Early warning method for deterioration of icu patient based on multi-modal vital sign fusion

Abstract

The present application relates to the field of multi-modal vital sign fusion, and particularly relates to an ICU patient condition deterioration early warning method based on multi-modal vital sign fusion. The content includes: collecting multi-modal physiological parameters and performing local mode feature analysis; based on the local mode feature, dynamically strengthening the multi-modal physiological parameters to obtain enhanced multi-modal physiological parameters, calculating normalized physiological parameters, and further constructing a multi-modal local mode fusion index; based on the multi-modal local mode fusion index, introducing a double-layer risk strengthening mechanism of local anomaly index amplification and overall trend high-order power enhancement, calculating a condition risk index, identifying priority attention patients and risk rapidly rising patients, and outputting clinical intervention information. The problems of difficult identification of local weak anomalies and dense anomaly changes, difficult quantification of nonlinear coupling, difficult representation of short-time mutation of overall physiological state, and poor sensitivity of traditional methods to dynamic changes of patient condition deterioration and poor clinical interpretability are solved.

Description

Technical Field

[0001] This invention relates to the field of multimodal vital sign fusion, and more particularly to an early warning method for the deterioration of the condition of ICU patients based on multimodal vital sign fusion. Background Technology

[0002] In ICU settings, patients' conditions change rapidly and their physiological states fluctuate complexly. Clinically, continuous monitoring of vital signs is typically achieved using electrocardiogram (ECG) monitors, pulse oximeters, and respiratory monitoring equipment, combined with medical experience to assess the risk of their condition. Traditional ICU monitoring methods mainly include fixed-threshold alarm methods and multi-parameter linear scoring methods. Fixed-threshold alarm methods usually set upper and lower limits for individual indicators such as heart rate, pulse oximeter, and respiratory rate, triggering an alarm when the monitored value exceeds the threshold. Multi-parameter linear scoring methods, on the other hand, linearly weight or summarize multiple vital signs to form a comprehensive risk level.

[0003] However, the above-mentioned technologies still have significant limitations: on the one hand, the fixed threshold method can only identify vital signs that are already obviously abnormal, and it is difficult to identify changes in the condition that have not yet reached the alarm threshold but have shown continuous and weak abnormalities in advance, which can easily lead to missed detection of early deterioration of the condition; on the other hand, the multi-parameter linear scoring method is difficult to characterize the nonlinear coupling relationship between different physiological parameters, and cannot effectively reflect the risk changes brought about by the dense occurrence of local abnormalities and the short-term sudden changes in the overall state. Summary of the Invention

[0004] This invention provides an early warning method for the deterioration of ICU patients' conditions based on multimodal vital sign fusion, in order to solve the technical problems of traditional ICU condition monitoring technology, such as difficulty in early identification of local weak abnormalities and abnormally dense changes, difficulty in effectively quantifying the nonlinear coupling relationship between different vital signs, difficulty in continuously representing short-term changes in the overall physiological state, and insufficient sensitivity and poor clinical interpretability of traditional fixed threshold and linear scoring methods to dynamic changes in the deterioration of patients' conditions.

[0005] The present invention provides an early warning method for the deterioration of ICU patients based on multimodal vital sign fusion, which specifically includes the following steps:

[0006] S1. Collect multimodal physiological parameters and perform local pattern feature analysis to obtain local abnormal amplitude and local abnormal frequency; based on the local abnormal amplitude and local abnormal frequency, dynamically enhance the multimodal physiological parameters, calculate the enhanced multimodal physiological parameters, and generate a structured set of physiological features.

[0007] S2. Based on the enhanced multimodal physiological parameters, normalized physiological parameters are calculated, and multimodal local pattern fusion indices are obtained. Based on the multimodal local pattern fusion indices, a two-layer risk enhancement mechanism of amplifying local abnormality index and enhancing the overall trend with higher power is introduced to calculate the disease risk index. Based on the disease risk index, priority patients and patients with rapidly rising risk are identified, and structured clinical intervention information is output.

[0008] Preferably, real-time acquired multimodal physiological parameters are used as the calculation data source. A sliding window is introduced to calculate the local median and local standard deviation, and the magnitude of the local abnormality is obtained. Combined with the patient's historical multimodal physiological parameters, a historical baseline value is constructed. Based on the historical baseline value, the proportion of abnormal samples within the sliding window is calculated to obtain the local abnormality frequency.

[0009] Preferably, the local abnormal amplitude and local abnormal frequency are combined and a squared mapping is introduced to construct an abnormal amplification factor, which nonlinearly enhances the multimodal physiological parameters to obtain the enhanced multimodal physiological parameters.

[0010] Preferably, a one-to-one correspondence is established between the enhanced multimodal physiological parameters, the local median and local abnormal frequency of their respective sliding windows according to the same sampling time point and their respective sliding windows, forming a multimodal joint input space, and outputting a structured set of physiological features.

[0011] Preferably, the enhanced multimodal physiological parameters are normalized to obtain normalized physiological parameters, which are used to quantify the local abnormality amplitude of the enhanced multimodal physiological parameters; the normalized physiological parameters are then weighted and fused to obtain the multimodal local pattern fusion index.

[0012] Preferably, the specific implementation process of the dual-layer risk enhancement mechanism of local anomaly exponential amplification and overall trend high-order power enhancement includes:

[0013] Based on normalized physiological parameters, a local abnormality index mapping term is constructed. Combined with the local abnormality frequency and higher-order trend power mapping term, a disease risk index is calculated and generated.

[0014] Preferably, the higher-order trend power mapping term is constructed by performing a higher-order power mapping on the difference between the multimodal local pattern fusion index at the current time point and the previous trend observation time point.

[0015] Preferably, for all ICU monitored patients, the disease risk index and the change in disease risk index are calculated at the same monitoring time point; the disease risk index and the change in disease risk index are sorted in descending order to screen patients for priority attention and patients with rapidly rising risk, and structured clinical intervention information is output, including patient identification, current disease risk index, current risk ranking, change in disease risk index, ranking of changes, priority attention marker, and rapidly rising risk marker.

[0016] The beneficial effects of the technical solution of the present invention are:

[0017] 1. By coupling and mapping the amplitude and frequency of local abnormalities, an abnormality amplification factor is constructed, and multimodal physiological parameters are dynamically enhanced. This allows weak abnormalities to be amplified in advance before the disease shows obvious threshold exceedances, overcoming the problem of insufficient sensitivity of traditional fixed threshold monitoring to early disease deterioration and improving the ability to identify hidden and persistent abnormalities.

[0018] 2. By using exponential mapping and higher-order trend power mapping to simultaneously enhance the accumulation of local abnormalities and the abrupt changes in the overall state, a continuous quantitative description of the dynamic changes in the patient's condition can be achieved, so that the calculated disease risk index has dynamic sensitivity, continuity and clinical interpretability.

[0019] 3. By ranking the disease risk index and the change in the disease risk index, the system can continuously identify patients requiring priority attention and those with rapidly rising risk, and output structured clinical intervention information. This provides medical staff with risk ranking, trend analysis, and intervention priority criteria, which helps improve the efficiency of ICU patient monitoring and the accuracy of clinical decision-making. Attached Figure Description

[0020] Figure 1 This is a flowchart of the early warning method for the deterioration of ICU patients based on multimodal vital sign fusion, as described in this invention. Detailed Implementation

[0021] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] The following description, in conjunction with the accompanying drawings, details the specific scheme of the early warning method for the deterioration of ICU patients based on multimodal vital sign fusion provided by this invention.

[0024] See attached document Figure 1 The diagram illustrates a flowchart of an early warning method for deterioration of ICU patients based on multimodal vital sign fusion, according to an embodiment of the present invention. The method includes the following steps:

[0025] S1. Collect multimodal physiological parameters and perform local pattern feature analysis to obtain local abnormal amplitude and local abnormal frequency; based on the local abnormal amplitude and local abnormal frequency, dynamically enhance the multimodal physiological parameters, calculate the enhanced multimodal physiological parameters, and generate a structured set of physiological features.

[0026] First, strictly following ICU critical care clinical monitoring guidelines, multimodal physiological parameters were collected using routine ICU patient monitoring equipment, including electrocardiogram monitors, pulse oximeters, and respiratory rate sensors. The collected multimodal physiological parameters included six categories of physiological characteristics: heart rate, RR interval, pulse oximetry, pulse oximetry fluctuation amplitude, respiratory cycle, and respiratory depth. The data collection process required ensuring synchronized sampling cycles to guarantee that different types of multimodal physiological parameters were time-aligned. Simultaneously, historical multimodal physiological parameters and corresponding disease deterioration outcome data of patients are retrieved to construct a personal physiological baseline database. Historical multimodal physiological parameters are collected from routine ICU patient monitoring equipment and stored in the ICU monitoring database. Disease deterioration outcome data is structured event information from the electronic medical record system, ICU nursing record system, and medical order system, generated as binary labeled data after alignment by time windows. Specifically, using a specific monitoring time window as the unit (e.g., 0–24h, 24–48h after ICU admission), it is determined whether a clinical event related to disease deterioration occurred within each monitoring time window. If one of the following events occurs within the corresponding monitoring time window: escalation of mechanical ventilation, hemodynamic instability, use of vasopressors, cardiopulmonary resuscitation, transfer to higher-level life support, or a clearly recorded worsening event by the physician, the corresponding label within the monitoring time window is assigned a value of 1, indicating a disease deterioration event has occurred. If none of the above events occur within the corresponding time window, a value of 0 is assigned, indicating no disease deterioration event has occurred. This forms a binary sequence that changes over time to characterize the patient's disease deterioration status in different monitoring periods.

[0027] Next, using real-time acquired multimodal physiological parameters as the data source, a sliding window statistical method was employed to analyze the local pattern features of the multimodal physiological parameters, obtaining the local abnormal amplitude and local abnormal frequency. Specifically, based on the theory of short-term stationarity of physiological signals, combined with the sampling frequency and clinical signal fluctuation characteristics, such as the near-stationary characteristics of slow-varying characteristics like human heart rate, blood oxygen, and respiration within a 3-7 second time range, a sliding window duration was selected. For example, if the sampling frequency is 1 Hz and the sliding window duration is 5 seconds, then a single window contains 5 consecutive sampling time points. Within each sliding window, the local median and local standard deviation of the multimodal physiological parameters were calculated, and the local abnormal amplitude was calculated based on the local median and local standard deviation. Simultaneously, the number of abnormal samples deviating from the historical baseline was counted, and the proportion of abnormal samples within the sliding window was further calculated, which was used as the local abnormal frequency. ,in, For the sliding window index, the first Each sliding window consists of multiple consecutive sampling time points. This is an index of physiological characteristic categories for multimodal physiological parameters; the specific method for setting the historical baseline value is as follows: a sliding window statistical analysis is performed on the patient's historical multimodal physiological parameters to obtain the local mean and local standard deviation, and then the standard deviation method (mean ± standard deviation) is used. Double the standard deviation, Pick Determine the historical baseline value.

[0028] Furthermore, based on the theory of short-term stationarity of physiological signals and the idea of ​​local fluctuation analysis, the amplitude and frequency of local anomalies are explicitly combined to generate an anomaly amplification factor. This factor is then used to enhance the high-anomaly density window through squared mapping, thereby increasing the sensitivity to local anomalies and obtaining enhanced multimodal physiological parameters. This achieves the enhancement of weak anomalies and the suppression of stable baselines. The calculation formula is as follows: , in, For the first Physiological characteristics at time points Enhanced multimodal physiological parameters; For the first Physiological characteristics at time points Multimodal physiological parameters, such as heart rate, blood oxygen, or respiratory depth; Index for continuous sampling time points; Indicates a point in time The corresponding sliding window index; Indicates the first Local standard deviation of a sliding window; This is a local anomaly magnitude assessment item, used to reflect the magnitude at the current time point. The greater the deviation from the local trend, the more significant the abnormality in the current multimodal physiological parameters. For the anomalous frequency term, the effect of anomalous density is enhanced through squared mapping; As an aberration amplification factor, it enables the enhanced multimodal physiological parameters to simultaneously reflect the deviation amplitude and aberration frequency, improve the sensitivity to weak aberrations, and maintain extreme value control.

[0029] Furthermore, the enhanced multimodal physiological parameters The local median of its sliding window and local abnormal frequency A one-to-one correspondence is established between the same sampling time point and its corresponding sliding window. The enhanced signal of each sampling point is combined with the local statistical features of the sliding window to form a multimodal joint input space. Subsequently, based on the alarm limits of clinical monitoring equipment, commonly used physiological safety ranges in the ICU, and hospital parameter management standards, data that are missing or exceed physiological limits are filtered out, retaining valid structured samples, and outputting a structured set of physiological features. , .

[0030] S2. Based on the enhanced multimodal physiological parameters, normalized physiological parameters are calculated, and multimodal local pattern fusion indices are obtained. Based on the multimodal local pattern fusion indices, a two-layer risk enhancement mechanism of amplifying local abnormality index and enhancing the overall trend with higher power is introduced to calculate the disease risk index. Based on the disease risk index, priority patients and patients with rapidly rising risk are identified, and structured clinical intervention information is output.

[0031] After obtaining the structured set of physiological characteristics Subsequently, the enhanced multimodal physiological parameters were analyzed. Normalization is performed; the normalization calculation formula is as follows: Normalized physiological parameters were obtained. This is used to quantify the local abnormality amplitude of enhanced multimodal physiological parameters; based on normalized physiological parameters, a multimodal local pattern fusion index is constructed to characterize the changing trend of the patient's overall physiological state at the same time point, and the calculation formula is as follows: ,in, For time points Multimodal local pattern fusion index For the first The fusion weights for physiological features range from [value range missing]. The data was obtained by: using historical multimodal physiological parameters and corresponding disease deterioration outcomes of ICU patients, Pearson correlation coefficient analysis was used to statistically determine the correlation strength between each physiological characteristic and disease deterioration events, and then the maximum-minimum normalization method was used to map each correlation strength to... The interval is further subjected to summation and normalization to obtain the desired result. Multimodal physiological feature fusion weights are used to represent the relative contribution of different physiological features to disease risk; This represents the total number of categories of physiological characteristics.

[0032] After calculating the multimodal local pattern fusion index, based on the theory of short-term dynamic evolution of physiological signals and the coupling analysis of multimodal local anomalies, a two-layer risk enhancement mechanism is introduced, which combines local anomaly exponential amplification and overall trend power enhancement. This mechanism jointly maps the local anomaly frequency, the amplitude of local anomalies in the enhanced multimodal physiological parameters, and the overall physiological state trend changes to construct a disease risk index calculation formula. The specific implementation process is as follows: Exponential mapping enhances the contribution of weak anomalies in risk calculation, allowing early local anomalies to be continuously amplified and accumulated; simultaneously, power-law mapping of trend changes strengthens the response capability to sudden changes in the overall physiological state within a short period. This enables the disease risk index to reflect not only single abnormal events but also the aggregation and diffusion process of anomalies over time, thereby achieving sensitive quantification of early disease deterioration and forming a disease risk index that combines dynamic sensitivity and clinical interpretability. The disease risk index calculation formula is as follows: , in, For time points The disease risk index ranges from [0.5, 1). The higher the disease risk index, the higher the risk of the patient's condition worsening. The trend observation interval is used to measure the overall change in multimodal physiological parameters over a short period of time, mainly capturing local trend abrupt changes. Based on ICU clinical monitoring standards, the effective observation interval for short-term fluctuations in slowly varying physiological parameters such as heart rate, blood oxygenation, and respiration is 3 to 7 seconds. Therefore, as a specific implementation, this is set here... The value is 5 seconds; The abnormal frequency term is used to nonlinearly amplify the impact of local abnormal frequencies on the disease risk index, so as to ensure that the high abnormal density window contributes more to the disease risk index. This is a local anomaly index mapping term, used to amplify physiological characteristics that slightly deviate from the local median after normalization; Represented as a time point Multimodal local pattern fusion metrics; It is a higher-order trend power mapping term, used to enhance the impact of rapid changes in the overall physiological state on the disease risk index, so that rapid changes in the disease are amplified in the disease risk index. It's the Sigmoid function, which compresses the disease risk index to a specific value. Interval.

[0033] Finally, for all ICU patients under monitoring, the disease risk index and its change at the same monitoring time point were calculated for each patient. Subsequently, a current risk ranking was generated based on the disease risk index from high to low, and the top three patients were designated as priority patients. Simultaneously, a risk escalation rate ranking was generated based on the change in disease risk index from high to low, and the top three patients with positive changes were designated as rapidly escalating patients. The final output was structured clinical intervention information, including patient identification, current disease risk index, current risk ranking, change in disease risk index, ranking of changes, priority patient identification, and rapid escalation identification. This enabled continuous monitoring and intervention guidance of dynamic changes in the patient's condition, providing medical staff with structured and actionable risk ranking, trend analysis, and clinical decision support information. Among these, the change in disease risk index... ,in, The time interval selected in the calculation of the change in the disease risk index is used to measure the rate of change in disease risk over a future period of time. As a specific example, a value of 60 seconds can be used. Indicates a point in time The disease risk index.

[0034] In summary, a method for early warning of deterioration in ICU patients based on multimodal vital sign fusion has been developed.

[0035] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0036] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0037] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention 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 the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An early warning method for the deterioration of ICU patients based on multimodal vital sign fusion, characterized in that, Specifically, the following steps are included: S1. Collect multimodal physiological parameters and perform local pattern feature analysis to obtain local abnormal amplitude and local abnormal frequency; based on the local abnormal amplitude and local abnormal frequency, dynamically enhance the multimodal physiological parameters, calculate the enhanced multimodal physiological parameters, and generate a structured set of physiological features. S2. Based on the enhanced multimodal physiological parameters, normalized physiological parameters are calculated, and multimodal local pattern fusion index is obtained; based on the multimodal local pattern fusion index, a two-layer risk enhancement mechanism of local abnormality index amplification and overall trend high-order power enhancement is introduced to calculate the disease risk index. Based on the disease risk index, priority patients and patients with rapidly rising risk are identified, and structured clinical intervention information is output.

2. The method for early warning of deterioration in ICU patients based on multimodal vital sign fusion according to claim 1, characterized in that, Using real-time acquired multimodal physiological parameters as the data source, a sliding window is introduced to calculate the local median and local standard deviation, and obtain the magnitude of the local abnormality. Combined with the patient's historical multimodal physiological parameters, a historical baseline value is constructed. Based on the historical baseline value, the proportion of abnormal samples within the sliding window is calculated to obtain the frequency of the local abnormality.

3. The method for early warning of deterioration in ICU patients based on multimodal vital sign fusion according to claim 2, characterized in that, By combining the amplitude and frequency of local abnormalities and introducing a squared mapping, an abnormality amplification factor is constructed to nonlinearly enhance the multimodal physiological parameters, thereby obtaining the enhanced multimodal physiological parameters.

4. The method for early warning of deterioration in ICU patients based on multimodal vital sign fusion according to claim 3, characterized in that, The enhanced multimodal physiological parameters, the local median and local abnormal frequency of their respective sliding windows are established in a one-to-one correspondence according to the same sampling time point and the corresponding sliding window to form a multimodal joint input space, and a structured set of physiological features is output.

5. The method for early warning of deterioration in ICU patients based on multimodal vital sign fusion according to claim 4, characterized in that, The enhanced multimodal physiological parameters were normalized to obtain normalized physiological parameters, which were used to quantify the local abnormal amplitude of the enhanced multimodal physiological parameters. The normalized physiological parameters are weighted and fused to obtain the multimodal local pattern fusion index.

6. The method for early warning of deterioration in ICU patients based on multimodal vital sign fusion according to claim 5, characterized in that, The specific implementation process of the dual-layer risk enhancement mechanism, which involves exponential amplification of local anomalies and power-law enhancement of the overall trend, includes: Based on normalized physiological parameters, a local abnormality index mapping term is constructed. Combined with the local abnormality frequency and higher-order trend power mapping term, a disease risk index is calculated and generated.

7. The method for early warning of deterioration in ICU patients based on multimodal vital sign fusion according to claim 6, characterized in that, The higher-order trend power mapping term is constructed by performing a higher-order power mapping on the difference between the multimodal local pattern fusion index at the current time point and the previous trend observation time point.

8. The method for early warning of deterioration in ICU patients based on multimodal vital sign fusion according to claim 7, characterized in that, For all ICU patients under monitoring, the disease risk index and the change in the disease risk index were calculated at the same monitoring time point. The disease risk index and the change in the disease risk index were sorted in descending order to screen patients for priority attention and patients with rapidly rising risk. Structured clinical intervention information was output, including patient identification, current disease risk index, current risk ranking, change in disease risk index, ranking of changes, priority attention marker, and rapid rise marker.

Images