A method and system for intelligent assessment of nutrition for senile comorbidity

Abstract

This invention discloses an intelligent nutritional assessment method and system for elderly patients with comorbidities, relating to the field of nutritional assessment technology. By constructing a multimodal nutritional response sequence, identifying response divergence segments and establishing a shift conflict structure, calculating the shift contribution of each modality and identifying the dominant modality, and then reconstructing trends based on the dominant modality, this method transforms existing assessment methods that rely on direct comprehensive scoring or averaging into one that first identifies persistent multimodal conflict relationships and response differences, and then distinguishes and corrects the impact of shifts. This avoids the dilution of abnormal signals in highly sensitive modalities by the temporary stable performance of low-sensitive modalities. Simultaneously, it prioritizes the retention of more indicative modal information when multimodal responses are asynchronous, reducing assessment distortions caused by edema, inflammation, recording bias, and comorbid interference. Furthermore, the corrected nutritional evolution trend more accurately reflects the nutritional change process, enabling early identification and accurate judgment of nutritional risks.

Description

Technical Field

[0001] This invention relates to the field of nutritional assessment technology, specifically to an intelligent nutritional assessment method and system for comorbidities in the elderly. Background Technology

[0002] As the population ages, comorbidity of two or more chronic diseases is becoming increasingly common among the elderly. Because elderly individuals with comorbidities often experience reduced intake, decreased digestive and absorptive capacity, increased chronic inflammatory burden, weakened mobility, and multiple medications, their nutritional status is often more complex and susceptible to the combined effects of various disease factors. Current methods for assessing nutritional status in the elderly with comorbidities typically involve collecting information on weight changes, height, body mass index, dietary intake, laboratory test results, activity levels, and daily living abilities, combined with nutritional screening tools and diagnostic criteria. This comprehensive assessment determines whether the elderly with comorbidities are at nutritional risk, whether malnutrition has occurred, and its severity. Based on this assessment, a nutritional intervention plan is developed, followed by reassessment. Overall, current methods for assessing nutritional status in the elderly with comorbidities have established a comprehensive assessment pathway based on multi-source data collection, which can improve the comprehensiveness of the assessment results to a certain extent.

[0003] However, in practical applications, the inventors discovered that while existing methods for assessing nutritional status in the elderly with comorbidities incorporate data from multiple modalities, such as weight, dietary records, laboratory indicators, and activity monitoring, the response speed, accuracy, and susceptibility to comorbid factors vary among these modalities. For example, some modalities can reflect the trend of nutritional status deterioration earlier, while others exhibit response lag, observational distortion, or recording bias. In such cases, directly combining, averaging, or superimposing data from various modalities can easily dilute the abnormal signals already appearing in highly sensitive modalities with the temporary stability of low-sensitive modalities, resulting in superficially stable but actually distorted assessment results. Especially in the context of comorbidities in the elderly, due to the intertwining of factors such as edema, inflammation, cognitive decline, activity limitation, and multiple medications, different modalities are more likely to continuously exhibit inconsistent directions, asynchronous responses, and different sources of deviation. Existing technologies generally lack mechanisms to identify and correct persistent conflicting relationships, differences in response timing, and differences in the sources of deviation between multimodal data. This can easily lead to the true nutritional risks being masked during the fusion process, resulting in delayed risk identification, distorted trend judgments, and mismatches in subsequent intervention decisions. Summary of the Invention

[0004] The purpose of this invention is to solve the problems mentioned in the background art above, and to propose a method and system for intelligent assessment of nutritional comorbidities in the elderly.

[0005] A first aspect of this invention provides a method for intelligent assessment of nutritional comorbidities in the elderly, the method comprising: S1: Obtain multimodal nutrition-related data of the target elderly comorbid subjects within a continuous time range, and perform time alignment and unified processing on the multimodal nutrition-related data to construct a multimodal nutrition response sequence; S2: Based on the multimodal nutrient response sequence, identify response divergence segments where the response directions of each modality are inconsistent within a continuous time range, and construct an offset conflict structure reflecting the relationship between each modality within the response divergence segment; S3: Based on the offset conflict structure, calculate the offset contribution of each mode in the response divergence segment, and identify the dominant mode that dominates the nutrient change based on the offset contribution; S4: Based on the dominant mode, the trend of the multimodal nutrition response sequence is reconstructed to obtain the corrected nutrition evolution trend, and the nutrition assessment results of the target elderly comorbid subjects are output according to the corrected nutrition evolution trend.

[0006] Optionally, the steps for constructing a multimodal nutrient response sequence are as follows: We collected weight data, dietary intake data, laboratory indicator data, and activity monitoring data from the target elderly comorbid subjects over a continuous time period, and recorded the corresponding collection timestamp for each data point. The continuous time range is divided into multiple continuous time slices according to a preset time interval, and each modal data is mapped to the corresponding time slice according to the timestamp; when there are multiple pieces of the same modal data in a certain time slice, the average value in the time slice is taken as the modal value in the time slice; For modal data that is missing in a certain time slice, the most recent data in the adjacent time slice is used to fill the missing data. When there is no modal data in either adjacent time slice, the time slice is marked as a missing time slice and removed in subsequent processing. The modal data within each time slice are normalized according to their maximum and minimum values ​​within the continuous time range, so that the different modal data are converted to a unified numerical range. The normalized modal data within each time slice are arranged sequentially according to the time slice order to construct a multimodal nutrient response sequence.

[0007] Optionally, the steps of identifying response divergence segments where the modal response directions are inconsistent over a continuous time range, and constructing an offset conflict structure reflecting the relationship between the modes within the response divergence segments, are as follows: According to the chronological order of the time slices, the numerical change of each mode in the multimodal nutrient response sequence between two adjacent time slices is calculated, and the response direction of each mode in each time slice is determined according to the sign of the numerical change. When the numerical change is greater than zero, the response direction of the mode in the corresponding time slice is recorded as increasing; when the numerical change is less than zero, the response direction of the mode in the corresponding time slice is recorded as decreasing; when the numerical change is equal to zero, the response direction of the mode in the corresponding time slice is recorded as remaining unchanged. Within each time slice, the response directions of each mode are compared pairwise. When the response directions of any two modes are different, and the response direction of at least one of the modes is not constant, it is determined that there is a directional divergence between the two modes within the time slice. The number of mode pairs with directional divergence within each time slice is counted, and time slices with a number of mode pairs greater than zero are marked as divergence time slices. Multiple consecutive divergent time slices are merged into candidate divergent segments, and the number of mode pairs corresponding to each time slice in each candidate divergent segment is counted. When the number of mode pairs corresponding to each time slice in a candidate divergent segment is greater than zero, the candidate divergent segment is determined as the response divergent segment. For each response divergence segment, extract all mode pairs with directional divergence corresponding to each time slice within the response divergence segment, and record the start time slice, end time slice, duration slice number, and numerical change difference within the response divergence segment for each mode pair. Each mode within the response divergence segment is used as a structural node, and the mode pairs with directional divergence are used as structural connection relationships. The start time slice, end time slice, duration slice number, and numerical change difference of each mode pair are written into the corresponding structural connection relationship to construct an offset conflict structure that reflects the relationship between each mode.

[0008] Optionally, the steps for calculating the shift contribution of each mode in the response divergence segment and identifying the dominant mode of dominant trophic changes based on the shift contribution are as follows: Based on the offset conflict structure, the conflict propagation dominance index and the stable structure destruction index of each mode in the response divergence section are calculated. The conflict propagation dominance index and the stable structure destruction index are added together to obtain the offset contribution index of each mode in the response divergence section. The mode with the largest offset contribution index is taken as the dominant mode that dominates the trophic change.

[0009] Optionally, the calculation steps for the conflict propagation dominance index are as follows: For any mode to be calculated, extract all mode pairs directly connected to the mode to be calculated from the offset conflict structure, and read the start time slice and end time slice corresponding to each mode pair in the current response divergence segment. Determine the continuous time range between the start time slice and the end time slice corresponding to each mode pair as an associated conflict arc segment of the mode to be calculated, thereby obtaining the set of associated conflict arc segments of the mode to be calculated in the current response divergence segment. Combine any two different associated conflict arc segments in the set of associated conflict arc segments in pairs, calculate the time slice distance between the end time slice of the previous associated conflict arc segment and the start time slice of the next associated conflict arc segment, and add one to the time slice distance and take the reciprocal as the start and end bonding amount between the two associated conflict arc segments. The smaller the time slice distance between the two associated conflict arc segments, the larger the start and end bonding amount. Calculate the arc length of each associated conflict arc segment. The arc length of each associated conflict arc segment is obtained by subtracting the starting time slice number from the ending time slice number and then adding one. For any two different associated conflict arc segments, calculate the absolute value of the difference between their arc lengths and divide the absolute value by the sum of their arc lengths to obtain the length fold between the two associated conflict arc segments.

[0010] Optionally, the calculation steps for the conflict propagation dominance index also include: For any two different associated conflict arc segments, multiply the beginning and end bonding amount between the two associated conflict arc segments by the length folding amount to obtain the bonding and folding product value between the two associated conflict arc segments, and use the bonding and folding product values ​​between all associated conflict arc segments to form the bonding and folding product set of the mode to be calculated in the current response divergence segment. Sort the bonding and flipping product values ​​in the bonding and flipping product set according to their numerical values, and select multiple bonding and flipping product values ​​that can be connected end to end on the associated conflict arc segment number as a propagation chain. Then, multiply all the bonding and flipping product values ​​in the propagation chain continuously to obtain the strongest propagation chain value of the mode to be calculated. The time slices covered by all associated conflict arcs corresponding to the mode to be calculated in the current response divergence segment are merged and counted to obtain the total number of deduplicated covered time slices. The total number of covered time slices is then divided by the total number of time slices in the current response divergence segment to obtain the segment coverage ratio of the mode to be calculated. Multiply the strongest propagation chain value by the segment coverage ratio to obtain the propagation coverage product value. Then multiply the strongest propagation chain value by the segment coverage ratio to obtain the propagation coverage product value. Add 1 to the propagation coverage product value as the numerator and use the propagation coverage product value as the denominator. Divide the denominator by the numerator to obtain the conflict propagation dominance index of the mode to be calculated in the current response divergence segment.

[0011] Optionally, the calculation steps for the stability structure failure index are as follows: For any mode to be calculated, read the start time slice of the current response divergence segment, and search for the consecutive non-divergence time slices adjacent to the current response divergence segment in reverse time order from the start time slice. The consecutive non-divergence time slices are determined as the stable reference segments corresponding to the current response divergence segment. Within each time slice of the stable reference segment, the normalized values ​​of all modes in the time slice are sorted in ascending order. The position of the mode to be calculated in the sorting result is recorded, as well as the directly adjacent modes to the left and right of the mode to be calculated. Based on the position of the mode to be calculated in each time slice of the stable reference segment, the median is taken as the stable center position of the mode to be calculated, the minimum value is taken as the lower boundary of the stable position of the mode to be calculated, and the maximum value is taken as the upper boundary of the stable position of the mode to be calculated. At the same time, the occurrence frequency of the directly adjacent modes to the left and right of the mode to be calculated in the stable reference segment is counted respectively. The directly adjacent mode to the left with the most occurrence frequency is determined as the stable left anchor mode, and the directly adjacent mode to the right with the most occurrence frequency is determined as the stable right anchor mode. Within each time slice of the current response divergence segment, the normalized values ​​of all modes in the time slice are sorted in ascending order, and the current position of the mode to be calculated in the time slice is recorded. When the current position is between the lower boundary and the upper boundary of the stable sequence, the escape depth of the corresponding time slice is recorded as zero. When the current position is less than the lower boundary of the stable sequence, the difference between the lower boundary of the stable sequence and the current position is divided by the total number of modes minus one to obtain the escape depth of the corresponding time slice. When the current position is greater than the upper boundary of the stable sequence, the difference between the current position and the upper boundary of the stable sequence is divided by the total number of modes minus one to obtain the escape depth of the corresponding time slice.

[0012] Optionally, the calculation steps for the stability structure failure index further include: Within each time slice of the current response divergence segment, extract the current left-side direct adjacent mode and the current right-side direct adjacent mode of the mode to be calculated, and compare them with the stable left anchor mode and the stable right anchor mode respectively. The side that is different from the corresponding stable anchor mode is marked as having undergone anchor replacement. Count the number of sides that have undergone anchor replacement within the time slice, and divide the number of sides by the total number of comparable sides within the time slice to obtain the anchor replacement degree corresponding to the time slice. Within each time slice of the current response divergence segment, determine whether the stable left anchor mode appears to the right of the current mode to be calculated, and whether the stable right anchor mode appears to the left of the current mode to be calculated; count the number of anchor points that have undergone the above left-right position swap, and divide the number of anchor points by the total number of stable anchor points to obtain the anchor point mirror flip degree corresponding to the time slice; For each time slice in the current response divergence segment, first subtract the position escape depth, anchor replacement degree, and anchor mirror flip degree corresponding to the time slice from 1 respectively, then multiply the three subtraction results consecutively, and then subtract the product result from 1 to obtain the local structural damage coefficient corresponding to the time slice. Within the current response divergence zone, find the time slice in which the local structural failure coefficient of the mode to be calculated first exceeds zero, and determine this time slice as the failure initiation time slice of the mode to be calculated; at the same time, find the earliest time slice in all modes where the local structural failure coefficient exceeds zero, and determine this earliest time slice as the earliest failure initiation time slice; add one to the difference between the failure initiation time slice of the mode to be calculated and the earliest failure initiation time slice, and take the reciprocal to obtain the failure initiation leading factor of the mode to be calculated; Optionally, the calculation steps for the stability structure failure index further include: Within the current response divergence segment, count the consecutive time segments in which the local structural failure coefficient corresponding to the mode to be calculated is continuously greater than zero, and take the longest consecutive time segment as the longest consecutive failure time segment; divide the number of time slices of the longest consecutive failure time segment by the total number of time slices in the current response divergence segment to obtain the continuous failure retention ratio of the mode to be calculated. For each time slice in the current response divergence segment, subtract the local structural failure coefficient corresponding to the time slice from 1, and multiply the subtraction results for all time slices consecutively. Then, normalize the result of the multiplication according to the total number of time slices in the current response divergence segment, and then subtract the normalized result from 1 to obtain the cumulative structural failure intensity of the mode to be calculated. Multiply the failure initiation leading factor, the successive failure retention ratio, and the cumulative structural failure intensity consecutively, and take the square root of the result to obtain the stable structural failure index of the mode to be calculated in the current response divergence segment.

[0013] A second aspect of this invention provides an intelligent nutritional assessment system for comorbidities in the elderly, the system comprising: Processing module: Acquires multimodal nutrition-related data of the target elderly comorbidity subject within a continuous time range, performs time alignment and unified processing on the multimodal nutrition-related data, and constructs a multimodal nutrition response sequence; Response bifurcation module: Based on the multimodal nutrient response sequence, it identifies response bifurcation segments where the response directions of each modality are inconsistent within a continuous time range, and constructs an offset conflict structure reflecting the relationship between each modality within the response bifurcation segment; Dominant module: Based on the offset conflict structure, calculate the offset contribution of each mode in the response divergence segment, and identify the dominant mode that dominates the nutrient change based on the offset contribution; Assessment module: Based on the dominant mode, the trend of the multimodal nutrition response sequence is reconstructed to obtain the corrected nutrition evolution trend, and the nutrition assessment results of the target elderly comorbid subjects are output according to the corrected nutrition evolution trend.

[0014] The beneficial effects of this invention are: This invention proposes an intelligent method and system for assessing nutritional comorbidities in the elderly. By constructing a multimodal nutritional response sequence and further identifying response divergence segments, a shift conflict structure reflecting the relationships between different modalities is established. The shift contribution of each modality in the response divergence segments is then calculated, and the dominant modality is identified. Finally, the trend of the multimodal nutritional response sequence is reconstructed based on the dominant modality. This transforms existing assessment methods that directly score or average multimodal data into one that first identifies persistent conflict relationships and response differences between multimodalities, and then distinguishes and corrects the shift influence of different modalities. This effectively prevents nutritional abnormality signals already present in highly sensitive modalities from being detected by low-sensitivity modalities. This approach addresses the issue of dilution caused by the temporary stability of the state. Furthermore, by structurally characterizing the differences in response timing and the sources of bias and introducing a dominant modality mechanism, it prioritizes the retention of more indicative modal information about nutritional changes when multimodal data responses are asynchronous, reducing assessment distortion caused by edema, inflammation, recording bias, or comorbid interference. Moreover, by reconstructing trends to obtain corrected nutritional evolution trends, it can more realistically reflect the nutritional changes in elderly individuals with comorbidities, thereby enabling early identification and accurate judgment of nutritional risks, avoiding misjudgments of apparent stability but actual deterioration, and ultimately improving the reliability of nutritional assessment results and providing a more suitable basis for subsequent intervention decisions. Attached Figure Description

[0015] Figure 1 A flowchart of a method for intelligent nutritional assessment of comorbidities in the elderly provided in an embodiment of the present invention. Detailed Implementation

[0016] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0017] This invention provides a method for intelligent assessment of nutritional comorbidities in the elderly. See also... Figure 1 , Figure 1 A flowchart illustrating a method for intelligent nutritional assessment of comorbidities in the elderly, provided as an embodiment of the present invention. The method includes the following steps: S1: Obtain multimodal nutrition-related data of the target elderly comorbid subjects within a continuous time range, and perform time alignment and unified processing on the multimodal nutrition-related data to construct a multimodal nutrition response sequence; S2: Based on the multimodal nutrient response sequence, identify response divergence segments where the response directions of each modality are inconsistent within a continuous time range, and construct an offset conflict structure reflecting the relationship between each modality within the response divergence segment; S3: Based on the offset conflict structure, calculate the offset contribution of each mode in the response divergence segment, and identify the dominant mode that dominates the nutrient change based on the offset contribution; S4: Based on the dominant mode, the trend of the multimodal nutrition response sequence is reconstructed to obtain the corrected nutrition evolution trend, and the nutrition assessment results of the target elderly comorbid subjects are output according to the corrected nutrition evolution trend.

[0018] This invention provides an intelligent nutritional assessment method for elderly patients with comorbidities. By constructing a multimodal nutritional response sequence and further identifying response divergence segments, a shift conflict structure reflecting the relationships between modalities is established. The method then calculates the shift contribution of each modality in the response divergence segments and identifies the dominant modality. Finally, based on the dominant modality, the multimodal nutritional response sequence is reconstructed to reflect trends. This transforms existing assessment methods that directly score or average multimodal data into methods that first identify persistent conflict relationships and response differences between modalities, and then differentiate and correct the shift effects of different modalities. This effectively prevents nutritional abnormalities already present in highly sensitive modalities from being masked by less sensitive modalities. This addresses the issue of dilution caused by the temporary stability of modal data. Furthermore, by structurally characterizing the differences in response timing and the sources of bias and introducing a dominant modality mechanism, it is possible to prioritize the retention of more indicative modal information about nutritional changes when multimodal data responses are asynchronous, reducing assessment distortion caused by edema, inflammation, recording bias, or comorbid interference. Moreover, by reconstructing trends to obtain corrected nutritional evolution trends, it is possible to more realistically reflect the nutritional changes in elderly individuals with comorbidities, thereby enabling early identification and accurate judgment of nutritional risks, avoiding misjudgments of apparent stability but actual deterioration, and ultimately improving the reliability of nutritional assessment results and providing a more suitable basis for subsequent intervention decisions.

[0019] In one embodiment, S1: The steps of acquiring multimodal nutrition-related data of the target elderly comorbidity subject within a continuous time range, and performing time alignment and unified processing on the multimodal nutrition-related data to construct a multimodal nutrition response sequence are as follows: Multimodal nutrition-related data were collected from the target elderly comorbid subjects over a continuous time period. The multimodal nutrition-related data included weight data, dietary intake data, laboratory indicator data, and activity monitoring data, and a corresponding collection timestamp was recorded for each data point. The continuous time range is divided into multiple continuous time slices according to a preset time interval, and each modal data is mapped to the corresponding time slice according to the timestamp; when there are multiple pieces of the same modal data in a certain time slice, the average value in that time slice is taken as the value of that modality in that time slice; For modal data that is missing in a certain time slice, the most recent data of that modality in the adjacent time slice is used to fill it; when there is no modal data in either of the adjacent time slices, the time slice is marked as a missing time slice and removed in subsequent processing; The modal data within each time slice are normalized according to their maximum and minimum values ​​within the continuous time range, so that the different modal data are converted to a unified numerical range. The normalized modal data within each time slice are arranged sequentially according to the time slice order to construct a multimodal nutrient response sequence.

[0020] It should be noted that, firstly, weight data, dietary intake data, laboratory indicator data, and activity monitoring data of the target elderly comorbid subjects are collected separately within a continuous time frame, and a corresponding collection timestamp is attached to each data point. Specifically, weight data refers to the weight value measured daily or periodically to reflect changes in body weight; dietary intake data refers to the records of energy, protein, and food types consumed in three meals or throughout the day to reflect intake status; laboratory indicator data specifically includes test results such as albumin, prealbumin, hemoglobin, electrolytes, or inflammation-related indicators to reflect the body's metabolic and nutritional status; activity monitoring data specifically includes data such as steps, activity duration, energy consumption, or bed rest time obtained through wearable devices to reflect the body's function and activity level. For example, weight data is collected once every morning, dietary intake data is recorded according to three meals a day, laboratory indicator data is tested every three days, and activity monitoring data is recorded hourly by wearable devices. Subsequently, the entire continuous time range is divided according to preset time intervals, such as dividing it into multiple continuous time slices by "day," and the above-mentioned data are assigned to the corresponding time slices based on their respective collection timestamps. If there are multiple records for a certain modality on a certain day, the average of these records is taken as the value of that modality on that day. For example, if an elderly person's activity monitoring device records the number of steps and activity intensity for multiple time periods on a certain day, the activity values ​​for each time period on that day can be averaged to obtain a unified value for the activity monitoring data on that day. For missing data within a certain time slice, the nearest data of the same modality from adjacent time slices is used to fill the gap. For example, if a person did not measure their weight on day 5, but had weight records on days 4 and 6, the nearest weight record in time is selected to fill the gap. If there is no data for that modality before or after a time slice, the time slice is marked as a missing time slice and removed to avoid data gaps that cannot be effectively judged in the subsequently constructed sequence. After time alignment, each modal data point is normalized according to its maximum and minimum values ​​over the entire continuous time range. This transforms data with different units and orders of magnitude into a unified numerical range. For example, weight might be expressed in kilograms, dietary intake in kilocalories, albumin in concentration, and activity monitoring in steps or activity duration. After normalization, all of these data can be converted into directly comparable standardized values. Finally, the normalized weight data, normalized dietary intake data, normalized laboratory indicator data, and normalized activity monitoring data from each time slice are arranged sequentially according to the time slice sequence, forming a multimodal nutritional response sequence corresponding to this elderly comorbidity subject.This processing method can transform multi-source nutrient-related data with different collection frequencies, scattered collection times, and inconsistent data units into a continuous, regular, and comparable sequence expression at the same time scale. This provides a unified data foundation for subsequent identification of response divergence segments between different modalities, construction of offset conflict structures, and calculation of offset contributions, avoiding distortion of subsequent evaluation results due to time misalignment, data loss, or differences in units.

[0021] In one embodiment, S2: Based on the multimodal nutrient response sequence, the step of identifying response divergence segments where the response directions of each modality are inconsistent within a continuous time range, and constructing an offset conflict structure reflecting the relationship between each modality within the response divergence segments, is as follows: According to the chronological order of the time slices, the numerical change of each mode in the multimodal nutrient response sequence between two adjacent time slices is calculated, and the response direction of each mode in each time slice is determined according to the sign of the numerical change. When the numerical change is greater than zero, the response direction of the mode in the corresponding time slice is recorded as increasing; when the numerical change is less than zero, the response direction of the mode in the corresponding time slice is recorded as decreasing; when the numerical change is equal to zero, the response direction of the mode in the corresponding time slice is recorded as remaining unchanged. Within each time slice, the response directions of each mode are compared pairwise. When the response directions of any two modes are different, and the response direction of at least one of the modes is not constant, it is determined that the two modes have a directional divergence within the time slice. The number of mode pairs with directional divergence within each time slice is counted, and time slices with a number of mode pairs greater than zero are marked as divergence time slices. Multiple consecutive divergent time slices are merged into candidate divergent segments, and the number of mode pairs corresponding to each time slice in each candidate divergent segment is counted. When the number of mode pairs corresponding to each time slice in a candidate divergent segment is greater than zero, the candidate divergent segment is determined as the response divergent segment. For each response divergence segment, extract all mode pairs with directional divergence corresponding to each time slice within the response divergence segment, and record the start time slice, end time slice, duration slice number, and numerical change difference within the response divergence segment for each mode pair. Each mode within the response divergence segment is used as a structural node, and the mode pairs with directional divergence are used as structural connection relationships. The start time slice, end time slice, duration slice number, and numerical change difference of each mode pair are written into the corresponding structural connection relationship to construct an offset conflict structure that reflects the relationship between each mode.

[0022] It's important to note that the approach first identifies the temporal direction of change for each modality, then determines whether these directions remain inconsistent over consecutive time periods, rather than directly synthesizing the multimodal data. Specifically, the changes in weight, dietary intake, laboratory indicators, and activity monitoring between adjacent time slices are calculated sequentially, and the response direction of each modality in the corresponding time slice is determined as increasing, decreasing, or remaining unchanged. Subsequently, the response directions of each modality are compared pairwise within each time slice. If any two modal directions differ and at least one is not unchanged, the modality pair is considered to have a directional divergence within that time slice, and the time slice with the directional divergence is marked as a divergence time slice. Consecutive divergence time slices are then merged into response divergence segments, and all modality pairs with directional divergence are extracted within each response divergence segment. The start time slice, end time slice, duration of each modality pair, and the difference in numerical change are recorded. Finally, using each modality as a structural node and the modality pairs with directional divergence as structural connections, an offset conflict structure is constructed. For example, if a subject experiences a continuous decline in weight and activity levels over three consecutive days, while their dietary intake consistently increases, a persistent discrepancy between "dietary intake-weight" and "dietary intake-activity monitoring" can be identified, and these three days can be designated as the response discrepancy segment. This approach allows for the precise extraction of when inconsistencies occur between multimodalities, their duration, and their degree of inconsistency. This avoids averaging out prematurely appearing anomalous signals during subsequent direct fusion, thus providing a more reliable foundation for identifying the dominant modality and reconstructing true nutritional trends.

[0023] In one embodiment, S3: Based on the offset conflict structure, the steps of calculating the offset contribution of each mode in the response divergence segment and identifying the dominant mode that dominates the trophic change based on the offset contribution are as follows: Based on the offset conflict structure, the conflict propagation dominance index and the stable structure destruction index of each mode in the response divergence section are calculated. The conflict propagation dominance index and the stable structure destruction index are added together to obtain the offset contribution index of each mode in the response divergence section. The mode with the largest offset contribution index is taken as the dominant mode that dominates the trophic change.

[0024] In one implementation, the calculation steps for the conflict propagation dominance index are as follows: For any mode to be calculated, extract all mode pairs directly connected to the mode to be calculated from the offset conflict structure, and read the start time slice and end time slice corresponding to each mode pair in the current response divergence segment. Determine the continuous time range between the start time slice and the end time slice corresponding to each mode pair as an associated conflict arc segment of the mode to be calculated, thereby obtaining the set of associated conflict arc segments of the mode to be calculated in the current response divergence segment. Combine any two different associated conflict arc segments in the set of associated conflict arc segments, calculate the time slice distance between the end time slice of the previous associated conflict arc segment and the start time slice of the next associated conflict arc segment, and add one to the time slice distance and take the reciprocal as the start and end bonding amount between the two associated conflict arc segments. The smaller the time slice distance between the two associated conflict arc segments, the larger the start and end bonding amount. Calculate the arc length of each associated conflict arc segment. The arc length of each associated conflict arc segment is obtained by subtracting the starting time slice number from the ending time slice number and then adding one. For any two different associated conflict arc segments, calculate the absolute value of the difference between their arc lengths and divide the absolute value by the sum of their arc lengths to obtain the length fold between the two associated conflict arc segments. For any two different associated conflict arc segments, multiply the beginning and end bonding amount between the two associated conflict arc segments by the length folding amount to obtain the bonding folding product value between the two associated conflict arc segments, and use the bonding folding product values ​​between all associated conflict arc segments to form the bonding folding product set of the mode to be calculated in the current response divergence segment. Sort the bonding and flipping product values ​​in the bonding and flipping product set according to their numerical values, and select multiple bonding and flipping product values ​​that can be connected end to end on the associated conflict arc segment number as a propagation chain. Then, multiply all the bonding and flipping product values ​​in the propagation chain continuously to obtain the strongest propagation chain value of the mode to be calculated. The time slices covered by all associated conflict arcs corresponding to the mode to be calculated within the current response divergence segment are merged and counted to obtain the total number of deduplicated covered time slices. The total number of covered time slices is then divided by the total number of time slices in the current response divergence segment to obtain the segment coverage ratio of the mode to be calculated. Multiply the strongest propagation chain value by the segment coverage ratio to obtain the propagation coverage product value. Then, add 1 to the propagation coverage product value as the numerator and use the propagation coverage product value as the denominator. Divide the denominator by the numerator to obtain the conflict propagation dominance index of the mode to be calculated in the current response divergence segment.

[0025] It's important to clarify that the conflict propagation dominance index doesn't simply indicate the magnitude of a mode's changes or the frequency of its involvement in conflicts. Instead, it measures whether a mode, within a response divergence segment, possesses the dominant ability to sequentially connect and expand multiple conflict relationships over time. More specifically, it measures whether, once a mode diverges directionally from other modes, this divergence is merely localized, sporadic, and short-lived, or whether it continuously connects around the mode to form a propagation chain spanning multiple time slices and involving multiple mode pairs, further covering a larger response divergence segment. The reason a higher conflict propagation dominance index indicates a greater likelihood that the mode is a key mode dominating nutrient changes is that the index calculation considers two levels of characteristics: firstly, whether the conflict arcs associated with the mode exhibit strong beginning-end cohesion and length reversal, meaning whether the conflict relationships formed around the mode can be sequentially connected and continuously expanded, rather than appearing isolated and brief; secondly, the extent to which these conflict relationships cover the entire response divergence segment, meaning whether this conflict expansion spans a relatively long divergence timeframe. Therefore, a large conflict propagation dominance index for a particular modality indicates that not only are there numerous conflicts surrounding that modality, but these conflicts are also continuous in time, structurally expanding from one modality pair to another, and occupying a large coverage area within the entire response divergence segment. This suggests that the modality is not merely passively involved in local conflicts, but rather acts as a core driving force causing persistent inconsistencies across multiple modalities. For example, in the assessment of an elderly individual with comorbidities, the dietary intake modality initially diverged from the weight modality, and subsequently from the activity monitoring modality. These divergence segments were adjacent, long-lasting, and collectively covered most of the response divergence segment, while the laboratory indicator modality only briefly diverged from the weight modality on one particular day. In this case, the conflict propagation dominance index for the former would be significantly larger. This indicates that the dietary intake modality has a stronger propagation dominance role in the formation and continuation of the overall multimodal inconsistency state, and is more likely to be the primary focus when identifying the dominant nutritional change modality.

[0026] S3: Based on the offset conflict structure, the steps for calculating the offset contribution of each mode in the response divergence segment and identifying the dominant mode that dominates trophic changes based on the offset contribution are as follows: Based on the offset conflict structure, the conflict propagation dominance index and the stable structure destruction index of each mode in the response divergence section are calculated. The conflict propagation dominance index and the stable structure destruction index are added together to obtain the offset contribution index of each mode in the response divergence section. The mode with the largest offset contribution index is taken as the dominant mode that dominates the trophic change.

[0027] In one implementation, the calculation steps for the stability structure failure index are as follows: For any mode to be calculated, read the starting time slice of the current response divergence segment, and search in reverse time from the beginning of the starting time slice for consecutive non-divergence time slices adjacent to the current response divergence segment. The consecutive non-divergence time slices are determined as the stable reference segments corresponding to the current response divergence segment. Within each time slice of the stable reference segment, the normalized values ​​of all modes in that time slice are sorted in ascending order. The position of the mode to be calculated in the sorting result is recorded, as well as the directly adjacent modes to the left and right of the mode to be calculated. Based on the position of the mode to be calculated in each time slice of the stable reference segment, the median is taken as the stable center position of the mode to be calculated, the minimum value is taken as the lower boundary of the stable position of the mode to be calculated, and the maximum value is taken as the upper boundary of the stable position of the mode to be calculated. At the same time, the occurrence frequency of the directly adjacent modes to the left and right of the mode to be calculated within the stable reference segment is counted. The directly adjacent mode to the left with the most occurrence frequency is determined as the stable left anchor mode, and the directly adjacent mode to the right with the most occurrence frequency is determined as the stable right anchor mode. Within each time slice of the current response divergence segment, the normalized values ​​of all modes in that time slice are sorted in ascending order, and the current position of the mode to be calculated in that time slice is recorded. When the current position is between the lower boundary and the upper boundary of the stable sequence, the escape depth of the corresponding time slice is recorded as zero. When the current position is less than the lower boundary of the stable sequence, the difference between the lower boundary of the stable sequence and the current position is divided by the total number of modes minus one to obtain the escape depth of the corresponding time slice. When the current position is greater than the upper boundary of the stable sequence, the difference between the current position and the upper boundary of the stable sequence is divided by the total number of modes minus one to obtain the escape depth of the corresponding time slice. Within each time slice of the current response divergence segment, extract the current left-side direct adjacent mode and the current right-side direct adjacent mode of the mode to be calculated, and compare them with the stable left anchor mode and the stable right anchor mode respectively. The side that is different from the corresponding stable anchor mode is marked as having undergone anchor replacement. Count the number of sides that have undergone anchor replacement in the time slice, and divide the number of sides by the total number of comparable sides in the time slice to obtain the anchor replacement degree corresponding to the time slice. Within each time slice of the current response divergence segment, determine whether the stable left anchor mode appears to the right of the current mode to be calculated, and whether the stable right anchor mode appears to the left of the current mode to be calculated; count the number of anchor points where the above left and right positions are swapped, and divide the number of anchor points by the total number of stable anchor points to obtain the anchor point mirror flip degree corresponding to the time slice; For each time slice in the current response divergence segment, first subtract the position escape depth, anchor replacement degree, and anchor mirror flip degree corresponding to the time slice from 1 respectively, then multiply the three subtraction results consecutively, and then subtract the product result from 1 to obtain the local structural damage coefficient corresponding to the time slice. Within the current response divergence zone, find the time slice in which the local structural failure coefficient of the mode to be calculated first exceeds zero, and determine this time slice as the failure initiation time slice of the mode to be calculated; at the same time, find the earliest time slice in all modes where the local structural failure coefficient exceeds zero, and determine this earliest time slice as the earliest failure initiation time slice; add one to the difference between the failure initiation time slice of the mode to be calculated and the earliest failure initiation time slice, and take the reciprocal to obtain the failure initiation leading factor of the mode to be calculated. Within the current response divergence segment, count the consecutive time segments in which the local structural failure coefficient corresponding to the mode to be calculated is continuously greater than zero, and take the longest consecutive time segment as the longest consecutive failure time segment; divide the number of time slices of the longest consecutive failure time segment by the total number of time slices in the current response divergence segment to obtain the continuous failure retention ratio of the mode to be calculated. For each time slice in the current response divergence segment, subtract the local structural damage coefficient corresponding to that time slice from 1, and multiply the subtraction results of all time slices consecutively; then normalize the result of the consecutive multiplication according to the total number of time slices in the current response divergence segment, and then subtract the normalized result from 1 to obtain the cumulative structural damage intensity of the mode to be calculated. The stability structural failure index of the mode to be calculated in the current response divergence segment is obtained by continuously multiplying the failure initiation leading factor, the successive failure retention ratio, and the cumulative structural failure intensity, and taking the square root of the multiplication result.

[0028] It's important to clarify that the Stability Structure Disruption Index (SSEI) is not used to simply indicate the magnitude of a mode's numerical change, nor to indicate the number of divergences a mode has made from other modes. Rather, it measures the degree to which a mode, upon entering the response divergence zone, disrupts the previously stable relative structure of the multimodal structures. The "stable structure" here refers to the relative positional and adjacency relationships formed by the modes within consecutive non-divergence time slices before the response divergence zone. For example, a mode typically occupies a middle position among several modes for a long period, with its left and right adjacent modes remaining relatively stable. "Structural disruption," on the other hand, refers to the mode beginning to deviate from its original stable position within the divergence zone, exceeding its previously stable positional range, or having its anchor modes on either side replaced, or even experiencing a reversal of left-right relationships. A higher SSEI indicates that the mode has not merely experienced ordinary fluctuations, but rather disrupted the previously stable relative structure earlier, more deeply, and more persistently. Therefore, it is more likely that this mode is the key mode causing changes in the overall assessment pattern. This is because the index calculation considers three levels of information simultaneously: First, whether the modality is the first to deviate from its original stable structure, i.e., whether it has the characteristic of "first destruction"; second, whether this destruction can last for a relatively long time, rather than just appearing briefly in a certain time slice; and third, whether this destruction penetrates multiple levels such as order escape, replacement of adjacent anchor points, and left-right mirror flipping, i.e., whether the destruction is thorough enough. For example, in an elderly person with comorbidities, the weight modality and laboratory modality basically maintain their original relative positions after the start of the divergence segment, with only slight fluctuations in values, while the activity monitoring modality is the first to jump out of its originally stable middle order, followed by the replacement of its left and right adjacent modalities, and this change continues for most of the entire divergence segment. In this case, the stable structure destruction index corresponding to the activity monitoring modality will be larger, which means that in this stage, it is more likely that the activity monitoring modality is the one that truly breaks the original stable assessment structure and promotes the rearrangement of multimodal relationships. Therefore, the larger the stability structure disruption index, the stronger, earlier, and more persistent the disturbance of the original stable state by the mode, and the more suitable it is as an important basis for identifying the dominant nutrient change mode.

[0029] In one embodiment, the conflict propagation dominance index and the stable structure disruption index are added together to obtain the offset contribution index of each mode in the response divergence segment, and the mode with the largest offset contribution index is taken as the dominant mode that dominates the nutrient change.

[0030] It should be noted that adding the conflict propagation dominance index and the stability structure disruption index to obtain the offset contribution index essentially merges two different roles of the same mode within the response divergence segment into a unified evaluation result. The conflict propagation dominance index reflects whether the mode can form a continuously expanding conflict relationship around itself, that is, whether the mode "brings up, connects, and continuously propagates" divergences among multiple modes. The stability structure disruption index reflects whether the mode has broken the originally stable relative position structure before the response divergence segment appeared, that is, whether the mode has changed the originally stable multimodal arrangement relationship earliest, deepest, and most persistently. Directly adding these two indices is equivalent to simultaneously considering "the mode's propagation dominance capability in the conflict network" and "the mode's ability to disrupt the original stable pattern," thus obtaining the comprehensive contribution of the mode to the deviation of the overall nutritional judgment from the normal stable state within the current response divergence segment. This comprehensive contribution is the offset contribution index. Since both sub-indices are defined as dimensionless quantities between zero and one, the summed offset contribution index, although its numerical range expands to between zero and two, can still be directly used for comparison between different modes. That is, it's not simply about how much a particular mode changes, but rather whether it has continuously driven the diffusion of multimodal divergence and whether it has truly disrupted the original stable structural position. Only when both aspects are strong simultaneously will the offset contribution index of that mode be larger. Subsequently, the offset contribution index of each mode within the response divergence segment is calculated, and these offset contribution indices are compared. The mode with the largest value is identified as the dominant mode driving trophic changes, because this mode has the strongest overall effect, indicating that it is more likely to be the center causing the continuous inconsistency of multimodal changes and also more likely to be the key factor that first changes the original stable pattern. For example, in the assessment of an elderly patient with comorbidities over several consecutive days, the dietary intake modality successively diverged from the weight and activity monitoring modalities, and these divergences almost covered the entire response divergence segment. Simultaneously, the dietary intake modality was the first to deviate from its original stable ordination position, causing replacement of its adjacent modalities. While the laboratory modality also showed divergence, the divergence time was short, the propagation range was small, and the disruption of the original relative position structure was not significant. Therefore, the conflict propagation dominance index and the stable structure disruption index of the dietary intake modality would be relatively large, and the resulting offset contribution index would be the largest. Thus, the dietary intake modality can be identified as the dominant modality of nutritional changes within this response divergence segment. This approach moves beyond simply using average fusion or observing the amplitude changes of a single modality to determine which is more important. Instead, it identifies the modality that truly plays a dominant role in nutritional changes from two different levels: "propagation divergence" and "disruption of stability." This makes the identification of the dominant modality more consistent with the actual evolution of nutritional status in elderly patients with comorbidities under complex multimodal conditions.

[0031] In one embodiment, S4: Based on the dominant mode, the multimodal nutritional response sequence is reconstructed to obtain a corrected nutritional evolution trend, and the nutritional assessment results of the target elderly comorbidity subject are output according to the corrected nutritional evolution trend. Extract the normalized values ​​corresponding to each time slice within the response divergence segment of the dominant mode, and arrange the normalized values ​​in the order of the time slices to form the dominant trend sequence; For the remaining modes in the multimodal nutrient response sequence, the difference between the normalized value corresponding to each time slice and the value corresponding to the dominant trend sequence in the same time slice is calculated to obtain the deviation value sequence of each remaining mode relative to the dominant trend sequence. For each other mode, in each time slice within the response divergence segment, if the absolute value of the deviation value corresponding to the other mode is less than a preset deviation threshold, the normalized value corresponding to the other mode in that time slice is retained; if the absolute value of the deviation value corresponding to the other mode is greater than or equal to the preset deviation threshold, the normalized value corresponding to the other mode in that time slice is replaced with the average of the normalized value of the other mode in that time slice and the value corresponding to the dominant trend sequence, to obtain the corrected sequence of each other mode; The values ​​of the dominant trend sequence and the modified sequences of the other modes are averaged at each time slice to obtain the reconstructed values ​​for each time slice. The reconstructed values ​​for each time slice are then arranged in chronological order to form the modified nutrient evolution trend. The modified nutrient evolution trend is compared with a preset evaluation threshold. When the modified nutrient evolution trend decreases over multiple consecutive time slices and the terminal value is lower than the preset evaluation threshold, an evaluation result of increased nutrient risk is output. When the modified nutrient evolution trend remains stable or increases over multiple consecutive time slices and the terminal value is higher than or equal to the preset evaluation threshold, an evaluation result of stable nutrient status is output.

[0032] It should be noted that, firstly, the normalized values ​​of the dominant mode in each time slice within the response divergence segment are extracted and arranged in chronological order to form a dominant trend sequence. This dominant trend sequence can be understood as the most trustworthy main line of change within the current divergence segment. Subsequently, for each of the remaining modes, the difference between its normalized value in each time slice and the corresponding value of the dominant trend sequence in the same time slice is calculated, thereby obtaining the deviation value sequence of the mode relative to the dominant trend sequence. The purpose of this step is to determine the extent to which the remaining modes deviate from the dominant change direction in each time slice. Next, within each time slice, the absolute value of the deviation corresponding to each of the other modes is compared with a preset deviation threshold. If the absolute value of the deviation is small, it indicates that although the mode is not the dominant mode, it is basically consistent with the dominant trend within that time slice, so its original normalized value is retained. If the absolute value of the deviation reaches or exceeds the preset deviation threshold, it indicates that the mode has significantly deviated from the dominant trend within that time slice, which may easily mask or interfere with subsequent overall judgments. In this case, its original value is no longer used directly, but is replaced by the average of the current normalized value of the mode and the value corresponding to the dominant trend, thereby pulling the excessively deviated mode back towards the dominant trend, and obtaining the corrected sequence of each of the other modes. Then, within each time slice, the values ​​corresponding to the dominant trend sequence and the corrected sequences of each of the other modes are averaged again to obtain the reconstructed value of that time slice. The reconstructed values ​​of all time slices are arranged in chronological order to finally form the corrected nutrient evolution trend. This corrected nutrient evolution trend is not a single result of a particular mode, nor is it a simple average of unprocessed multimodal results. Instead, it is an overall trend obtained by "using the dominant mode as an anchor point and correcting for modes with excessive deviations," thus better reflecting the true direction of nutrient changes within the response divergence segment. Finally, this corrected nutrient evolution trend is compared with a preset assessment threshold: if the trend continues to decline over multiple consecutive time slices, and the value corresponding to the last time slice is lower than the preset assessment threshold, an assessment result of increased nutrient risk is output; if the trend remains stable or gradually increases over multiple consecutive time slices, and the final value is higher than or equal to the preset assessment threshold, an assessment result of stable nutrient status is output.For example, in the assessment of an elderly individual with comorbidities, the activity monitoring modality was identified as the dominant modality. Its normalized values ​​over five consecutive time slices were 0.62, 0.58, 0.55, 0.49, and 0.44, showing a continuous downward trend. Meanwhile, the weight modality showed little change in the corresponding time slices, and the dietary intake modality even briefly increased in two time slices. Directly averaging these modalities might weaken the true downward trend. However, according to the processing method described in this step, the activity monitoring modality is first used to form a dominant trend sequence. Then, the deviation of other modalities from this dominant trend is compared. Time slices with excessive deviations are averaged and replaced to correct the trend. The final reconstructed overall nutritional evolution trend still shows a significant downward trend, and the endpoint is below the preset assessment threshold. Therefore, the system can output an assessment result indicating increased nutritional risk. This method avoids the problem of the dominant risk signal being diluted by other modalities due to direct averaging, ensuring that the final output retains multimodal information while highlighting the core trend that truly dominates current nutritional changes.

[0033] Based on the same inventive concept, this invention also provides an intelligent nutritional assessment system for comorbidities in the elderly. It includes: Processing module: Acquires multimodal nutrition-related data of the target elderly comorbidity subject within a continuous time range, performs time alignment and unified processing on the multimodal nutrition-related data, and constructs a multimodal nutrition response sequence; Response bifurcation module: Based on the multimodal nutrient response sequence, it identifies response bifurcation segments where the response directions of each modality are inconsistent within a continuous time range, and constructs an offset conflict structure reflecting the relationship between each modality within the response bifurcation segment; Dominant module: Based on the offset conflict structure, calculate the offset contribution of each mode in the response divergence segment, and identify the dominant mode that dominates the nutrient change based on the offset contribution; Assessment module: Based on the dominant mode, the trend of the multimodal nutrition response sequence is reconstructed to obtain the corrected nutrition evolution trend, and the nutrition assessment results of the target elderly comorbid subjects are output according to the corrected nutrition evolution trend.

[0034] The intelligent nutritional assessment system for elderly comorbidities provided in this invention constructs a multimodal nutritional response sequence and further identifies response divergence segments. Based on this, it establishes a shift conflict structure reflecting the relationships between different modalities, calculates the shift contribution of each modality in the response divergence segments, identifies the dominant modality, and finally reconstructs the trend of the multimodal nutritional response sequence based on the dominant modality. This transforms the existing assessment method of directly scoring or averaging multimodal data into an assessment method that first identifies persistent conflict relationships and response differences between multimodalities, and then distinguishes and corrects the shift influence of different modalities. This effectively avoids nutritional abnormality signals already appearing in highly sensitive modalities being misinterpreted by less sensitive modalities. This addresses the issue of dilution caused by the temporary stability of modal data. Furthermore, by structurally characterizing the differences in response timing and the sources of bias and introducing a dominant modality mechanism, it is possible to prioritize the retention of more indicative modal information about nutritional changes when multimodal data responses are asynchronous, reducing assessment distortion caused by edema, inflammation, recording bias, or comorbid interference. Moreover, by reconstructing trends to obtain corrected nutritional evolution trends, it is possible to more realistically reflect the nutritional changes in elderly individuals with comorbidities, thereby enabling early identification and accurate judgment of nutritional risks, avoiding misjudgments of apparent stability but actual deterioration, and ultimately improving the reliability of nutritional assessment results and providing a more suitable basis for subsequent intervention decisions.

[0035] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should still fall within the scope of the claims of the present invention.

Claims

1. A method for intelligent assessment of nutritional comorbidities in the elderly, characterized in that, Includes the following steps: S1: Obtain multimodal nutrition-related data of the target elderly comorbid subjects within a continuous time range, and perform time alignment and unified processing on the multimodal nutrition-related data to construct a multimodal nutrition response sequence; S2: Based on the multimodal nutrient response sequence, identify response divergence segments where the response directions of each modality are inconsistent within a continuous time range, and construct an offset conflict structure reflecting the relationship between each modality within the response divergence segment; S3: Based on the offset conflict structure, calculate the offset contribution of each mode in the response divergence segment, and identify the dominant mode that dominates the nutrient change based on the offset contribution; S4: Based on the dominant mode, the trend of the multimodal nutrition response sequence is reconstructed to obtain the corrected nutrition evolution trend, and the nutrition assessment results of the target elderly comorbid subjects are output according to the corrected nutrition evolution trend.

2. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 1, characterized in that, The steps for constructing multimodal nutrient response sequences are as follows: Multimodal nutrition-related data were collected from the target elderly patients with comorbidities over a continuous time period, and the corresponding collection timestamp was recorded for each data point. The continuous time range is divided into multiple continuous time slices according to a preset time interval, and each modal data is mapped to the corresponding time slice according to the timestamp; when there are multiple pieces of the same modal data in a certain time slice, the average value in the time slice is taken as the modal value in the time slice; For modal data that is missing in a certain time slice, the most recent data in the adjacent time slice is used to fill the missing data. When there is no modal data in either adjacent time slice, the time slice is marked as a missing time slice and removed in subsequent processing. The modal data within each time slice are normalized according to their maximum and minimum values ​​within the continuous time range, so that the different modal data are converted to a unified numerical range. The normalized modal data within each time slice are arranged sequentially according to the time slice order to construct a multimodal nutrient response sequence.

3. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 1, characterized in that, The steps for identifying response divergence segments where the modal response directions are inconsistent over a continuous time range, and constructing an offset conflict structure reflecting the relationship between the modes within these segments, are as follows: According to the chronological order of the time slices, the numerical change of each mode in the multimodal nutrient response sequence between two adjacent time slices is calculated, and the response direction of each mode in each time slice is determined according to the sign of the numerical change; when the numerical change is greater than zero, the response direction of the mode in the corresponding time slice is recorded as rising. When the change in value is less than zero, the response direction of the mode in the corresponding time slice is recorded as decreasing; When the change in value is zero, the response direction of the mode in the corresponding time slice is recorded as unchanged; Within each time slice, the response directions of each mode are compared pairwise. When the response directions of any two modes are different, and the response direction of at least one of the modes is not constant, it is determined that there is a directional divergence between the two modes within the time slice. The number of mode pairs with directional divergence within each time slice is counted, and time slices with a number of mode pairs greater than zero are marked as divergence time slices. Multiple consecutive divergent time slices are merged into candidate divergent segments, and the number of mode pairs corresponding to each time slice in each candidate divergent segment is counted. When the number of mode pairs corresponding to each time slice in a candidate divergent segment is greater than zero, the candidate divergent segment is determined as the response divergent segment. For each response divergence segment, extract all mode pairs with directional divergence corresponding to each time slice within the response divergence segment, and record the start time slice, end time slice, duration slice number, and numerical change difference within the response divergence segment for each mode pair. Each mode within the response divergence segment is used as a structural node, and the mode pairs with directional divergence are used as structural connection relationships. The start time slice, end time slice, duration slice number, and numerical change difference of each mode pair are written into the corresponding structural connection relationship to construct an offset conflict structure that reflects the relationship between each mode.

4. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 1, characterized in that, The steps for calculating the migration contribution of each mode in the response divergence region and identifying the dominant mode that dominates trophic changes based on the migration contribution are as follows: Based on the offset conflict structure, the conflict propagation dominance index and the stable structure destruction index of each mode in the response divergence section are calculated. The conflict propagation dominance index and the stable structure destruction index are added together to obtain the offset contribution index of each mode in the response divergence section. The mode with the largest offset contribution index is taken as the dominant mode that dominates the trophic change.

5. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 4, characterized in that, The calculation steps for the conflict propagation dominance index are as follows: For any mode to be calculated, extract all mode pairs directly connected to the mode to be calculated from the offset conflict structure, and read the start time slice and end time slice corresponding to each mode pair in the current response divergence segment. Determine the continuous time range between the start time slice and the end time slice corresponding to each mode pair as an associated conflict arc segment of the mode to be calculated, thereby obtaining the set of associated conflict arc segments of the mode to be calculated in the current response divergence segment. Combine any two different associated conflict arc segments in the set of associated conflict arc segments in pairs, calculate the time slice distance between the end time slice of the previous associated conflict arc segment and the start time slice of the next associated conflict arc segment, and add one to the time slice distance and take the reciprocal as the start and end bonding amount between the two associated conflict arc segments. The smaller the time slice distance between the two associated conflict arc segments, the larger the start and end bonding amount. Calculate the arc length of each associated conflict arc segment. The arc length of each associated conflict arc segment is obtained by subtracting the starting time slice number from the ending time slice number and then adding one. For any two different associated conflict arc segments, calculate the absolute value of the difference between their arc lengths and divide the absolute value by the sum of their arc lengths to obtain the length fold between the two associated conflict arc segments.

6. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 5, characterized in that, The calculation steps for the conflict propagation dominance index also include: For any two different associated conflict arc segments, multiply the beginning and end bonding amount between the two associated conflict arc segments by the length folding amount to obtain the bonding and folding product value between the two associated conflict arc segments, and use the bonding and folding product values ​​between all associated conflict arc segments to form the bonding and folding product set of the mode to be calculated in the current response divergence segment. Sort the bonding and flipping product values ​​in the bonding and flipping product set according to their numerical values, and select multiple bonding and flipping product values ​​that can be connected end to end on the associated conflict arc segment number as a propagation chain. Then, multiply all the bonding and flipping product values ​​in the propagation chain continuously to obtain the strongest propagation chain value of the mode to be calculated. The time slices covered by all associated conflict arcs corresponding to the mode to be calculated in the current response divergence segment are merged and counted to obtain the total number of deduplicated covered time slices. The total number of covered time slices is then divided by the total number of time slices in the current response divergence segment to obtain the segment coverage ratio of the mode to be calculated. Multiply the strongest propagation chain value by the segment coverage ratio to obtain the propagation coverage product value. Add 1 to the propagation coverage product value as the numerator and the propagation coverage product value as the denominator. Divide the denominator by the numerator to obtain the conflict propagation dominance index of the mode to be calculated in the current response divergence segment.

7. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 5, characterized in that, The steps for calculating the stability failure index are as follows: For any mode to be calculated, read the start time slice of the current response divergence segment, and search for the consecutive non-divergence time slices adjacent to the current response divergence segment in reverse time order from the start time slice. The consecutive non-divergence time slices are determined as the stable reference segments corresponding to the current response divergence segment. Within each time slice of the stable reference segment, the normalized values ​​of all modes in the time slice are sorted in ascending order. The position of the mode to be calculated in the sorting result is recorded, as well as the directly adjacent modes to the left and right of the mode to be calculated. Based on the position of the mode to be calculated in each time slice of the stable reference segment, the median is taken as the stable center position of the mode to be calculated, the minimum value is taken as the lower boundary of the stable position of the mode to be calculated, and the maximum value is taken as the upper boundary of the stable position of the mode to be calculated. At the same time, the occurrence frequency of the directly adjacent modes to the left and right of the mode to be calculated in the stable reference segment is counted respectively. The directly adjacent mode to the left with the most occurrence frequency is determined as the stable left anchor mode, and the directly adjacent mode to the right with the most occurrence frequency is determined as the stable right anchor mode. Within each time slice of the current response divergence segment, the normalized values ​​of all modes in the time slice are sorted in ascending order, and the current position of the mode to be calculated in the time slice is recorded. When the current position is between the lower boundary and the upper boundary of the stable sequence, the escape depth of the corresponding time slice is recorded as zero. When the current position is less than the lower boundary of the stable sequence, the difference between the lower boundary of the stable sequence and the current position is divided by the total number of modes minus one to obtain the escape depth of the corresponding time slice. When the current position is greater than the upper boundary of the stable sequence, the difference between the current position and the upper boundary of the stable sequence is divided by the total number of modes minus one to obtain the escape depth of the corresponding time slice.

8. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 7, characterized in that, The calculation steps for the stability failure index also include: Within each time slice of the current response divergence segment, extract the current left-side direct adjacent mode and the current right-side direct adjacent mode of the mode to be calculated, and compare them with the stable left anchor mode and the stable right anchor mode respectively. The side that is different from the corresponding stable anchor mode is marked as having undergone anchor replacement. Count the number of sides that have undergone anchor replacement within the time slice, and divide the number of sides by the total number of comparable sides within the time slice to obtain the anchor replacement degree corresponding to the time slice. Within each time slice of the current response divergence segment, determine whether the stable left anchor mode appears to the right of the current mode to be calculated, and whether the stable right anchor mode appears to the left of the current mode to be calculated; count the number of anchor points that have undergone the above left-right position swap, and divide the number of anchor points by the total number of stable anchor points to obtain the anchor point mirror flip degree corresponding to the time slice; For each time slice in the current response divergence segment, first subtract the position escape depth, anchor replacement degree, and anchor mirror flip degree corresponding to the time slice from 1 respectively, then multiply the three subtraction results consecutively, and then subtract the product result from 1 to obtain the local structural damage coefficient corresponding to the time slice. Within the current response divergence zone, find the time slice in which the local structural failure coefficient of the mode to be calculated first exceeds zero, and determine this time slice as the failure initiation time slice of the mode to be calculated. At the same time, find the earliest time slice in all modes where the local structural failure coefficient exceeds zero, and determine this earliest time slice as the earliest failure initiation time slice. Add one to the difference between the failure initiation time slice of the mode to be calculated and the earliest failure initiation time slice, and take the reciprocal to obtain the failure initiation leading factor of the mode to be calculated.

9. The intelligent nutritional assessment method for comorbidities in the elderly according to claim 8, characterized in that, The calculation steps for the stability failure index also include: Within the current response divergence segment, count the consecutive time segments in which the local structural failure coefficient corresponding to the mode to be calculated is continuously greater than zero, and take the longest consecutive time segment as the longest consecutive failure time segment; divide the number of time slices of the longest consecutive failure time segment by the total number of time slices in the current response divergence segment to obtain the continuous failure retention ratio of the mode to be calculated. For each time slice in the current response divergence segment, subtract the local structural damage coefficient corresponding to the time slice from 1, and multiply the subtraction results of all time slices consecutively; then normalize the result of the multiplication according to the total number of time slices in the current response divergence segment, and then subtract the normalized result from 1 to obtain the cumulative structural damage intensity of the mode to be calculated. The stability structural failure index of the mode to be calculated in the current response divergence segment is obtained by continuously multiplying the failure initiation leading factor, the successive failure retention ratio, and the cumulative structural failure intensity, and taking the square root of the multiplication result.

10. The intelligent nutritional assessment system for comorbidities in the elderly according to claim 1, characterized in that, The system includes: Processing module: Acquires multimodal nutrition-related data of the target elderly comorbidity subject within a continuous time range, performs time alignment and unified processing on the multimodal nutrition-related data, and constructs a multimodal nutrition response sequence; Response bifurcation module: Based on the multimodal nutrient response sequence, it identifies response bifurcation segments where the response directions of each modality are inconsistent within a continuous time range, and constructs an offset conflict structure reflecting the relationship between each modality within the response bifurcation segment; Dominant module: Based on the offset conflict structure, calculate the offset contribution of each mode in the response divergence segment, and identify the dominant mode that dominates the nutrient change based on the offset contribution; Assessment module: Based on the dominant mode, the trend of the multimodal nutrition response sequence is reconstructed to obtain the corrected nutrition evolution trend, and the nutrition assessment results of the target elderly comorbid subjects are output according to the corrected nutrition evolution trend.

Images