A neonatal state recognition method based on colostrum nutrition immune characteristics fusion

Abstract

The application relates to the technical field of multi-source physiological data processing, in particular to a newborn state recognition method based on colostrum nutrition and immune feature fusion. The method comprises the following steps: acquiring colostrum nutrition indexes, immune indexes, actual feeding time, intake and newborn state data of the same mother-infant pairing object at multiple time points after delivery; constructing nutrition tracks and immune tracks according to postpartum physiological time and determining corresponding phase change time; generating nutrition exposure tracks and immune exposure tracks according to the actual feeding time and the intake; constructing a newborn adaptation state track and determining a state phase change time through individual baseline correction; and determining the newborn adaptation state according to the time sequence relationship of the three types of phase change time, the response time lag and the track fluctuation degree after the phase change. The application can reduce the influence of time mismatch of multi-source data and single-point detection limitation on the state recognition accuracy.

Description

Technical Field

[0001] This application relates to the field of multi-source physiological data processing technology, and in particular to a method for neonatal status identification based on the fusion of colostrum nutritional and immune characteristics. Background Technology

[0002] Colostrum is the milk secreted by the mother in the early postpartum period. Its nutritional and immune components gradually change over time, while the newborn's physiological indicators are also undergoing a continuous adaptation process. Current technologies typically analyze the composition of a single colostrum sample, then splice the results with the newborn's gestational age, birth weight, or vital signs. The spliced ​​features are then input into a classification model to output nutritional levels, health risks, or feeding recommendations. This approach primarily focuses on whether the test values ​​at a specific point in time are abnormal. By associating data from different sources in the order they were recorded, machine learning algorithms are used to establish a mapping relationship between input features and output labels.

[0003] However, in existing technologies, the time of colostrum collection, the actual intake time of newborns, and the time of newborn status detection are usually not consistent. Furthermore, it is difficult to distinguish between temporary fluctuations and phase transitions in the absolute detection values ​​of different components. This makes it easy to incorrectly associate colostrum samples that have not yet been ingested with changes in the newborn's status that have already occurred by directly splicing data. This makes it impossible to accurately reflect the true correspondence between the dynamic change stages of colostrum and the newborn's adaptation process. Summary of the Invention

[0004] This application provides a method for neonatal status identification based on the fusion of colostrum nutritional and immune characteristics to solve the above-mentioned problems. The method includes: S1. Obtain colostrum nutritional indicators and colostrum immune indicators at multiple sampling points after delivery for the same mother-infant pair, as well as the corresponding actual feeding time, intake and newborn status data. S2. The colostrum nutritional indicators and colostrum immune indicators are fitted according to the postpartum physiological time to obtain the nutritional trajectory and immune trajectory. The nutritional phase transition time and immune phase transition time are determined based on the degree of improvement of the residual of the segmented fitting compared with the single-segment fitting and the change of the trajectory slope before and after the segmentation. S3. Based on the actual feeding time and intake, convert the nutritional trajectory and immune trajectory into the actual nutritional exposure trajectory and immune exposure trajectory received by the newborn, respectively. S4. Perform individual baseline correction on the newborn state data, construct the newborn adaptation state trajectory, and determine the state phase transition time of the adaptation state trajectory; S5. Based on the temporal relationship between the nutritional phase transition time, the immune phase transition time, and the state phase transition time, combined with the response time lag between the nutritional exposure trajectory, the immune exposure trajectory, and the adaptive state trajectory, as well as the degree of trajectory fluctuation after the state phase transition, determine the newborn's adaptive state.

[0005] Optionally, the multiple postpartum sampling points include no fewer than four valid sampling points; Before performing the trajectory fitting, the mass weight of each detection value is determined based on the interval between collection and detection of the colostrum sample, storage temperature, number of freeze-thaw cycles, coefficient of variation of parallel detection, and calibration deviation of the detection equipment. The contribution of each detection value to the trajectory fitting is then adjusted according to the mass weight.

[0006] Optionally, intra-individual standardization can be performed on the colostrum nutritional and immune indicators of the same mother-infant pair. The sampling time point of the first Standardized values ​​of each indicator It is determined according to the following formula: ; in, For the corresponding detection value, For the first The median of an indicator at multiple sampling points For the first The interquartile differences of an indicator at multiple sampling points It is a non-zero constant.

[0007] Optionally, the weights of each indicator can be determined based on the number of valid sampling points, the missing rate, and the stability of parallel detection. The standardized values ​​of the colostrum nutritional indicators are fused according to the corresponding indicator weights to obtain the nutritional status value at each sampling time point, and the standardized values ​​of the colostrum immune indicators are fused to obtain the immune status value at each sampling time point. Using the postpartum physiological time relative to the delivery time at each sampling point as the independent variable, the nutritional status value and immune status value are locally fitted to obtain the nutritional trajectory and immune trajectory.

[0008] Optionally, determining the nutritional phase transition time or the immune phase transition time includes: Candidate phase transition moments from the nutrient trajectory or immune trajectory are selected sequentially, and the first fitting residual is calculated when the trajectory is fitted using a single segment. And the second fitting residuals when fitting before and after the candidate phase transition time, respectively. Calculate the residual improvement rate using the following formula. : ; in, It is a non-zero constant; The candidate phase transition moment is determined as the phase transition moment of the corresponding trajectory when the residual improvement rate reaches a preset improvement threshold, the difference in trajectory slope before and after the candidate phase transition moment reaches a preset slope threshold, and the trend of change after the phase transition continues for a preset duration.

[0009] Optionally, converting the nutrient trajectory and immune trajectory into nutrient exposure trajectory and immune exposure trajectory includes: Determine the nutritional and immune status values ​​of the colostrum used in each actual feeding; Based on the feeding time, intake, and time elapsed after each feeding, the nutritional exposure contribution and immune exposure contribution of each feeding to the different collection times under different conditions were determined. The nutritional exposure contribution and immune exposure contribution corresponding to each actual feeding are summed to obtain the nutritional exposure trajectory and immune exposure trajectory.

[0010] Optionally, the nutritional exposure trajectory and the immune exposure trajectory are determined according to the following formulas: ; ; in, For the nutrient exposure trajectory, For the immune exposure trajectory, This refers to the actual number of feedings. For the first The actual feeding time. For the first The corresponding intake amount for each actual feeding. and These are the nutritional status and immune status values ​​corresponding to the colostrum fed to the individual. and These are the response functions for nutritional exposure and immune exposure over time after feeding, respectively. When the state acquisition time is earlier than the corresponding feeding time, the exposure contribution of that feeding is zero.

[0011] Optionally, the newborn status data includes fast response status data and slow response status data. The fast response status data includes at least one of heart rate stability, blood oxygen stability, body temperature change, and feeding completion rate. The slow response status data includes at least one of weight change, feeding interval change, and excretion status. The corresponding individual baselines are subtracted from the fast response state data and the slow response state data respectively to construct the fast response trajectory and the slow response trajectory. The newborn adaptation state trajectory is obtained by weighted fusion based on the effective data ratio of the two types of state data in the corresponding time period.

[0012] Optionally, within a preset time delay range, the candidate time delay values ​​are changed sequentially, and the first degree of correspondence between the rate of change of the nutritional exposure trajectory and the rate of change of the adaptive state trajectory, and the second degree of correspondence between the rate of change of the immune exposure trajectory and the rate of change of the adaptive state trajectory are calculated respectively. The candidate time delay corresponding to the highest first degree of correspondence is determined as the nutritional response time delay, and the candidate time delay corresponding to the highest second degree of correspondence is determined as the immune response time delay. Based on the deviation between the state phase transition time and the nutritional and immune phase transition times after correction by the corresponding response time delay, the phase transition correspondence of the three types of trajectories is determined.

[0013] Optionally, based on the degree of phase transition correspondence, the consistency of the direction of change between the nutritional exposure trajectory and the immune exposure trajectory and the adaptive state trajectory, and the degree of fluctuation of the adaptive state trajectory after the phase transition, the newborn is determined to be in one of the following states: synchronous adaptation, delayed adaptation, weak response, asynchronous trajectory, or unstable adaptation. When the number of valid sampling points or the confidence level of the trajectory phase transition does not reach the corresponding threshold, the system outputs a data shortage status and determines the supplementary sampling time based on the candidate phase transition interval of the current trajectory.

[0014] Through the above technical solution, this application achieves the following beneficial effects: This application constructs colostrum nutritional and immune trajectories according to postpartum physiological time, and determines the corresponding phase transition time through segmented fitting. This can reflect the dynamic process of colostrum nutritional and immune components transitioning from one stage of change to another, and reduce the identification bias caused by random fluctuations at a single sampling point.

[0015] This application does not directly concatenate nutritional and immune indicators into a unified feature vector. Instead, it performs trajectory fitting and phase transition detection separately to obtain the timing of nutritional and immune phase transitions and the sequential relationship between them. This is beneficial for identifying situations where changes in colostrum nutrition and immune function are asynchronous.

[0016] This application converts the colostrum nutrition trajectory and immune trajectory into the actual nutrition exposure trajectory and immune exposure trajectory received by the newborn based on the actual feeding time, intake amount and time elapsed after feeding. This avoids directly substituting the newborn's intake time with the colostrum sampling time and reduces the risk of miscorrelation between colostrum test data and newborn status data.

[0017] By performing individual baseline correction on newborn status data and processing fast-response status data and slow-response status data separately, this application can reduce the impact of differences in gestational age, birth weight and initial birth status on the identification results, making the adaptation status more reflective of the actual changes in newborns after receiving colostrum.

[0018] This application integrates the phase transition time relationship of three types of trajectories, the nutrient response time lag, the immune response time lag, and the degree of trajectory fluctuation after the state phase transition to identify the state. This avoids making judgments based on a single indicator or fixed threshold, and is helpful in distinguishing different states such as synchronous adaptation, delayed adaptation, weak response, trajectory asynchrony, and unstable adaptation.

[0019] This application can output intermediate results such as the timing of nutritional phase transition, immune phase transition, state phase transition, response lag, and stability after phase transition, which makes it easier to determine whether the identification state mainly comes from asynchronous colostrum trajectories, delayed neonatal response, or continuous state fluctuations, thus reducing the problem of black box classification results being difficult to interpret.

[0020] By combining sample storage conditions, detection repeatability, equipment calibration deviation, and the number of effective sampling points to determine data quality and trajectory phase transition confidence, this application can reduce the impact of abnormal detection values ​​on trajectory fitting results; when there is insufficient effective data, it can output the data shortage state and determine the time for supplementary sampling, avoiding the forced generation of low-confidence state recognition results.

[0021] This application outputs the newborn's adaptation status to the dynamic changes of colostrum and the corresponding confidence level, rather than directly giving a specific disease diagnosis conclusion. The technical solution focuses on solving data processing problems such as time alignment of multi-source data, trajectory phase transition detection and state recognition, and has a clear data processing process and technical effect. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 The flowchart illustrates a neonatal status identification method based on the fusion of colostrum nutritional and immune characteristics, as provided in one embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0025] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0026] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0027] Example 1: In existing technologies, the identification of a newborn's adaptation status typically relies on the test values ​​of a single colostrum sample, which are then simply concatenated with the newborn's gestational age, birth weight, or vital signs before being input into a classification model. This approach primarily focuses on whether the test values ​​are abnormal at a specific point in time, failing to adequately consider the rate of change, the turning points, and the phase transition relationships between nutritional and immune components as colostrum transitions from the colostrum stage to the transitional milk stage. For example, two mothers may have the same immunoglobulin concentration at the same postpartum time, but one mother's indicator may be in a rapid decline phase while the other has entered a stable phase. A single-point concentration cannot distinguish the different stages of milk change they are in. Furthermore, the time of colostrum collection, the actual intake time of the newborn, and the time of newborn status monitoring are usually not consistent. If data is directly concatenated according to the recording order, it is easy to incorrectly associate colostrum samples that have not yet been ingested with changes in the newborn's status that have already occurred. Individual differences in the initial state of different newborns can also easily lead to biased identification results, and conventional models often lack the technical interpretability of the identification results.

[0028] Based on the above issues, please refer to Figure 1 The following is a flowchart of a neonatal status identification method based on the fusion of colostrum nutritional and immune characteristics, provided in an embodiment of this application. The method includes the following steps: Step 1: Obtain colostrum nutritional indicators and colostrum immune indicators from multiple sampling points after delivery for the same mother-infant pair, as well as the corresponding actual feeding time, intake, and newborn status data; Colostrum nutritional indicators can refer to parameters reflecting the content of nutrients in colostrum, specifically including at least two of the following: total protein, fat, lactose, total solids, human milk oligosaccharides, free amino acids, medium-chain fatty acids, or long-chain polyunsaturated fatty acids. Colostrum immune indicators can refer to parameters reflecting the content of immune-active components in colostrum, specifically including at least two of the following: secretory immunoglobulin A, immunoglobulin G, immunoglobulin M, lactoferrin, lysozyme, interleukin, transforming growth factor, or the number of colostrum immune-active cells. These indicators are obtained by analyzing collected colostrum samples using laboratory testing equipment. The actual feeding time can refer to the exact moment when the newborn actually ingests colostrum, and the intake amount can refer to the volume or mass of colostrum ingested by the newborn in a single feeding. These data come from the automatic recording of maternal and infant care record systems or intelligent feeding devices. Neonatal status data can refer to parameters that reflect the physiological and behavioral adaptation of newborns, including fast-response status data and slow-response status data. Fast-response status data includes at least one of heart rate stability, blood oxygen stability, body temperature change, and feeding completion rate, while slow-response status data includes at least one of weight change, feeding interval change, and excretion status.

[0029] Specifically, a unique mother-infant pairing identifier is first established for each mother and newborn to ensure the correspondence of data sources. For each colostrum sample, information such as collection time, testing time, storage time, and storage temperature is recorded, and the collection time is converted into postpartum physiological time using the delivery time as the time origin. For example, colostrum collected at 6 hours, 18 hours, 34 hours, 55 hours, and 82 hours postpartum are marked as different sampling time points. This invention does not require adjacent sampling time intervals to be the same, but to ensure the accuracy of trajectory fitting, each mother-infant pairing preferably has no less than four valid colostrum sampling time points. By acquiring multi-dimensional data from multiple time points, the dynamic changes in colostrum components over time and the evolutionary trend of the newborn's state can be fully captured, providing a sufficient data foundation for subsequent trajectory reconstruction and phase transition identification.

[0030] Step 2: According to the postpartum physiological time, the nutritional and immune indicators of colostrum are fitted to obtain the nutritional trajectory and immune trajectory. Based on the degree of improvement of the residual of the segmented fitting compared with the single-segment fitting and the change of trajectory slope before and after segmentation, the nutritional phase transition time and the immune phase transition time are determined. The nutritional trajectory can refer to a continuous curve showing the changes in colostrum nutritional indicators over postpartum physiological time, and the immune trajectory can refer to a continuous curve showing the changes in colostrum immune indicators over postpartum physiological time. These two trajectories are obtained through local regression or piecewise spline fitting of standardized values ​​from discrete sampling points. The nutritional phase transition moment and the immune phase transition moment can refer to the time points on the nutritional and immune trajectories where the rate or direction of change undergoes a sustained change, i.e., the turning point where the trajectory transitions from the first stage of change to the second stage.

[0031] Specifically, before fitting, each indicator can be standardized within individuals to eliminate the influence of absolute differences in levels among different mothers and highlight the relative changes in colostrum within the postpartum period for the same mother. Then, using postpartum physiological time as the independent variable, a weighted local fitting algorithm is used to construct continuous nutritional trajectories N(t) and immune trajectories I(t). When determining the phase transition time, candidate times in the trajectory are selected sequentially, and the first fitting residual for single-segment fitting is calculated for each. And the second fitting residual when performing two fitting operations before and after the candidate time point. And calculate the residual improvement rate. ,in This is a non-zero constant. If the residual improvement rate at a candidate time point reaches a preset improvement threshold, and the difference in trajectory slope before and after that time point reaches a preset slope threshold, while the trend of change after the phase transition continues for a preset duration, then that candidate time point is determined as the corresponding phase transition time. For example, if the residual of the piecewise fitting of the nutritional trajectory is significantly lower than that of the single-segment fitting around 41 hours postpartum, and the slope changes from a rapid decrease to a gradual decrease, then that time point is determined to be the nutritional phase transition time. In this way, the specific time points at which the nutritional and immune components of colostrum undergo phase transitions can be accurately identified, while preserving the possible asynchrony between the two, avoiding the limitations of a single threshold judgment.

[0032] Step 3: Based on the actual feeding time and intake, convert the nutritional trajectory and immune trajectory into the actual nutritional exposure trajectory and immune exposure trajectory received by the newborn, respectively. The nutritional exposure trajectory refers to the time series of the cumulative intensity of nutritional stimuli received by the newborn at different times, while the immune exposure trajectory refers to the time series of the cumulative intensity of immune stimuli received by the newborn at different times. These two trajectories are calculated based on the time of each feeding, the amount ingested, and the nutritional / immune status value corresponding to the colostrum fed, combined with a response function that decays over time.

[0033] Specifically, for the first The actual feeding time was determined. Intake and the nutritional status values ​​corresponding to the colostrum fed. and immune status value Nutritional exposure trajectory and immune exposure trajectory According to the formula and Confirmed, among which and These are the response functions for changes in nutritional and immune exposure over time after feeding, respectively. When the state collection time is earlier than the corresponding feeding time, the exposure contribution of that feeding is zero. For example, if a feeding occurs 20 hours postpartum with an intake of 10 ml, and the nutritional status value of the colostrum it contains is 0.5, then this feeding will have a positive exposure contribution for time points after 20 hours, but a zero contribution for time points before 20 hours. By introducing the actual feeding time and intake for conversion, alignment from the milk collection time axis to the intake time axis is achieved, effectively avoiding the use of colostrum test results from newborns who have not yet been fed to interpret previously observed changes in newborn status, and resolving the data mismatch problem caused by inconsistent sampling times from multiple data sources.

[0034] Step 4: Perform individual baseline correction on the neonatal status data, construct the neonatal adaptation status trajectory, and determine the state phase transition time of the adaptation status trajectory; Individual baseline correction can refer to subtracting the baseline value of the newborn before receiving colostrum exposure, or the baseline value within a preset period before each feeding, to eliminate the influence of individual differences in initial state. The newborn adaptation trajectory is a comprehensive state change curve formed after baseline correction and the integration of fast and slow response characteristics. The state phase transition point can refer to the turning point in the newborn adaptation trajectory from the initial fluctuating stage to a relatively stable stage, or from the stable stage to a stage of continuous deviation.

[0035] Specifically, neonatal status indicators are first divided into fast-response indicators (such as heart rate, blood oxygen, and body temperature) and slow-response indicators (such as weight changes and excretion changes), and fast-response trajectories and slow-response trajectories are constructed respectively. Then, based on the proportion of effective data for each indicator in the corresponding time period, a weighted fusion is performed to obtain a comprehensive neonatal adaptation status trajectory. When determining the state phase transition moment, a piecewise fitting method is also used to find the position where the trajectory change rate changes significantly and the trend continues. For example, if the neonatal adaptation trajectory shows a significant decrease in fluctuation amplitude and tends to be stable around 48 hours postpartum, and this change is not caused by a single abnormal detection point, and there is at least one effective colostrum feeding exposure near the phase transition position, then this moment is determined to be the state phase transition moment. Through individual baseline correction, the interference of initial differences such as gestational age and birth weight on the recognition results is reduced, making the recognition results more realistically reflect the dynamic adaptation process of neonates to colostrum changes.

[0036] Step 5: Based on the temporal relationship between the nutritional phase transition time, the immune phase transition time, and the state phase transition time, combined with the response time lag between the nutritional exposure trajectory, the immune exposure trajectory, and the adaptive state trajectory, as well as the degree of trajectory fluctuation after the state phase transition, determine the newborn's adaptive state.

[0037] The temporal relationship can refer to the time interval and sequence between the nutritional phase transition, the immune phase transition, and the state phase transition; the response lag can refer to the time delay corresponding to the highest correlation between the rate of change of the nutritional exposure trajectory or the rate of change of the adaptive state trajectory; the trajectory fluctuation degree can refer to stability indicators such as the variance of the adaptive state trajectory after the state phase transition, the difference in change between adjacent time points, or the fitting residual. The adaptive state of a newborn includes one of the following: synchronous adaptive state, delayed adaptive state, weak response state, trajectory asynchronous state, or unstable adaptive state.

[0038] Specifically, firstly, within a preset time lag range, candidate time lag values ​​are sequentially changed. The first degree of correspondence between the rate of change of the nutritional exposure trajectory and the rate of change of the adaptation trajectory, and the second degree of correspondence between the rate of change of the immune exposure trajectory and the rate of change of the adaptation trajectory, are calculated. The candidate time lag value with the highest degree of correspondence is determined as the nutritional response time lag and the immune response time lag, respectively. Next, the deviation between the state phase transition time and the nutritional and immune phase transition times corrected for the corresponding response time lags is calculated to determine the phase transition correspondence of the three types of trajectories. Finally, the state is determined by comprehensively considering the phase transition correspondence, the consistency of the change direction, and the degree of fluctuation after the state phase transition. For example, if both the colostrum nutritional and immune trajectories exhibit effective phase transitions, the neonatal adaptation trajectory shows a corresponding phase transition within the allowable time lag, and the state fluctuation gradually decreases after the phase transition with high trend consistency, it is determined to be a synchronous adaptation state. If the neonatal adaptation trajectory phase transition is significantly later than the colostrum trajectory phase transition and exceeds the reference time lag range, it is determined to be a delayed adaptation state. If colostrum exposure changes significantly but the neonatal response gain is lower than the reference range, it is determined to be a weak response state. Through the coupled analysis of multi-dimensional features, not only is the accurate identification of the neonatal adaptation status achieved, but the identification results are also given strong interpretability, which can clearly indicate whether the abnormal status is due to delayed nutritional phase transition, premature immune phase transition or insufficient neonatal response.

[0039] Example 2: In another embodiment, a method is provided for determining the quality weights of each detection value and adjusting the trajectory fitting contribution, comprising: Step 1: Multiple sampling points postpartum, including no fewer than four valid sampling points; In this context, effective sampling points refer to sampling time nodes that, after quality verification, are confirmed as usable for constructing colostrum nutritional and immune trajectories. Setting at least four effective sampling points is to meet the basic requirements of the piecewise fitting algorithm for data degrees of freedom, ensuring the ability to distinguish between single-segment linear changes and multi-segment phase transitions. If there are fewer than four effective sampling points, it becomes statistically difficult to significantly distinguish between random fluctuations and actual physiological stage transitions. For example, with only three sampling points, any two points can form a straight line, making it impossible to calculate the residual improvement of piecewise fitting compared to single-segment fitting, thus failing to accurately pinpoint the phase transition time. However, with six sampling points (e.g., 6h, 18h, 34h, 55h, 82h, and 96h postpartum), the system can fully capture the nonlinear trend of colostrum components changing from an initial high concentration to a lower concentration in transitional milk. By setting this threshold, sufficient temporal resolution can be provided for subsequent trajectory reconstruction, avoiding misjudgments of phase transition locations due to data sparsity.

[0040] Step 2: Before performing trajectory fitting, determine the mass weight of each detection value based on the interval between collection and detection of the colostrum sample, storage temperature, number of freeze-thaw cycles, coefficient of variation of parallel detections, and calibration deviation of the detection equipment, and adjust the contribution of each detection value to trajectory fitting according to the mass weight. The quality weight is a numerical coefficient used to quantify the reliability of each colostrum detection data point, typically ranging from 0 to 1. This weight is determined through a comprehensive evaluation of five key dimensions: First, the longer the interval between collection and detection, the higher the risk of degradation of bioactive components, resulting in a lower weight; second, if the storage temperature deviates from the preset cold chain range (e.g., above -20℃), protein denaturation will occur, further reducing the weight; third, the more freeze-thaw cycles, the more severe the cell structure damage, leading to an exponential decay in weight; fourth, the coefficient of variation in parallel detections reflects the repeatability of the experimental operation; a larger coefficient of variation means higher data dispersion and a lower weight; fifth, the calibration deviation of the detection equipment reflects the accuracy of the instrument system; a larger deviation results in a smaller weight. Specifically, the quality weight Q_(i,j) can be obtained by multiplying or weighted summing the sub-weights of the above five factors. For example, a sample stored at an ideal temperature but subjected to three freeze-thaw cycles with a parallel detection coefficient of variation as high as 15% might have a final quality weight of 0.4; while another sample, collected and tested rapidly without freeze-thaw cycles and with a coefficient of variation of only 2%, could have a quality weight of 0.95. For extreme outliers with quality weights below a certain threshold, the system can mark them as invalid and perform constrained interpolation instead of directly participating in the fitting. This mechanism effectively eliminates noise interference introduced by improper sample preservation or detection errors, ensuring that the generated nutrient and immune trajectories truly reflect the biological evolution of colostrum components, rather than the accumulation of experimental errors. This lays a solid data foundation for the subsequent accurate identification of phase transition moments.

[0041] Example 3: In another optional embodiment, this application also provides a specific implementation method for intra-individual standardization of colostrum nutritional and immune indicators. This method aims to address the problem of significant differences in the absolute levels of colostrum components among different mothers, avoiding the technical shortcomings of directly comparing across individuals or using uniform threshold normalization to mask the dynamic changes within the same mother.

[0042] Step 1: Perform intra-individual standardization of colostrum nutritional and immune indicators for the same mother-infant pair. The sampling time point of the first Standardized values ​​of each indicator It is determined according to the following formula: ; in, The corresponding detection value can refer to the value at the th detection point. In the colostrum samples collected at the first sampling time point, for the first The raw concentration or count value of a specific indicator (such as total protein, secretory immunoglobulin A, etc.) obtained by laboratory testing. For the first The median of a metric across multiple sampling time points can refer to the median of the metric across all valid postpartum sampling time points (e.g., at least four time points) for the mother-infant pairing. After sorting the test values ​​of each indicator by size, the value in the middle position is used as the central tendency reference for that indicator for the mother, in order to eliminate inherent differences in baseline levels between individuals. For the first The interquartile range of an indicator at multiple sampling points can refer to the difference between the third quartile and the first quartile of the indicator at all valid sampling points. It is used to characterize the dispersion or fluctuation range of the indicator for the mother. Compared with the standard deviation, the interquartile range has stronger robustness to extreme outliers. This is a non-zero constant, an extremely small positive number (e.g., 10^-6), used to prevent division by zero errors when the interquartile range is zero due to identical values ​​across all sampling points. The above formula transforms the original absolute test value into a relative variation with respect to the mother's own distribution. A value greater than 0 indicates that the indicator level at that point in time is higher than the mother's average level; if A value less than 0 indicates a level below the mother's average. For example, consider a postpartum woman's lactoferrin levels at 6h, 18h, 34h, 55h, and 82h postpartum: 2.1 mg / mL, 1.8 mg / mL, 1.2 mg / mL, 0.9 mg / mL, and 0.8 mg / mL respectively. The median is 1.2 mg / mL, and the quartile range is 0.9 mg / mL. Let's define... If the value is 0.001, then the standardized value Z at 6 hours postpartum is approximately 0.999 (2.1-1.2) / (0.9+0.001), indicating that the lactoferrin level at this time is significantly higher than the average level for this mother. However, at 82 hours postpartum, Z is approximately -0.444 (0.8-1.2) / (0.9+0.001), indicating that the level is significantly lower than the average. This approach allows subsequent trajectory construction to focus on the process of change rather than absolute highs and lows, effectively shielding against baseline shifts caused by individual differences in the mother's physical condition. This step provides calibrated, statistically robust input data for subsequent multi-indicator fusion and trajectory fitting, resulting in a more reasonable weight allocation among different indicators.

[0043] Example 4: In another optional embodiment, a neonatal state recognition method based on the fusion of colostrum nutritional and immune features is provided, including the steps of determining index weights and performing local fitting to obtain a trajectory.

[0044] Step 1: Determine the weight of each indicator based on the number of valid sampling points, the missing rate, and the stability of parallel detection; The indicator weights are used to quantify the reliability contribution of each colostrum nutritional or immune indicator in subsequent trajectory construction. Specifically, the determination of indicator weights comprehensively considers the completeness of the data and the repeatability of the detection: the number of valid sampling points reflects the coverage density of the indicator on the postpartum timeline; the more points, the higher the base weight. The missing proportion characterizes the continuity of the data sequence; the lower the missing proportion, the less affected the indicator is by sampling omissions, and the weight is correspondingly increased. Parallel detection stability is usually reflected by the coefficient of variation (CV) of parallel sample detection results; the smaller the CV, the more stable the detection system is in measuring the indicator, and the greater the weight is assigned. For example, for the indicator secretory immunoglobulin A, if there are valid detection values ​​in all five sampling time points (missing proportion is 0), and the CV of three parallel detections is less than 5%, its indicator weight will be set to a higher value; conversely, if a cytokine indicator has missing data at two time points due to easy degradation and large fluctuations in parallel detections, its indicator weight will be significantly reduced. This multi-dimensional quality assessment mechanism can avoid the undue interference of low-quality or high-noise single indicators on the overall state assessment, and ensure that the subsequent fusion results are mainly driven by high-credibility indicators.

[0045] Step 2: The standardized values ​​of colostrum nutritional indicators are fused according to the corresponding indicator weights to obtain the nutritional status values ​​at each sampling time point, and the standardized values ​​of colostrum immune indicators are fused to obtain the immune status values ​​at each sampling time point. The fusion process refers to using the determined indicator weights to perform a weighted aggregation operation on the standardized values ​​of each indicator calculated in the above embodiments. Specifically, for each sampling point, the standardized values ​​of all colostrum nutritional indicators at that point are multiplied by their corresponding indicator weights and then summed to generate a scalar nutritional status value, which represents the overall level of colostrum nutritional characteristics at that moment. Similarly, the standardized values ​​of all colostrum immune indicators at that point are weighted and summed to generate an immune status value. This weighted fusion method differs from a simple arithmetic average; it ensures that highly reliable indicators (such as total protein, lactoferrin, and other relatively stable components) dominate the status value, while the contribution of low-reliability indicators is suppressed. For example, at a certain sampling point, if the weight of the fat indicator is 0.4 and the weight of a certain trace amino acid indicator is only 0.05, the final nutritional status value will be mainly determined by the standardized change trend of the fat indicator. The resulting nutritional and immune status values ​​not only eliminate the influence of different indicator dimensions and individual baseline differences, but also embed data quality information, providing high-quality discrete data points for constructing smooth and realistic biological trajectories.

[0046] Step 3: Using the postpartum physiological time relative to the delivery time at each sampling point as the independent variable, locally fit the nutritional status value and the immune status value to obtain the nutritional trajectory and the immune trajectory.

[0047] Local fitting aims to address trajectory reconstruction distortion caused by irregular sampling times (e.g., non-uniform intervals such as 6 hours, 18 hours, and 55 hours postpartum). Specifically, this method abandons the equidistant fitting approach using sample number as the independent variable, instead strictly using postpartum physiological time (i.e., the time difference between sampling time and delivery time) as the horizontal axis. Based on this, a local regression algorithm (such as weighted local scatter smoothing fitting LOESS or piecewise spline interpolation) is used, with the obtained discrete nutritional and immune status values ​​as the vertical axis observations, to construct a continuous function curve. During the fitting process, sampling points closer to the target estimation time contribute more to the fitting result; simultaneously, combined with the aforementioned quality weights, the pulling effect of low-quality data points on the curve shape is further reduced. Through this local fitting based on the true time axis, the true dynamic changes in colostrum nutritional and immune components over postpartum time can be accurately reproduced, capturing key stage characteristics such as the rapid decline phase and plateau phase, avoiding spurious fluctuations or turning points caused by uneven sampling intervals, and laying a solid data foundation for subsequent accurate identification of phase transition moments.

[0048] Example 5: In another embodiment, this application also provides a specific implementation method for determining the timing of the nutrient phase transition or the immune phase transition.

[0049] Step 1: Select candidate phase transition moments from the nutrient trajectory or immune trajectory in sequence, and calculate the first fitting residual when the trajectory is fitted using a single segment. And the second fitting residuals when fitting before and after the candidate phase transition time, respectively. ; Candidate phase transition moments can refer to any point on the postpartum physiological timeline where the nutritional or immune status values ​​of colostrum may undergo a phase transition. In practice, the system can generate a series of candidate moments by traversing between valid sampling points at a preset time step (e.g., every 1 hour or every 30 minutes). First fitting residual This can refer to the sum of squared errors between the actual detected values ​​and the fitted curve when the trajectory data over the entire observation period is treated as a whole and fitted using a single mathematical model (such as linear regression or polynomial fitting). It reflects the overall fitting deviation without considering the stage transition assumption. The second fitting residual... This could refer to dividing the trajectory data into two sub-intervals, with the currently selected candidate phase transition time as the boundary, and then independently fitting each of these two sub-intervals, resulting in the sum of the two fitting residuals. Through comparison... and This allows for the quantification of the improvement in trajectory description accuracy after introducing the segmentation assumption. For example, if a mother and infant's nutritional trajectory shows a rapid decline before 30 hours postpartum and then stabilizes after 30 hours, then the trajectory calculated at candidate time t=30h... will be significantly smaller than This indicates that the moment is very likely the actual phase transition point.

[0050] Step 2: Calculate the residual improvement rate using the following formula. : ;in, It is a non-zero constant; Among them, residual improvement rate It is a statistical metric used to measure the improvement in goodness of fit of piecewise fitting compared to single-piece fitting. The numerator in the formula... The denominator represents the total reduction in fitting error due to the introduction of the phase transition point. As a normalization factor, it eliminates the influence of different indicator dimensions and absolute error magnitude on the judgment result. For a very small nonzero constant (e.g., 10), -6 Its function is to prevent the first fitting residual from being... A division-by-zero error occurs when the result approaches zero, ensuring the numerical stability of the calculation. The value typically ranges from 0 to 1. The closer the value is to 1, the more significant the error reduction brought about by piecewise fitting, meaning the higher the reliability of this candidate time point as a phase transition time. For example, when It is 0.85. It is 0.20, and When the value is 0.01, P≈0.76 is calculated, indicating that this candidate time can explain about 76% of the original fitting error and has a very high probability of phase transition.

[0051] Step 3: The candidate phase transition moment is determined as the phase transition moment of the corresponding trajectory if the residual improvement rate reaches the preset improvement threshold, the difference in trajectory slope before and after the candidate phase transition moment reaches the preset slope threshold, and the trend of change after the phase transition continues for a preset duration.

[0052] Determining the final phase transition time requires simultaneously satisfying constraints in three dimensions to ensure the statistical significance, geometric abruptness, and physiological rationality of the identification results.

[0053] First, regarding statistical significance, the calculated residual improvement rate P must be greater than a preset improvement threshold (e.g., 0.4 or 0.5). This condition ensures that the identified inflection point is not a small fluctuation caused by random noise, but a statistically significant structural change.

[0054] Secondly, regarding geometric abruptness, the absolute value of the difference between the fitted slope of the trajectory before and after the candidate phase transition moment must reach a preset slope threshold. The slope represents the rate of change of nutritional or immune components over time, and the constraint on the slope difference aims to capture the fundamental shift in change, such as a change from a rapid decline to a slow plateau, or from a stable state to a rapid rise. If the slope difference is too small, even if the residuals improve, it may only be a slight curvature of the curve rather than a true phase transition.

[0055] Finally, regarding physiological rationality, the trend of change after the phase transition must persist for a preset duration. This is to filter out transient disturbances that occur momentarily and recover quickly (such as brief fluctuations in indicators caused by a single feeding). Only when the new trend of change is maintained for a sufficiently long time (e.g., lasting at least 6 hours or covering at least two subsequent sampling points) is it considered to have entered a new physiological phase.

[0056] Through the above multi-dimensional collaborative judgment, it is possible to effectively avoid misjudging random errors in colostrum detection as physiological stage transitions, thereby providing a high-confidence time benchmark for subsequent calculation of the nutrition-immunity phase transition interval and neonatal response lag.

[0057] Example 6: In another alternative embodiment, a method is provided for converting a trajectory into an exposed trajectory and superimposing contributions, comprising the following steps: Step 1: Determine the nutritional and immune status values ​​of the colostrum used in each actual feeding; The nutritional status value can refer to the quantitative representation of the nutritional trajectory constructed above at the corresponding sampling time point, reflecting the comprehensive level of various nutrients in the colostrum used in this feeding. The immune status value can refer to the quantitative representation of the immune trajectory at the corresponding sampling time point, reflecting the comprehensive activity of various immune factors in the colostrum used in this feeding. Since there is often a time difference between the colostrum collection time and the actual feeding time of the newborn, and the bioactivity of colostrum may decrease or change over time during storage, this step does not directly use the original detection value at the time of collection. Instead, based on the time interval between the actual feeding time and the collection time, combined with the storage conditions of the colostrum (such as temperature and freeze-thaw cycles), the data on the nutritional and immune trajectories are interpolated or corrected to determine the true status value of the colostrum at the time of the actual feeding. For example, if a feeding occurs 20 hours postpartum, and the most recent colostrum sampling points are 18 hours and 34 hours postpartum, then the values ​​at t=20h are extracted from the nutritional trajectory N(t) and immune trajectory I(t) through linear interpolation or spline interpolation, and used as the nutritional status values ​​for this feeding. and immune status value In this way, feature loss caused by sampling discreteness can be eliminated, ensuring the continuity and accuracy of the input data.

[0058] Step 2: Based on the feeding time, intake, and time elapsed after each feeding, determine the contribution of each feeding to nutritional exposure and immune exposure at different collection times. Among them, feeding time This can refer to the actual time point at which the newborn begins to ingest colostrum, used as the time series origin for exposure calculations; intake amount This can refer to the actual volume or mass of colostrum swallowed by the newborn during that feeding, used to reflect the dose-response effect; the elapsed time after feeding can refer to the time t between the current state data collection time and the feeding time. Time difference between This step involves introducing a response function. and This is used to simulate the metabolic dynamics of nutrient and immune components in newborns. Specifically, when the state acquisition time t is earlier than the feeding time... When t ≥ 0, the exposure contribution of that feeding is set to zero to strictly ensure causal logic and prevent the use of data from colostrum that has not yet been ingested to explain changes in the neonatal state that have already occurred; At that time, the contribution of exposure was determined by intake. State value ( or and response function value Jointly determined. Response function. and It typically exhibits a curve that decays over time, such as an exponential decay function. Or a piecewise linear function, its decay rate The settings can be configured based on the half-life or historical metabolic data of different components. For example, for immune factors such as lactoferrin, their retention time in the digestive tract is relatively short, corresponding to... The function decays relatively quickly; however, for some long-chain fatty acids, their absorption and metabolism are slower, corresponding to... The function decays slowly. This calculation method can transform discrete feeding events into continuous, time-dependent exposure signals, accurately characterizing the intensity of biostimuli actually received by newborns at different times.

[0059] Step 3: The nutritional exposure contribution and immune exposure contribution corresponding to each actual feeding are summed to obtain the nutritional exposure trajectory and immune exposure trajectory; Here, "overlay" refers to the summation of the exposure contributions from all actual feeding events that occurred during the study period over time. Since newborns typically experience multiple feedings during the observation period, and the composition, intake, and timing of each feeding differ, the total exposure level at any given time t is the vector sum of all historical feeding contributions. Specifically, this refers to the nutritional exposure trajectory. It is obtained by summing the nutritional exposure contributions generated at time t from the first to the Rth feedings; similarly, the immune exposure trajectory is... This is achieved by summing the immune exposure contributions from each feeding. This process maps discrete feeding event sequences to continuous physiological exposure trajectories. For example, if three feedings were performed at 6, 18, and 34 hours postpartum, the remaining contributions from each feeding after decaying over different time periods must be added when calculating the exposure trajectory value at 40 hours postpartum. The resulting nutritional and immune exposure trajectories completely decouple the colostrum sampling timeline from the neonatal physiological timeline, eliminating data mismatch issues caused by inconsistencies between sampling and feeding times. This allows subsequent state identification to be based on the actual nutritional and immune load received by the neonate, significantly improving the biological rationality and interpretability of the identification results.

[0060] Example 7: In another embodiment, this application also provides a specific implementation method for calculating the neonatal nutrition and immune exposure trajectory based on the response function, which mainly includes two core steps: determining the contribution of a single feeding and superimposing to generate the trajectory.

[0061] Step 1: The nutritional exposure trajectory and the immune exposure trajectory are determined according to the following formulas: ; ;in, For the nutrient exposure trajectory, For the immune exposure trajectory, This refers to the actual number of feedings. For the first The actual feeding time. For the first The corresponding intake amount for each actual feeding. and These are the nutritional status and immune status values ​​corresponding to the colostrum fed to the individual. and These are the response functions for nutritional exposure and immune exposure over time after feeding, respectively. In the above formula, each parameter has a clear physical meaning and data source. R can refer to the total number of effective feedings completed by the same mother-infant pair within the observation period, and its value comes from the actual feeding record log. It is the absolute moment when the r-th feeding occurs, and is usually converted to postpartum physiological time with the time of delivery as the time origin. It is the actual volume or mass of colostrum ingested by the newborn during this feeding, used to quantify the intensity of exposure. and It is not the original detection concentration, but the nutritional status and immune status values ​​of the colostrum fed in this batch after the aforementioned intra-individual standardization and index fusion processing, reflecting the relative level of this batch of colostrum in the overall change trajectory. and It is the core response function used to describe the effects of nutrients and immune components on the newborn's body over time after feeding. The decline in bioavailability or metabolic transformation that occurs over time. For example, It can be set as an exponential decay function. ,in This represents the half-life or average residence time of a specific nutrient in a newborn's body; similarly, The corresponding decay curves can be set based on the metabolic characteristics of macromolecules such as immunoglobulins. Through... and (or The product of () and () determines the theoretical maximum exposure load for a single feeding, which is then multiplied by the response function. (or This approach transforms discrete feeding events into exposure contributions that vary continuously over time. This allows the model to distinguish between different exposure patterns: high intake with rapid metabolism and low intake with long duration.

[0062] Step 2: When the status acquisition time is earlier than the corresponding feeding time, the exposure contribution of this feeding is zero.

[0063] This constraint establishes the causal boundary for exposure calculations, meaning that only feeding behaviors that have already occurred can contribute to the current neonatal state. In practice, the system compares the current state acquisition time t with all historical and planned future feeding times in real time. If t < This means that the feeding has not yet occurred, and the nutrients and immune components in the colostrum have not yet entered the newborn's body. Therefore, the corresponding term in the summation is forcibly set to zero. This mechanism effectively avoids logical errors caused by disordered data recording order or the intervention of predictive data, ensuring the exposure trajectory. and The value of t at any given time is the sum of all feeding events completed before time t. For example, if neonatal status data is collected 20 hours postpartum, and the third feeding is scheduled for 24 hours postpartum, then the value of the third feeding corresponds to... , , Although the parameters have been recorded, their contribution is not considered when calculating the exposure value at t=20. These events are directly ignored. The calculation of the response function value and cumulative contribution is only performed when the data collection time is later than the feeding time; otherwise, the feeding event is skipped. This ensures that the constructed exposure trajectory strictly conforms to physiological facts, providing an accurate time reference for subsequent analysis of the neonatal state's response time lag to colostrum exposure.

[0064] Example 8: In one possible implementation, this application also provides a specific method for constructing a neonatal adaptation state trajectory. This method addresses the significant differences in the timescale of response to colostrum exposure among different physiological indicators of newborns. Through classification processing and adaptive fusion, it eliminates interference from individual innate differences and constructs a state trajectory that accurately reflects the dynamic adaptation process.

[0065] Step 1: Obtain newborn status data and divide it into fast response status data and slow response status data. Fast response status data includes at least one of heart rate stability, blood oxygen stability, body temperature change, and feeding completion rate. Slow response status data includes at least one of weight change, feeding interval change, and excretion status. Rapid response status data refers to physiological or behavioral indicators that change significantly within minutes to hours after colostrum intake. These indicators serve to capture the newborn's acute response to nutrition and immune components in real time. For example, heart rate stability can be quantified by calculating the standard deviation of heart rate variability per unit time; blood oxygen stability can be determined based on the variance of blood oxygen saturation fluctuations; body temperature changes are recorded as the deviation of body surface temperature from a set threshold; and the feeding completion rate is defined as the ratio of actual intake to planned intake. These indicators are typically sampled at high frequencies and can sensitively reflect short-term physiological fluctuations.

[0066] Slow-response data refer to indicators that require hours to days of cumulative effect to show a trend of change. Their purpose is to assess a newborn's long-term growth adaptability and metabolic regulation. For example, weight changes are typically measured daily or every other day, reflecting long-term energy balance; changes in feeding intervals record the trend of time between two successful feedings, reflecting the rhythm of hunger recovery; and excretory status includes changes in urination frequency and stool consistency, reflecting the maturity of digestive and absorptive functions.

[0067] Specifically, the system extracts time-series data of the aforementioned indicators from the mother-infant pairing database. For the same mother-infant pair, fast response data may be distributed at minute or hourly densities, while slow response data may be distributed at hourly or daily densities. This classification avoids feature ambiguity caused by directly mixing high-frequency fluctuation signals with low-frequency trend signals, ensuring that subsequent analysis can separately capture acute stress responses and chronic adaptation processes.

[0068] Step 2: Subtract the corresponding individual baselines from the fast response state data and the slow response state data respectively to construct the fast response trajectory and the slow response trajectory; The individual baseline can refer to the basic state level exhibited by the newborn before being exposed to colostrum at the current stage, or at the beginning of a specific observation window. The purpose of subtracting the individual baseline is to eliminate the differences in initial state caused by congenital factors such as gestational age, birth weight, and delivery method, so that the constructed trajectory only reflects the dynamic change relative to its own baseline level, rather than the magnitude of the absolute value.

[0069] For rapid response status data, the individual baseline can be the average of various indicators within a preset period before each feeding (such as 30 minutes before feeding), or the stable value in the first hour after birth. Specifically, if a premature infant's baseline heart rate is naturally lower than that of a full-term infant, directly comparing absolute heart rate values ​​will misjudge its status. However, after deducting its own baseline heart rate, the resulting heart rate change curve can truly reflect its response to feeding stimuli.

[0070] For slow-response state data, the individual baseline can be the mean measurement on the first day after birth or the previous valid measurement. For example, when constructing a weight change trajectory, the birth weight or the average weight on the first day is subtracted from the weight measurements at each subsequent time point to obtain a cumulative weight change curve; when constructing an excretion state trajectory, the baseline frequency is subtracted from the current urination frequency to obtain an excretion increment sequence.

[0071] After baseline correction, the corrected fast response data and slow response data are time-aligned and smoothed to generate continuous fast response trajectories. and slow response trajectory These two trajectories represent the immediate adaptive behavior of newborns on a short timescale and the cumulative adaptive trend on a long timescale, respectively, and both have eliminated the influence of inherent individual differences.

[0072] Step 3: Weighted fusion is performed based on the proportion of effective data of the two types of state data in the corresponding time period to obtain the newborn adaptation state trajectory.

[0073] Weighted fusion can refer to using a weighting coefficient that changes over time. By linearly combining the fast response trajectory and the slow response trajectory, the final neonatal adaptation state trajectory is formed. The weighting coefficient It is dynamically determined based on the effective data ratio of fast response state data and slow response state data within a time window near the current time t.

[0074] The proportion of valid data reflects the completeness and reliability of the data. At certain times, due to sensor malfunctions or interruptions in care procedures, fast response data may be missing. In such cases, the proportion of valid fast response data decreases, and the weighting coefficient is affected. Automatic reduction makes the fusion results rely more on the relatively complete slow response trajectory; conversely, during the interval when the slow response index has not yet generated new measurements, the weighting coefficients... The increase makes the trajectory primarily dominated by high-frequency, fast-response data.

[0075] For example, in the first 24 hours postpartum, weight change data may only have 1-2 sampling points, resulting in a very low effective proportion, while heart rate and blood oxygen data are continuous and complete. Approaching 1, the adaptive trajectory primarily exhibits rapid response characteristics; as time progresses, weight and excretion data gradually become more abundant. Gradually adjust to the equilibrium value, the trajectory takes into account both short-term fluctuations and long-term trends.

[0076] The neonatal adaptation trajectory obtained through this adaptive weighted fusion mechanism It possesses both high sensitivity to acute physiological changes and robustness to long-term development trends, effectively solving the technical problem that single-timescale modeling cannot fully depict the complex adaptation process of newborns, and providing a high-quality data foundation for subsequent determination of state phase transition moments and analysis of multi-trajectory coupling relationships.

[0077] Example 9: In one possible implementation, a method for determining response delay and phase transition correspondence is provided, comprising: sequentially changing candidate delay values ​​within a preset delay range; calculating a first correspondence between the rate of change of the nutrient exposure trajectory and the rate of change of the adaptive state trajectory, and a second correspondence between the rate of change of the immune exposure trajectory and the rate of change of the adaptive state trajectory; determining the candidate delay value corresponding to the highest first correspondence as the nutrient response delay, determining the candidate delay value corresponding to the highest second correspondence as the immune response delay, and determining the phase transition correspondence of the three types of trajectories based on the deviation between the state phase transition time and the nutrient phase transition time and the immune phase transition time after correction by the corresponding response delay.

[0078] Step 1: Within the preset time delay range, change the candidate time delay values ​​sequentially, and calculate the first degree of correspondence between the rate of change of the nutritional exposure trajectory and the rate of change of the adaptive state trajectory, and the second degree of correspondence between the rate of change of the immune exposure trajectory and the rate of change of the adaptive state trajectory. The preset time lag range is a pre-defined time interval based on the physiological characteristics of newborns' absorption and metabolism of colostrum nutrients and regulation of the immune system, such as 2 hours to 72 hours postpartum. Candidate time lag values ​​are time offsets traversed within this range at fixed step sizes (e.g., 0.5 hours or 1 hour). The rate of change can refer to the first derivative of the trajectory data with respect to time, used to characterize the dynamic rate of change in nutritional exposure, immune exposure, or neonatal adaptation status per unit time, thereby eliminating the influence of individual baseline absolute value differences and focusing on matching trends. Both the first and second degree of correspondence use Pearson correlation coefficients or mutual information to quantify the linear correlation or statistical dependence of the two rate of change curves at different time offsets. Specifically, for each candidate time lag value... Tracing the nutritional exposure path Translate forward get Calculate its rate of change With the rate of change of the adaptive state trajectory The correlation coefficient is used as the first degree of correspondence; similarly, the immune exposure trajectory is used as the first degree of correspondence. Translate forward get Calculate its rate of change and The correlation coefficient is used as the second degree of correspondence. By traversing all candidate time delays and constructing time delay-correlation curves, the independent biological response timelines of nutrient absorption and immune regulation can be automatically captured, avoiding matching failures caused by fixed time delay assumptions.

[0079] Step 2: Determine the candidate time delay corresponding to the highest first degree of correspondence as the nutritional response time delay, and the candidate time delay corresponding to the highest second degree of correspondence as the immune response time delay. Based on the deviation between the state phase transition time and the nutritional and immune phase transition times after correction by the corresponding response time delay, determine the phase transition correspondence of the three types of trajectories.

[0080] Among them, nutrient response lag It is the candidate time delay that causes the first correspondence to reach its peak, characterizing the average physiological delay time required from changes in colostrum nutrient intake to a significant response in the newborn's state; immune response time delay. This is the candidate time delay at which the second corresponding degree reaches its peak, characterizing the delay time for the immune component to take effect. Since immune regulation typically involves a cytokine cascade, Often greater than The nutrient phase transition time after corresponding response time delay correction is: The immune phase transition time is ,in and These represent the phase transition times of the originally detected nutritional and immune trajectories, respectively. The degree of correspondence between the phase transitions of the three types of trajectories is calculated by determining the state phase transition times. Absolute deviation from the corrected colostrum phase transition time and This method, which determines time lag and corrects phase transition timing by maximizing the correlation of rate of change, effectively solves the temporal mismatch problem caused by inconsistent sampling times of multi-source data and lag in physiological response, providing a precise temporal alignment basis for accurately identifying the adaptation status type of newborns.

[0081] Example 10: In one optional implementation, a method for determining the type of adaptation state and outputting the supplementary sampling time in a neonatal state recognition method based on the fusion of colostrum nutritional and immune features is provided, including specific steps for determining the neonatal adaptation state based on multidimensional features.

[0082] Step 1: Based on the degree of phase transition correspondence, the consistency of the direction of change between the nutritional exposure trajectory and the immune exposure trajectory and the adaptive state trajectory, and the degree of fluctuation of the adaptive state trajectory after the phase transition, determine whether the newborn is in a synchronous adaptive state, a delayed adaptive state, a weak response state, a trajectory asynchronous state, or an unstable adaptive state. Among them, the degree of phase transition correspondence can refer to the absolute value of the time deviation between the nutritional phase transition time, the immune phase transition time, and the neonatal state phase transition time after response lag correction. The smaller the deviation, the more synchronous the transformation of the three types of trajectories. The consistency of change direction can refer to whether the change trends of the nutritional exposure trajectory, the immune exposure trajectory, and the adaptive state trajectory before and after the phase transition are in the same direction in terms of sign. For example, when nutritional exposure increases and immune exposure increases, the adaptive state trajectory should also show an improving trend. The degree of fluctuation can refer to the cumulative value of the variance of the adaptive state trajectory within the preset observation window or the difference in change between adjacent time points after the state phase transition time, which is used to characterize the stability of the newborn after entering a new stage.

[0083] Specifically, the system uses a preset rule engine to logically combine and judge the quantitative indicators of the above three dimensions. If the corresponding degree of phase transition is less than the preset time threshold, and the direction of change is completely consistent, while the fluctuation after the phase transition is lower than the stability threshold, it is judged as a synchronous adaptation state, indicating that the newborn's response to the colostrum change is timely and stable. If any deviation in the corresponding degree of phase transition exceeds the time threshold, that is, the newborn's phase transition is significantly lagging behind the colostrum phase transition after time lag correction, it is judged as a delayed adaptation state. If the nutritional or immune exposure trajectory changes significantly (the rate of change exceeds the threshold), but the change amplitude of the adaptation state trajectory is extremely small (the response gain is lower than the threshold), it is judged as a weak response state. If the interval between the nutritional phase transition time and the immune phase transition time is too large, causing the newborn's adaptation trajectory to only form an effective correspondence with one type of trajectory, while the other type of trajectory is severely mismatched, it is judged as a trajectory asynchrony state. If, although a phase transition is detected, the adaptation state trajectory after the phase transition still has continuous large oscillations or repeated switching, and the fluctuation exceeds the instability threshold, it is judged as an unstable adaptation state.

[0084] By fusing these multidimensional features, misjudgments caused by a single indicator (such as relying solely on time difference) can be avoided, enabling the output status labels to have clear clinical interpretability and directly point to specific abnormal links in the adaptation process.

[0085] Step 2: When the number of valid sampling points or the confidence level of the trajectory phase transition does not reach the corresponding threshold, output a data shortage status and determine the supplementary sampling time based on the candidate phase transition interval of the current trajectory.

[0086] Among them, the number of effective sampling time points can refer to the total number of colostrum detection points and newborn status detection points that meet the quality weight requirements; the trajectory phase transition confidence is a probability value or score calculated based on the residual improvement rate of piecewise fitting, the significance of slope change, and the phase transition duration, which is used to measure the reliability of the currently detected phase transition position.

[0087] Specifically, before classifying the states, the system first verifies the completeness and confidence of the data. If the number of valid sampling points is less than the preset minimum (e.g., less than 4), or the calculated confidence of the trajectory phase transition is lower than the preset confidence threshold, the system will skip the judgment logic of the above five states and directly output the data insufficiency state. Based on this, the system uses the currently constructed incomplete trajectory to predict the next candidate phase transition interval through local regression or trend extrapolation, and automatically calculates and recommends one or more supplementary sampling times within this interval (e.g., within the range before and after the predicted phase transition time). For example, if the current trajectory shows that a phase transition may exist between 40 and 60 hours postpartum but the data is sparse, the system can suggest supplementary sampling at the 48th hour.

[0088] This step, acting as a fallback mechanism, proactively manages the uncertainty caused by missing or low-quality data, avoiding the forced output of potentially erroneous state conclusions based on low confidence. At the same time, through precise supplementary sampling suggestions, it obtains key data to converge on the phase transition position with minimal detection cost, significantly improving the system's practicality and the reliability of the results.

[0089] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for neonatal status identification based on the fusion of colostrum nutritional and immune characteristics, characterized in that, include: S1. Obtain colostrum nutritional indicators and colostrum immune indicators at multiple sampling points after delivery for the same mother-infant pair, as well as the corresponding actual feeding time, intake and newborn status data. S2. The colostrum nutritional indicators and colostrum immune indicators are fitted according to the postpartum physiological time to obtain the nutritional trajectory and immune trajectory. The nutritional phase transition time and immune phase transition time are determined based on the degree of improvement of the residual of the segmented fitting compared with the single-segment fitting and the change of the trajectory slope before and after the segmentation. S3. Based on the actual feeding time and intake, convert the nutritional trajectory and immune trajectory into the actual nutritional exposure trajectory and immune exposure trajectory received by the newborn, respectively. S4. Perform individual baseline correction on the newborn state data, construct the newborn adaptation state trajectory, and determine the state phase transition time of the adaptation state trajectory; S5. Based on the temporal relationship between the nutritional phase transition time, the immune phase transition time, and the state phase transition time, combined with the response time lag between the nutritional exposure trajectory, the immune exposure trajectory, and the adaptive state trajectory, as well as the degree of trajectory fluctuation after the state phase transition, determine the newborn's adaptive state.

2. The method according to claim 1, characterized in that, The postpartum multiple sampling time points include no less than four valid sampling time points; Before performing the trajectory fitting, the mass weight of each detection value is determined based on the interval between collection and detection of the colostrum sample, storage temperature, number of freeze-thaw cycles, coefficient of variation of parallel detection, and calibration deviation of the detection equipment. The contribution of each detection value to the trajectory fitting is then adjusted according to the mass weight.

3. The method according to claim 2, characterized in that, Intra-individual standardization was performed on the colostrum nutritional and immune indicators of the same mother-infant pair. The sampling time point of the first Standardized values ​​of each indicator It is determined according to the following formula: ； in, For the corresponding detection value, For the first The median of an indicator at multiple sampling points For the first The interquartile differences of an indicator at multiple sampling points It is a non-zero constant.

4. The method according to claim 3, characterized in that, The weights of each indicator are determined based on the number of valid sampling points, the missing rate, and the stability of parallel detection. The standardized values ​​of the colostrum nutritional indicators are fused according to the corresponding indicator weights to obtain the nutritional status value at each sampling time point, and the standardized values ​​of the colostrum immune indicators are fused to obtain the immune status value at each sampling time point. Using the postpartum physiological time relative to the delivery time at each sampling point as the independent variable, the nutritional status value and immune status value are locally fitted to obtain the nutritional trajectory and immune trajectory.

5. The method according to claim 4, characterized in that, Determining the nutritional phase transition time or the immune phase transition time includes: Candidate phase transition moments from the nutrient trajectory or immune trajectory are selected sequentially, and the first fitting residual is calculated when the trajectory is fitted using a single segment. And the second fitting residuals when fitting before and after the candidate phase transition time, respectively. Calculate the residual improvement rate using the following formula. : ； in, It is a non-zero constant; The candidate phase transition moment is determined as the phase transition moment of the corresponding trajectory when the residual improvement rate reaches a preset improvement threshold, the difference in trajectory slope before and after the candidate phase transition moment reaches a preset slope threshold, and the trend of change after the phase transition continues for a preset duration.

6. The method according to claim 1, characterized in that, Converting the nutrient trajectory and immune trajectory into nutrient exposure trajectory and immune exposure trajectory includes: Determine the nutritional and immune status values ​​of the colostrum used in each actual feeding; Based on the feeding time, intake, and time elapsed after each feeding, the nutritional exposure contribution and immune exposure contribution of each feeding to the different collection times under different conditions were determined. The nutritional exposure contribution and immune exposure contribution corresponding to each actual feeding are summed to obtain the nutritional exposure trajectory and immune exposure trajectory.

7. The method according to claim 6, characterized in that, The nutritional exposure trajectory and the immune exposure trajectory are determined according to the following formulas: ； ； in, For the nutrient exposure trajectory, For the immune exposure trajectory, This refers to the actual number of feedings. For the first The actual feeding time. For the first The corresponding intake amount for each actual feeding. and These are the nutritional status and immune status values ​​corresponding to the colostrum fed to the individual. and These are the response functions for nutritional exposure and immune exposure over time after feeding, respectively. When the state acquisition time is earlier than the corresponding feeding time, the exposure contribution of that feeding is zero.

8. The method according to claim 1, characterized in that, The neonatal status data includes fast response status data and slow response status data. The fast response status data includes at least one of heart rate stability, blood oxygen stability, body temperature change, and feeding completion rate. The slow response status data includes at least one of weight change, feeding interval change, and excretion status. The corresponding individual baselines are subtracted from the fast response state data and the slow response state data respectively to construct the fast response trajectory and the slow response trajectory. The newborn adaptation state trajectory is obtained by weighted fusion based on the effective data ratio of the two types of state data in the corresponding time period.

9. The method according to claim 8, characterized in that, Within a preset time delay range, the candidate time delay values ​​are changed sequentially, and the first degree of correspondence between the rate of change of the nutritional exposure trajectory and the rate of change of the adaptive state trajectory, and the second degree of correspondence between the rate of change of the immune exposure trajectory and the rate of change of the adaptive state trajectory are calculated respectively. The candidate time delay corresponding to the highest first degree of correspondence is determined as the nutritional response time delay, and the candidate time delay corresponding to the highest second degree of correspondence is determined as the immune response time delay. Based on the deviation between the state phase transition time and the nutritional and immune phase transition times after correction by the corresponding response time delay, the phase transition correspondence of the three types of trajectories is determined.

10. The method according to claim 9, characterized in that, Based on the degree of phase transition correspondence, the consistency of the direction of change between the nutritional exposure trajectory and the immune exposure trajectory and the adaptive state trajectory, and the degree of fluctuation of the adaptive state trajectory after the phase transition, the newborn is determined to be in one of the following states: synchronous adaptation, delayed adaptation, weak response, asynchronous trajectory, or unstable adaptation. When the number of valid sampling points or the confidence level of the trajectory phase transition does not reach the corresponding threshold, the system outputs a data insufficiency status and determines the supplementary sampling time based on the candidate phase transition interval of the current trajectory.

Images