle; For the first The onset time of uterine contractions corresponding to each labor cycle; The peak phase positioning constant has a value greater than zero and less than one. An example value is... This invention does not limit the scope of the invention.

[0027] Equation (1) uses the peak time as the boundary to linearly normalize the ascending and descending segments respectively, so that the peak times of contractions of different durations are mapped to the same phase value. Thus, the same phase value corresponds to the same mechanical stage within the contraction during different labor cycles. When a contraction's peak time is equal to its starting time due to a recording abnormality, or the starting time of the next contraction is equal to the peak time of this contraction, making the denominator of Equation (1) zero, the phase value of the corresponding phase segment within that labor cycle is taken as its interval endpoint value. That is, the ascending segment is taken as zero, and the descending segment is taken as the value of the peak phase positioning constant. The labor cycle is then given a low weight or suspended for manual confirmation to avoid abnormal labor cycles affecting subsequent estimations. The labor phase coordinates are composed of the cycle index and the phase value, which run through the entire system.

[0028] Further, mark the peak contraction phase interval and the inter-contraction phase interval on the phase axis: The peak contraction phase interval is a closed interval from the peak phase positioning constant minus the peak phase half-width to the peak phase positioning constant plus the peak phase half-width; the inter-contraction phase interval is a left-closed, right-open interval from 1 minus the inter-contraction phase width to 1. The peak phase half-width is greater than zero and less than the smaller of the peak phase positioning constant and 1 minus the peak phase positioning constant. An example value is... The value of the intermittent phase width is greater than zero and less than one minus the peak phase positioning constant. An example value is... This invention does not limit the above values. Thus, the module outputs a data stream with the labor phase coordinates and their respective phase intervals for each sampling time.

[0029] The timing measurement module performs timing measurements and phase normalization. Its function is to convert ultrasound images cycle-by-cycle into timing data of multiple labor parameters under the labor phase coordinates established by the coordinate construction module. During the conversion, it subtracts the systematic offset introduced by the contraction phase, ensuring that the measurements of the contraction peak phase and the contraction interval phase are grouped into the same comparison domain. This module measures several labor parameters under the labor phase coordinates. The number of labor parameters is an integer not less than two. These parameters are selected from the fetal head progression angle, fetal head-perineal distance, fetal head-pubic symphysis distance, and cerebral midline angle. Automatic measurement is achieved using section recognition, structural segmentation, and geometric calculation. This embodiment does not limit the measurement method. Specifically, the positions of anatomical structures such as the fetal head contour, pubic symphysis, and perineum can be obtained through structural segmentation, and then calculated according to the geometric definitions of each labor stage parameter. For example, the fetal head progression angle is calculated as the angle between the long axis of the pubic symphysis and the tangent to the fetal head contour, and the fetal head-perineum distance is calculated as the distance from the perineum to the lowest point of the fetal head contour. The above-mentioned structural segmentation and geometric measurement are existing methods in this field, and those skilled in the art can use existing intrapartum ultrasound automatic measurement methods to achieve this, and output the structural segmentation confidence of the images on which each measurement is based. The measured values ​​of all labor stage parameters at each labor stage phase coordinate are decomposed according to equation (2):

[0030]

[0031] In formula (2): For the first Individual labor parameters at a cycle index of Phase value The measured value at the location; For the first The labor parameters at the first The actual progress component of each labor cycle varies only with the cycle index; For the first Phase offset of each labor parameter as the phase value changes; Measurement noise with a mean of zero; Index the labor parameters and retrieve all labor parameters. It is a cyclical index; This is the phase value.

[0032] Equation (2) separates the true process component, which changes only with the progress of labor, from the phase bias, which changes only with the phase, in the model. The true process component does not change with the phase within the same labor cycle, so the difference between the measured values ​​at the peak phase of contractions and the inter-contraction phase within the same labor cycle for the same mother can eliminate the true process component while retaining the phase bias. The phase bias is modeled as a piecewise constant with the peak phase bias at the peak phase of contractions and zero at the inter-contraction phase. Based on the inter-contraction phase, the peak phase bias is estimated as the median of the following set of paired differences, i.e., each peak phase of contractions and each inter-contraction phase within the same labor cycle are sampled to form a pair, and the median of all paired differences is taken, as shown in Equation (3):

[0033]

[0034] In formula (3): For the first Phase offset estimates of individual labor parameters at the peak of uterine contractions; The median calculation is used to suppress outliers in individual pairings; , The first Each labor parameter is in the same... Measurements taken at the peak of uterine contractions and at the intervals between contractions during each labor cycle. The phase value within the peak phase interval of uterine contractions. The phase value within the intermittent phase interval of uterine contractions; , These are the peak phase interval of uterine contractions and the intermittent phase interval of uterine contractions, respectively. To estimate the total number of labor cycles, only those labor cycles that simultaneously include both peak contraction phase sampling and intermittent contraction phase sampling are included.

[0035] One implementation method models the phase bias as two constants at the peak of contractions and the inter-contraction phase. Another implementation method further subdivides the phase interval into multiple sub-intervals, estimating the phase bias of each sub-interval using the median of the paired differences. This approximates the continuous change of phase bias with phase using piecewise constants; the finer the sub-interval division, the more accurate the approximation. Based on this, the measured value at the peak of contractions is subtracted from the estimated phase bias, while the measured value at the inter-contraction phase is not subtracted, resulting in the phase-normalized measured value. (The text then repeats the first instance.) The median of all phase-normalized measurements within a single labor cycle is the representative value of that labor parameter for that cycle; the absolute value of the phase offset characterizes the degree of uterine contraction pressure on the presenting part and is used as an auxiliary quantity to characterize the degree of uterine contraction pressure for reference output, but is not included in the aforementioned labor parameter.

[0036] Since the parameters of each production process differ in dimension, range of values, and direction of increase or decrease with progress, this module determines the monotonic calibration of the representative value to the equivalent fraction of the process for each production process parameter, and maps the representative value to the equivalent fraction of the process with the same range of values ​​and increasing with progress according to equation (4):

[0037]

[0038] In equation (4): For the first The labor parameters at the first The equivalent fraction of the process of each labor cycle, which takes a value greater than zero and less than one; For the first The labor parameters at the first The phase-normalized representative value of a labor cycle, that is, the median of all phase-normalized measurements of the labor parameter within the labor cycle. For the first The calibration slope coefficients of each labor process parameter are set to positive values ​​for labor process parameters that increase with progress and negative values ​​for labor process parameters that decrease with progress, so that the process equivalent score increases monotonically with the progress of labor. For the first A calibration center for individual labor parameters; It is a natural exponential function; It is a cyclical index; This is an index for labor process parameters.

[0039] Equation (4) is a logistic monotonic mapping. The calibration slope coefficient and calibration center are determined by the reference value range of the labor parameter under the known labor stage. The monotonic calibration of each labor parameter makes its process equivalent score take similar values ​​under the same labor stage, so it can be approximated as the observation of the same fetal head descent process. When the correspondence between a certain labor parameter and the fetal head descent process deviates from this approximation, its process equivalent score will deviate from the other parameters. This deviation is quantified by the process consistency described later, and the parameter divergence state is output when the deviation is significant, so as to identify and isolate the observation mismatch of individual parameters. After calibration, multiple labor parameters with different directions and dimensions are unified into process equivalent scores with the same direction and the same interval. Their time series data are sent to the time series fusion module as observations, and their dispersion is used by the trend assessment module to determine whether the parameters are divergent.

[0040] Event Grouping Module. This module performs event grouping, which, in parallel with temporal measurements, detects identifiable anatomical milestones in labor from the same ultrasound image sequence. It groups co-occurring milestones within a time window into high-confidence composite events, providing anchor points for trigger trajectory correction in temporal fusion. Detected labor milestones include at least one of fetal head engagement, completion of internal rotation, crowning, rupture of membranes, and fetal head remodeling. For each event, the detector outputs the event type, occurrence time, corresponding labor phase coordinates, and detection confidence level. Events with a detection confidence level below the confirmation threshold are suspended for manual confirmation. The detection of the labor milestone events can be achieved based on the structural segmentation results of the time-series measurement module: fetal head engagement and crowning can be determined by the relative position of the lowest point of the fetal head relative to the pubic symphysis or perineum exceeding a preset anatomical threshold; completion of internal rotation can be determined by the angle between the midline of the brain and the pelvis reaching a preset range; rupture of membranes can be determined by changes in the features of the amniotic cavity image; fetal head shaping change can be determined by the deformation of the fetal head contour exceeding a preset threshold; the detection confidence of each event is a synthesis of the corresponding structural segmentation confidence and decision margin; this embodiment does not limit the specific network structure and decision threshold of each event detector, and those skilled in the art can use conventional methods such as image segmentation, target detection, or classification to achieve this. To suppress false detections of single events due to image quality issues, this module does not directly trigger correction based on a single event. Instead, it groups co-occurring labor milestone events and uterine contraction events within a preset time window into composite events. The preset time window is measured in physical time and its duration is a preset value. Each composite event contains at least one labor milestone event. The anchoring confidence of the composite event is calculated using equation (5) with the complement of the product of the complements of the detection confidence of each member of the labor milestone event:

[0041]

[0042] In equation (5): The anchor confidence level for composite events has a value that is neither less than zero nor greater than one. The index is the member of the labor milestone event in the composite event, and the multiplication is performed only on the members of the labor milestone event; For the first The detection confidence level of each member of the labor milestone event shall be no less than zero and no greater than one. This indicates multiplying the terms within the parentheses together.

[0043] The uterine contraction event in the composite event is deterministically detected by the coordinate construction module using both energy and slope criteria, and serves as the time anchor of the composite event without being included in the multiplication. Equation (5) allows any member of the labor milestone event with high detection confidence to increase the anchoring confidence of the composite event, while the requirement for co-occurrence of multiple members ensures that the false detection of a single member is insufficient to cause the anchoring confidence to exceed the threshold. Only when the anchoring confidence is not less than the preset anchoring threshold (example value is...) (This invention is not limited to this point) When the composite event, along with its phase coordinates, is provided to the timing fusion module as an anchor point, the continuous process equivalent fractions and the discrete composite event anchor points converge at the timing fusion module.

[0044] The temporal fusion module performs temporal fusion, which involves taking the equivalent process scores from the multiple paths output by the temporal measurement module, recursively extrapolating along the labor progress axis under the constraints of the composite event anchor points provided by the event grouping module, and then backtracking and correcting at the anchor points, fusing them into a single fetal head descent trajectory. This module treats each observation as a noisy observation of the same, directly unmeasurable fetal head descent process. The latent state of labor in each labor cycle is denoted as a two-dimensional column vector, whose components are the normalized fetal head descent position and descent rate. The normalized fetal head descent position has a value that is not less than zero and not greater than one, with zero corresponding to the reference starting position and one corresponding to the crown. The descent rate is the incremental rate of the fetal head descent position between adjacent labor cycles. Since the coordinate construction module has placed the uterine contraction drive and intermittent slow variation in the uterine contraction peak phase and the uterine contraction intermittent phase, the evolution within a labor cycle is written as two segments: first through the uterine contraction peak phase sub-step, and then through the uterine contraction intermittent phase sub-step. The state transition is according to equation (6):

[0045]

[0046] In equation (6): , The first , No. The latent state vector of labor in each labor cycle; , These are the second-order state transition matrices for the peak contraction phase substep and the intermittent contraction phase substep, respectively. Both are upper triangular in form, with the diagonal element being a coefficient for maintaining the descent rate of the corresponding substep, and the upper triangular element being the integral step size of the descent position with respect to the descent rate in the corresponding substep. The integral step size of the peak contraction phase substep is greater than that of the intermittent contraction phase substep, and the latter is not less than zero, allowing for a larger step change in the descent position of the fetal head in the peak contraction phase substep, while only a gradual change occurs in the intermittent contraction phase substep. The descent rate maintenance coefficient of each substep is greater than zero and not greater than one. For process noise, its covariance matrix is ​​the process noise covariance matrix; It is a cyclical index.

[0047] This segmented form differs from the approach that uses the same dynamics throughout the entire process when physical time is the axis, and is consistent with the process of the fetal head moving forward during contractions and remaining basically stationary during intervals. Simultaneously, a soft constraint is applied to the fetal head descent position that is monotonically constant with the cycle exponent. This is achieved by setting the fetal head descent position to the larger of its estimated position and the position from the previous labor cycle after each state update. The equivalent scores of each labor stage parameter constitute the observation vector, and each equivalent score is considered a noisy observation of the same fetal head descent position. The observation equation is given by equation (7):

[0048]

[0049] In equation (7): For the first The observation vector for each labor cycle, with each component being the progress equivalent score of each labor parameter; The observation matrix has the same number of rows as the number of labor parameters and two columns. Each row is set to one or zero, indicating that the equivalent score for each process is the observation of the fetal head descent position. For observation noise, its covariance matrix is ​​a diagonal matrix, the first... The diagonal element is the first The labor parameters at the first The variance of the observed noise for each labor cycle is calculated by dividing the observation noise baseline coefficient by the sum of the confidence level of the labor parameter in that labor cycle and a preset minimum positive number. Thus, the lower the confidence level of the labor parameter, the greater its observed noise variance. It is a cyclical index; This is an index for labor process parameters.

[0050] Based on the state transition of equation (6) and the observation of equation (7), forward recursion is performed along the exponentially increasing direction of the cycle using standard Kalman filtering. That is, for each labor cycle, the estimate of the previous labor cycle is first predicted by the state transition matrix, and then updated by the observation vector of the current labor cycle according to the Kalman gain, to obtain the forward estimate of the latent state of labor for each labor cycle and its error covariance. The Kalman gain is determined by the prediction error covariance, the observation matrix and the observation noise covariance in a conventional way, which is well known in the art, and its stepwise expression is not listed in this embodiment. The initial value of the forward recursion is taken as the reference starting position of the fetal head descent and the descent rate is zero. The initial error covariance is taken as a diagonal matrix, and its diagonal elements are taken as larger values ​​representing the initial uncertainty. The example value is 1000, and the specific value can be determined by the noise level calibration. The forward recursion is performed online for each labor cycle. The estimate of each labor cycle only uses the observations up to that labor cycle. Therefore, the deviation caused by the insufficient image quality in the early stage of labor is propagated backward along the progression axis in the forward trajectory.

[0051] To correct early biases accumulated along the progression axis in the forward trajectory, at composite event anchor points that meet the threshold, the forward estimates of each labor cycle before the anchor point are corrected using standard backward smoothing along the direction of decreasing cycle exponent. Specifically, the estimates at the anchor point and subsequent cycles are back-substituted with the backward smoothing gain, and the estimates and error covariance of each labor cycle are updated cycle by cycle towards the earlier stage of labor. The backward smoothing gain is determined conventionally by the forward estimation error covariance of each labor cycle, the state transition matrix, and the prediction error covariance of the next labor cycle, which is well-known in the art, and its stepwise expression is not listed in this embodiment. Composite events correspond to confirmed anatomical stages in the fetal head descent process. These events, as high-confidence evidence, are advanced towards the earlier stage of labor through backward smoothing, correcting estimates obtained before the anchor point based on low-confidence observations. Corrections that reduce the fetal head descent position relative to the previous labor cycle are not adopted to comply with the soft constraint of non-regression in fetal head descent. The fusion result will be... The position of fetal head descent in each labor cycle is denoted as the fused fetal head descent position for that labor cycle: if there is a composite event anchor point that meets the threshold up to the present, the backward smoothed estimate is used; if there is no composite event anchor point that meets the threshold, the forward estimate is used. The resulting fused fetal head descent position sequence uses the cycle index as the axis of labor progress and serves as the input to the trend assessment module.

[0052] The aforementioned observation weights are determined by the confidence levels of each labor parameter, which are synthesized from three aspects: first, the imaging quality of the image on which the labor parameter is based, taken from the structural segmentation confidence level; second, the phase purity of the measurement, whose value increases as the phase value of the measurement is located in the center of its respective phase interval; and third, the periodic consistency of the measurement, whose value decreases as the change of the measurement relative to the previous labor cycle increases. When only one of these factors is used, the confidence level is taken from that factor. If any of the three factors is lower, the labor parameter is downweighted. It should be distinguished that fetal head shaping causes a systematic shift in the direction of labor parameters such as the fetal head progression angle, which is different in nature from the random errors caused by imaging quality. The detector for the fetal head shaping jump event outputs a fetal head shaping grade according to the degree of shaping. The fetal head shaping grade is an integer not less than zero, and zero indicates no shaping. When a fetal head shaping jump is detected, the corresponding labor parameters are subtracted from their phase-normalized measurements by a systematic offset determined by the fetal head shaping grade, as shown in equation (8):

[0053]

[0054] In equation (8): The measured value after applying systematic bias correction is used to replace the phase-normalized measured value before correction when calculating the phase-normalized representative value of this labor parameter in subsequent calculations. For the first Phase-normalized measurements of individual labor parameters; For the first Grades of fetal head shaping during each labor cycle; For the first The systematic bias coefficient for each labor process parameter corresponding to the unit fetal head shaping grading; It is a cyclical index; This is an index for labor process parameters.

[0055] Therefore, the systematic deviation determined by the tire head plasticity grading is explicitly subtracted by equation (8), while the unreliability of randomness is still dealt with by increasing the variance of observation noise. The two are handled separately.

[0056] Trend Assessment Module. This module performs trend assessment, which is based on the process trajectory obtained by time series fusion and the equivalent scores of each process to determine the direction and consistency of each labor process parameter, and to converge the continuous process into one of the following labor progress trend states: trend state, direction indeterminate state, and parameter divergence state. The acquisition frequency is adjusted according to the state and the feedback of the detected events. First, the direction of change of each labor process parameter is determined: for the equivalent score of each process, the median of the difference quotient between any two labor cycles is taken as the rate of change of its relative cycle index. The median slope is used to make the direction estimation insensitive to outliers of individual labor cycles, as shown in Equation (9):

[0057]

[0058] In equation (9): For the first The rate of change of each production process parameter relative to the cycle index; To perform the median calculation; , The first The labor parameters at the first , No. Equivalent fraction of the process in each labor cycle; , To meet Not less than one Less than and Not greater than The cyclical index pair; This represents the total number of current labor cycles. This is an index for labor process parameters.

[0059] When the total number of current labor cycles is less than two, it is insufficient to estimate the rate of change using equation (9). In this case, the direction of each labor parameter is not calculated, and the labor progress trend state is directly output as an undetermined state. The labor progress trend state is generated again in the following manner after the total number of current labor cycles has accumulated to no less than two. For the first... Each labor process parameter is configured with a dead zone and hysteresis: when its rate of change is greater than the dead zone, its direction is determined to be advancing; when its rate of change is less than the opposite of the dead zone, its direction is determined to be retreating; when the absolute value of its rate of change is not greater than the dead zone, its direction is determined to be undetermined. The dead zone is a preset multiple of the noise fluctuation of the equivalent fraction of the labor process parameter in the steady segment. The hysteresis is used to suppress the direction reversal of the rate of change at the boundaries of the dead zone. It can be implemented using a Schmitt triggering method: when the rate of change crosses the upper or lower boundary of the dead zone from inside to outside, the direction is immediately updated to advancing or retreating; when the rate of change returns to the dead zone from outside, the direction is reversed to undetermined only after a preset number of labor cycles fall within the dead zone. An example value for the preset number is 2, so that the direction does not repeatedly switch due to small fluctuations near the threshold.

[0060] Secondly, it is determined whether the parameters of each labor stage are consistent: Since the equivalent scores of each stage increase with the progress after calibration, the rates of change of each stage can be directly compared. Therefore, the confidence level of each labor stage parameter in the current labor cycle is used as the weight, and the weighted variance of each rate of change relative to its weighted mean is taken as the consistency of the process, as shown in equation (10):

[0061]

[0062] In formula (10): The weighted variance of the rate of change of each production process parameter represents the consistency of the process; the smaller the value, the more consistent the trend of each parameter. For indexing labor parameters From one to Summation of all terms, This represents the total number of labor parameters. For the first The rate of change of each production process parameter; For the first The weights of each labor process parameter are determined by taking the confidence level of that parameter in the current labor cycle and normalizing it so that the sum of the weights is one. This is the weighted average of each rate of change, with each rate of change having its own weight.

[0063] Using confidence levels as weights, labor parameters with reliable imaging bear a significant weight in the consistency assessment. Based on the direction, progress consistency, and abnormal labor conditions described later for each labor parameter, the labor progress trend state is generated according to the following rules: when all labor parameters are in the advancing direction, the progress consistency is less than the first threshold, and the abnormal labor conditions are not met, a trend state is output; when a labor parameter is in the regressing direction, or the progress consistency is greater than the second threshold, or the abnormal labor conditions are met, a parameter divergence state is output; in other cases, a direction-indeterminate state is output; the first threshold is not greater than the second threshold. The above rules incorporate abnormal labor conditions as a component of the parameter divergence state determination into the same decision, ensuring that each input corresponds to a unique labor progress trend state; the labor progress trend state takes one of the trend state, direction-indeterminate state, or parameter divergence state, and does not map to the delivery method, diagnosis, or medical orders.

[0064] To maintain assessment even when imaging quality deteriorates for individual labor parameters, when the first When the confidence level of a labor process parameter is lower than a preset threshold, the weight of that labor process parameter is reduced, and the weighted average of the confidence levels of the equivalent scores of the remaining labor process parameters is used as its extrapolated value for judgment, as shown in equation (11):

[0065]

[0066] In equation (11): The first one that was demoted The labor parameters at the first Extrapolated equivalent fractions of the progress of each labor cycle; For the first The labor parameters at the first Equivalent fraction of the process in each labor cycle; For the first The labor parameters at the first Confidence level of each labor cycle; summation of the results divided by the first cycle. Other than the individual labor parameters, the remaining labor parameters are used; It is a cyclical index; , This is an index for labor process parameters.

[0067] Since the equivalent scores of each process are aligned in direction and scale after calibration, the confidence-weighted mean of the remaining labor parameters is a robust estimate of the common process. Based on this, the downweighted labor parameters are extrapolated, thus the assessment can still be maintained even when individual labor parameters fail. The abnormal labor conditions used for judgment are determined by the contraction conversion efficiency and the advance-retreat ratio. The contraction conversion efficiency is the ratio of the net fetal head progression in a labor cycle to the integral of the contraction intensity in that labor cycle, as shown in equation (12):

[0068]

[0069] In equation (12): For the first The uterine contraction conversion efficiency of each labor cycle; , The first in the fusion result , No. The fused fetal head descent position for each labor cycle is the output value after forward recursion and backward smoothing optimization, which is different from the true value of the fetal head descent position in the aforementioned latent state of labor. The difference between the two is the net fetal head progression in that labor cycle. The envelope energy of the uterine contraction signal. Physical time; The baseline energy of uterine contractions is the baseline level of the envelope energy during the intercontinental period, which can be taken as the same baseline level as the baseline threshold used to detect uterine contraction events. This means taking the positive part, that is, treating negative values ​​as zero and non-negative values ​​as themselves; , The first The start and end times of uterine contractions corresponding to each labor cycle; the denominator of equation (12) is the integral of the uterine contraction intensity of that labor cycle. It is a cyclical index.

[0070] The advance-retreat ratio is the ratio of the magnitude of the advance during contractions to the magnitude of the retreat during the intervals between contractions. It is calculated by taking a specified labor parameter that monotonically increases with the progress of labor and measuring it in its unnormalized form, as shown in equation (13):

[0071]

[0072] In equation (13): For the first The advance-to-retreat ratio of each labor cycle; For the specified labor parameters in the first The median of all measurements at the peak phase of uterine contractions during each labor cycle; , The specified labor parameters are respectively in the first... , No. Median of all measurements during the intermittent phases of uterine contractions in each labor cycle; To prevent the preset minimum positive number with a denominator of zero, the same minimum positive number as described in equation (7) is used; the numerator is the forward shift of the uterine contraction period relative to the previous interval, and the denominator is the backward shift of the uterine contraction period relative to the current interval; the specified labor parameters are, for example, taken as the fetal head progression angle; It is a cyclical index.

[0073] when When the value equals 1, there is no intermittent phase measurement from the previous labor cycle. The median intermittent phase from the previous labor cycle is then taken as the median intermittent phase of the current labor cycle. Unnormalized measurements are used here to preserve the phase difference between contraction advance and intermittent regression. Therefore, the abnormal labor condition is defined as follows: the integral of contraction intensity is not less than the preset intensity and the contraction conversion efficiency is less than the preset efficiency, or the advance-retreat ratio is not less than the preset advance-retreat ratio threshold. This condition is incorporated into the generation of parameter divergence states according to the aforementioned decision rules.

[0074] Data acquisition frequency adjustment. Furthermore, this module also adjusts the data acquisition frequency based on the progress of labor. The median slope of the fusion fetal head descent position sequence from the first labor cycle to a preset number of labor cycles is used as the individual baseline rate; the preset number is 5 (example value). The median slope of the fusion fetal head descent position sequence over the most recent few labor cycles is used as the current descent rate; the number of recent few labor cycles is 3 (example value). The attenuation ratio is defined as the ratio of the difference between the individual baseline rate and the current descent rate to the individual baseline rate. When the attenuation ratio is not less than the attenuation ratio threshold (example value is...), the attenuation rate is adjusted accordingly. When the current rate of decline drops to less than half of the individual baseline rate and the number of continuous labor cycles is not less than the preset number of attenuation cycles (example value is 3), a trend attenuation flag is added to the output trend state. The trend attenuation flag is a flag added to the trend state rather than an independent labor progress trend state. The above example values ​​can be determined based on clinical data statistics, and this invention does not limit them. Then, the acquisition frequency of the ultrasound image sequence is adjusted according to formula (14):

[0075]

[0076] In equation (14): The acquisition frequency of the ultrasound image sequence; , These are the lower and upper limits of the sampling frequency, respectively. For a monotonically saturated function with values ​​greater than zero and less than one, the logistic function is taken as an example. Uncertainty characterizing the reliability of the current labor progress trend; This is a marker quantity for labor milestone events; it is set to one if a labor milestone event occurs, and zero otherwise. This is the uncertainty gain coefficient; This is the event-driven gain coefficient.

[0077] The uncertainty is obtained by first converting the three-state criterion scores into the posterior probabilities of the trend state, direction-indeterminate state, and parameter divergence state through flexibility maximization, and then taking their normalized entropy, as shown in equation (15):

[0078]

[0079] In equation (15): The uncertainty is defined as a value that is neither less than zero nor greater than one. For state index Summing of terms from one to three; , , The posterior probabilities are the trend state, the direction-indeterminate state, and the parameter divergence state, respectively. They are obtained by flexible maximization of the criterion scores of the three states, that is, by dividing each criterion score by the temperature coefficient, taking the exponent, and normalizing. Flexible maximization is well known in the art. It is the natural logarithm.

[0080] The criteria scores for the three states are as follows: the trend state score is calculated by subtracting the process consistency from the first threshold, dividing by the first threshold, and subtracting the product of the direction weight coefficient and the number of backtracking parameters. The direction-indeterminate state score is fixed at zero and serves as a relative benchmark for the intermediate reference state between the trend state and the parameter divergence state. The parameter divergence state score is calculated by subtracting the second threshold from the process consistency, dividing by the second threshold, and adding the product of the direction weight coefficient and the number of backtracking parameters. The direction weight coefficient is a preset positive number used to adjust the penalty intensity of the backtracking direction on the trend state score and the parameter divergence state score. An example value is... This invention is not limited; the number of backtracking parameters is the number of process parameters that are determined to be backtracking in the current direction; the temperature coefficient (example value) The value of the exponent should be avoided to prevent it from being too small, which could lead to overflow in the exponent calculation. In practice, a conventional numerical stabilization method can be used: subtract the maximum value from each criterion score before taking the exponent. This process does not change the values ​​of the posterior probability and uncertainty. Dividing by the natural logarithm of three normalizes the uncertainty to a value neither less than zero nor greater than one. The posterior probability is a normalized weight constructed based on the criterion scores to quantify the relative reliability of the three states. Its monotonicity satisfies the following: the higher the criterion score of a certain state relative to the others, the larger the normalized weight corresponding to that state. The uncertainty serves as a measure of the concentration of the normalized weight distribution, used to drive the adjustment of the collection frequency, and is not used as a diagnostic judgment of the labor progress trend. When the criterion scores of the three states are close, the posterior distribution tends to be uniform and the uncertainty approaches one. When the criterion score of a certain state is significantly dominant, the posterior distribution is concentrated and the uncertainty approaches zero. According to equations (14) and (15), when the trend of labor progress is clear and there are no labor milestone events, the collection frequency tends to the lower limit; when the trend of labor progress is unclear or a labor milestone event occurs, the collection frequency tends to the upper limit. The output of the trend assessment and the detected events are thus fed back to the collection stage, so that the collection behavior on which the coordinate construction module is based is adjusted with the progress of labor. Each module forms a closed loop that combines forward processing and feedback adjustment.

[0081] Combination Figure 2Based on the collaboration of the above modules, the processing performed by the system proceeds sequentially through five steps along the data flow order: coordinate construction, time series measurement, event grouping, time series fusion, and trend evaluation. These five steps correspond to the coordinate construction module, time series measurement module, event grouping module, time series fusion module, and trend evaluation module. Specifically: S1 Coordinate Construction: Constructing labor phase coordinates based on uterine contraction events, with the cycle index as the axis of labor progress; S2 Time Series Measurement: Measuring various labor parameters under the labor phase coordinates, estimating and subtracting phase offsets using paired uterine contraction peak phase measurements and uterine contraction interval phase measurements to obtain time series data; S3 Event Grouping: Detecting labor milestone events and grouping co-occurring labor milestone events and uterine contraction events within a preset time window into composite events; S4 Time Series Fusion: Establishing a labor latent state space model in the form of segmented phase intervals for the process equations, forward recursively extrapolating along the increasing direction of the cycle index, and correcting the estimates before the anchor point along the decreasing direction of the cycle index at the composite event anchor point, obtaining a fusion result with the cycle index as the axis of labor progress; S5 Trend Assessment: Generating and outputting a labor progress trend state as one of the following: trend state, direction indeterminate state, or parameter divergence state, based on the fusion result and the directional relationship and confidence level of each time series data; Further, in the trend assessment step, adjusting the acquisition frequency according to the state and the detected events to form feedback. The above modules can be deployed on the ultrasound equipment or on the computing device connected to it. This invention does not limit the deployment method.

[0082] The thresholds, weighting coefficients, integration step sizes, descent rate maintenance coefficients, dead zones, calibration parameters, example values, and initial values ​​of each covariance matrix given above are all examples; those skilled in the art can determine them through calibration experiments, historical data statistics, or simulation analysis based on the specific performance of the ultrasonic equipment, acquisition conditions, and target indicators. The determination methods for parameters not given unique values ​​have been described in the corresponding paragraphs. This invention does not limit the specific values ​​of the above parameters.

[0083] The above descriptions are specific embodiments of the present invention and are not intended to limit the invention. Equivalent substitutions or modifications made based on the technical solutions of the present invention are all within the protection scope of the present invention.

Claims

1. A system for fusing multi-parameter time-series ultrasound data during labor and assessing the progress of labor, characterized in that: include: The coordinate construction module is used to acquire uterine contraction signals and ultrasound image sequences during labor, detect uterine contraction events from the uterine contraction signals, take the interval defined by two adjacent uterine contraction events as the labor cycle, assign an increasing cycle index to each labor cycle according to the order of occurrence, assign a phase value to the sampling time within the labor cycle to represent its relative position within the labor cycle, and construct the labor phase coordinates by the cycle index and the phase value. The time-series measurement module is used to measure multiple labor parameters of the ultrasound image sequence under the labor phase coordinates, and to perform phase normalization on each of the labor parameters according to the phase interval to which the phase value belongs, so as to obtain phase-normalized time-series data. The event grouping module is used to detect labor milestone events in the ultrasound image sequence, assign the labor milestone events to the labor phase coordinates corresponding to their occurrence time, and group the labor milestone events and the uterine contraction events that co-occur within a preset time window into composite events. The time series fusion module is used to perform bidirectional fusion of the time series data after phase normalization according to the phase interval, using the composite event as the anchor point, and forward recursion and backward propagation, to obtain a fusion result with the cycle index as the axis of labor progress. The trend assessment module is used to generate and output the labor progress trend state based on the fusion results and the directional relationship and confidence of each of the time series data under the labor phase coordinates. The labor progress trend state is one of the following: trend state, direction indeterminate state, and parameter divergence state.

2. The system according to claim 1, characterized in that, The time-series measurement module performs phase normalization, including: dividing the phase value into phase intervals according to its position within the labor cycle, wherein the phase interval includes at least the peak phase of uterine contractions and the intermittent phase of uterine contractions; for the same labor parameter, using the measured values ​​at the peak phase of uterine contractions and the measured values ​​at the intermittent phase of uterine contractions for the same mother within the same labor cycle, estimating the phase offset of the labor parameter as the phase value changes; and subtracting the phase offset from the measured value of the labor parameter to obtain the phase-normalized time-series data.

3. The system according to claim 1, characterized in that, The time series measurement module is also used to determine the monotonic calibration of the measured value to the process equivalent score for each of the labor parameters, and to map each of the labor parameters to a process equivalent score with the same value range and which increases with the progress of labor through the monotonic calibration. The process equivalent score is used as the observation of the time series fusion module, and the dispersion between them is used by the trend evaluation module to determine the parameter divergence state.

4. The system according to claim 2, characterized in that, The time-series fusion module is used to establish a latent state space model of labor. The latent state of labor in this model includes the position and rate of fetal head descent. The time-series data after phase normalization are used as the observations of the latent state of labor. The process equation of the latent state space model allows the position of fetal head descent to advance stepwise relative to the previous labor cycle at the peak of contraction, constrains the position of fetal head descent to change slowly relative to the previous labor cycle at the interval of contraction, and applies a soft constraint to the position of fetal head descent that is monotonically constant with the cycle exponent. The observation noise covariance of each observation is determined by the confidence level of the corresponding labor parameter. The lower the confidence level, the larger the observation noise covariance.

5. The system according to claim 4, characterized in that, The time-series fusion module forward procures the forward estimate of the latent state of labor along the direction of increasing periodic exponent, and then propagates backward along the direction of decreasing periodic exponent at the composite event anchor point. The forward estimates of each labor cycle before the anchor point are corrected by the forward estimates at and after the anchor point to obtain the fusion result after bidirectional fusion. Furthermore, corrections that reduce the position of the fetal head descent relative to the previous labor cycle are not adopted to comply with the monotonic soft constraint.

6. The system according to claim 4, characterized in that, The confidence level of each of the labor parameters is synthesized from at least one of image quality, phase purity, and periodic consistency, wherein the image quality is taken from the structural segmentation confidence level of the image on which the labor parameter is based, the phase purity increases as the measured phase value is located in the center of its respective phase interval, and the periodic consistency decreases as the change of the measurement relative to the previous labor cycle increases; when the event grouping module detects a change in fetal head shaping, it deducts systematic bias from the corresponding labor parameter according to the fetal head shaping grade to distinguish it from random unreliability characterized by the image quality.

7. The system according to claim 3, characterized in that, The trend assessment module generates the labor progress trend state, including: estimating the rate of change of the equivalent score of each labor parameter relative to the cycle index; determining the direction of the labor parameter as undetermined when the absolute value of the rate of change is not greater than a preset dead zone; otherwise, determining it as advancing or retreating according to the sign of the rate of change, and setting hysteresis to suppress directional reversal at the dead zone boundary; calculating the dispersion of the rate of change of each labor parameter weighted by the confidence level of each labor parameter, as the process consistency; and generating the labor progress trend state according to the following rules: when the total number of labor cycles is less than two, output the undetermined state; when the direction of each labor parameter is advancing, the process consistency is less than a first threshold, and the abnormal labor force condition is not met, output the trend state; when there is a labor parameter in retreat, or when the process consistency is greater than a second threshold, or when the abnormal labor force condition is met, output the parameter divergence state; in other cases, output the undetermined state; the first threshold is not greater than the second threshold.

8. The system according to claim 7, characterized in that, The trend assessment module is further configured to: when the confidence level of any of the labor process parameters is lower than a preset threshold, reduce the weight of the labor process parameter, and use the confidence-weighted average of the progress equivalent scores of the other labor process parameters as the extrapolated value of the progress equivalent score of the labor process parameter to participate in the generation of the labor progress trend status.

9. The system according to claim 7, characterized in that, The abnormal labor force conditions are determined by the contraction conversion efficiency and the advance-retreat ratio: For each labor cycle, the ratio of the net progress of the fetal head descent position relative to the previous labor cycle to the integral of the contraction intensity within that labor cycle is calculated as the contraction conversion efficiency, and the ratio of the advance amplitude of a specified labor parameter at the peak of the contraction to the retreat amplitude at the interval of the contraction is calculated as the advance-retreat ratio; the abnormal labor force conditions are: the integral of the contraction intensity is not less than a preset intensity and the contraction conversion efficiency is less than a preset efficiency, or the advance-retreat ratio is not less than a preset advance-retreat ratio threshold.

10. The system according to claim 7, characterized in that, The trend assessment module is also used to: obtain an individual baseline rate from the fusion result from the first labor cycle after the start of labor to the preset number of labor cycles; when the decay ratio of the rate of change of the fusion result relative to the individual baseline rate and the number of continuous labor cycles meets a preset condition, a trend decay flag is added while outputting the trend state. The acquisition frequency of the ultrasound image sequence is adjusted based on the uncertainty characterizing the reliability of the labor progress trend and the labor milestone events.

Images