Intensive care monitoring method and system based on artificial intelligence

Abstract

The invention relates to an intensive care monitoring method and system based on artificial intelligence, and relates to the technical field of intensive care monitoring, and the method comprises the steps: obtaining the individual feature information and individual monitoring records of a target patient, constructing an individual monitoring baseline, and determining a stable monitoring state and a key monitoring state according to the individual monitoring baseline; collecting multi-dimensional real-time physiological monitoring data of a target patient, and judging whether the current monitoring state is a stable monitoring state or a key monitoring state in combination with the individualized monitoring baseline; based on the current monitoring state, the collection frequency of the multi-dimensional physiological parameters is determined, and the multi-dimensional physiological parameters of the patient are collected and monitored according to the frequency. The problems that in traditional intensive care monitoring, due to the fact that a general monitoring standard is adopted, individualized adaptation is lacked, and the monitoring frequency is fixed, key state response lags behind, monitoring precision is insufficient, and efficiency is low are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intensive care monitoring, in particular to an intensive care monitoring method and system based on artificial intelligence. BACKGROUND

[0002] With the development of intelligent intensive care, dynamic monitoring and accurate response of the physiological state of critical patients have become key technical problems to guarantee the treatment effect. At present, the traditional intensive care monitoring method mostly uses unified monitoring standards and fixed collection frequencies, which is difficult to adapt to individual differences and dynamic changes in the patient's condition, and is prone to insufficient monitoring accuracy, missing key state response, or resource redundancy, which not only reduces the utilization value of intensive care monitoring data, but also increases the work burden and monitoring cost of medical staff. SUMMARY

[0003] In order to solve the above technical problems, the present application provides an intensive care monitoring method and system based on artificial intelligence, which improves the insufficient monitoring accuracy, key state response lag and resource waste caused by lack of individual adaptation and fixed monitoring frequency in traditional intensive care monitoring.

[0004] The embodiments of the present application disclose the following technical solutions: In a first aspect, the embodiments of the present application provide an intensive care monitoring method based on artificial intelligence, which comprises: obtaining individual characteristic information and individual monitoring records of a target patient, constructing an individualized monitoring baseline, and determining a stable monitoring state and a key monitoring state of the target patient according to the individualized monitoring baseline; collecting multi-dimensional real-time physiological monitoring data of the target patient, determining a current monitoring state of the target patient according to the multi-dimensional real-time physiological monitoring data and the individualized monitoring baseline, the current monitoring state being a stable monitoring state or a key monitoring state; based on the current monitoring state, determining a multi-dimensional parameter collection frequency of multi-dimensional physiological parameters of the target patient, and collecting and monitoring the multi-dimensional physiological parameters of the target patient according to the multi-dimensional parameter collection frequency.

[0005] In a second aspect, the embodiments of the present application provide an intensive care monitoring system based on artificial intelligence, which comprises: an individualized monitoring baseline construction module, configured to obtain individual characteristic information and individual monitoring records of a target patient, construct an individualized monitoring baseline, and determine a stable monitoring state and a key monitoring state of the target patient according to the individualized monitoring baseline; A real-time monitoring state determining module is configured to collect multi-dimensional real-time physiological monitoring data of the target patient, and determine a current monitoring state of the target patient according to the multi-dimensional real-time physiological monitoring data and the individualized monitoring baseline, the current monitoring state being a stable monitoring state or a key monitoring state; A parameter collection frequency adapting module is configured to determine a multi-dimensional parameter collection frequency of the multi-dimensional physiological parameters of the target patient based on the current monitoring state, and collect and monitor the multi-dimensional physiological parameters of the target patient according to the multi-dimensional parameter collection frequency.

[0006] The one or more technical solutions provided in the present application have at least the following technical effects or advantages: The present application provides an intensive care monitoring method and system based on artificial intelligence, which realizes the precision and individualization of intensive care monitoring through the cooperative operation of constructing an individualized monitoring baseline, dividing monitoring states, and dynamically adjusting the collection frequency. First, the individual characteristic information and individual monitoring records of the target patient are obtained, and the individualized monitoring baseline is constructed by combining the data of the similar patient group; second, the similar reference patient group is retrieved based on the individualized monitoring baseline, and the stable state and abnormal state combination set are extracted to divide the monitoring state; then, the multi-dimensional real-time physiological monitoring data is collected, and the current monitoring state is determined by matching the monitoring state combination set; finally, according to the stable monitoring state or the key monitoring state, the mapping relationship model or the high-frequency collection strategy matrix is called respectively, and the collection frequency of each physiological parameter is dynamically adjusted to realize the targeted monitoring and monitoring of the target patient.

[0007] The technical solutions of the present application solve the problems of insufficient precision and resource waste caused by the generalization of monitoring standards and the fixation of collection frequency in traditional intensive care monitoring by fusing multiple steps such as individualized baseline construction, state precision division, and dynamic collection frequency adjustment, and realize the individualization and efficiency of monitoring, thereby providing technical support for the precise monitoring of critical patients. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0009] Figure 1 A flowchart of an intensive care monitoring method based on artificial intelligence provided by the embodiments of the present application is shown; Figure 2 A structure diagram of an intensive care monitoring system based on artificial intelligence provided by the embodiments of the present application is shown.

[0010] In the drawings, the components represented by the respective reference numerals are explained as follows: An individualized monitoring baseline construction module 01, a real-time monitoring state determination module 02, and a parameter collection frequency adaptation module 03. DETAILED DESCRIPTION

[0011] The present application provides an intensive care monitoring method and system based on artificial intelligence, which is used to solve the technical problems that the intensive care monitoring in the prior art relies on general criteria, lacks individualized adaptation, and has fixed monitoring frequency, resulting in insufficient accuracy of patient state judgment, delayed response to critical states, and resource waste and redundant monitoring.

[0012] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0013] In the description of the present application, the terms "first", "second" are used only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0014] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present application can be implemented without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope in accordance with the principles and characteristics disclosed in the present application.

[0015] Embodiment one, as shown in the accompanying Figure 1 The present application provides an intensive care monitoring method based on artificial intelligence, which comprises the following steps: S110: obtaining individual feature information and individual monitoring records of a target patient, constructing an individualized monitoring baseline, and determining a stable monitoring state and a key monitoring state of the target patient according to the individualized monitoring baseline; In the embodiment of the present application, in the scene of intensive care monitoring, in order to realize the precise monitoring of the target patient, the individualized monitoring baseline needs to be constructed and the monitoring state needs to be divided to provide the targeted monitoring basis.

[0016] Specifically, first, the disease type information of the target patient is collected, and the historical individual feature information set of the historical patients with the same disease is retrieved according to the disease type information of the patient. These historical individual feature information covers patient features in multiple dimensions such as different ages, disease courses, and basic health conditions.

[0017] At the same time, the historical individual feature information set is subjected to cluster analysis, and patients with similar features are classified into a category to obtain multiple patient feature categories.

[0018] Further, the individual feature information of the target patient is obtained, which is matched with the multiple patient feature categories to determine the matching patient feature category to which the target patient belongs.

[0019] At the same time, the historical physiological monitoring data of the same type of patients in the matching patient feature category is obtained, the statistical distribution range of each physiological parameter is extracted as the category reference monitoring range, and the individual numerical characteristics of each physiological parameter in the individual monitoring record of the target patient are combined to calculate the individual deviation coefficient.

[0020] Further, the individual deviation coefficient is used to individualize the category reference monitoring range to obtain the normal value interval of each physiological parameter of the target patient, and form the individualized monitoring baseline.

[0021] On this basis, based on the individualized monitoring baseline of the target patient, a reference patient group with similar individualized monitoring baseline is retrieved, and the historical physiological monitoring trajectory of the group is obtained.

[0022] Further, the physiological parameter state combination of the historical monitoring point marked with a stable identifier in the historical physiological monitoring trajectory is extracted to form a stable state combination set, and a stable monitoring state of the target patient is established.

[0023] Similarly, the physiological parameter state combination of the historical monitoring point marked with an abnormal identifier in the historical physiological monitoring trajectory is extracted to form an abnormal state combination set, and a key monitoring state of the target patient is established.

[0024] This step provides a basis for the subsequent determination of the real-time monitoring state of the target patient and the adjustment of the parameter acquisition frequency by constructing the individualized monitoring baseline and dividing the monitoring state, which guarantees the accuracy and pertinence of the intensive care monitoring.

[0025] The step S110 in the method provided by the embodiment of the present application comprises: collecting disease type information of the target patient, and obtaining a historical individual feature information set of a historical patient with a same disease type according to the disease type information; performing cluster analysis on the historical individual feature information set to obtain a plurality of patient feature categories; obtaining individual feature information of the target patient, matching the individual feature information of the target patient with the plurality of patient feature categories, and determining a matching patient feature category to which the target patient belongs; constructing an individual monitoring baseline for the target patient based on the matching patient feature category and the individual monitoring record of the target patient.

[0026] retrieving a reference patient population with a similar individual monitoring baseline based on the individual monitoring baseline of the target patient; obtaining a historical physiological monitoring trajectory of the reference patient population, the historical physiological monitoring trajectory having a plurality of historical monitoring points, each of the historical monitoring points having a stable identifier and an abnormal identifier; performing physiological parameter state combination extraction on the historical monitoring points marked as stable identifiers to obtain a stable state combination set; performing physiological parameter state combination extraction on the historical monitoring points marked as abnormal identifiers to obtain an abnormal state combination set; establishing a stable monitoring state of the target patient based on the stable state combination set, and establishing a key monitoring state of the target patient based on the abnormal state combination set.

[0027] In the embodiments of the present application, in order to construct an individual monitoring baseline suitable for the target patient and accurately divide the monitoring state, the monitoring data of the same type of patients and the individual features need to be hierarchically processed, and the individualization of the monitoring standard and the accuracy of the state determination are realized through multi-dimensional analysis.

[0028] Specifically, first, the disease type information corresponding to the target patient is collected from the clinical records of the target patient. For example, if the patient is diagnosed with acute myocardial infarction, the historical individual feature information set of all patients with acute myocardial infarction is retrieved from the historical intensive care monitoring database according to the diagnosis result.

[0029] The historical individual feature information set includes the age, gender, underlying diseases (such as whether combined with hypertension, diabetes), disease stage, treatment plan, and other multi-dimensional individual feature information of the patient, providing a comprehensive data basis for subsequent classification.

[0030] Further, the obtained historical individual feature information set is subjected to cluster analysis. That is, the hierarchical clustering algorithm in the prior art is used to calculate the Euclidean distance between different patient individual feature information, and the features with high similarity are classified into one category, and are gradually iteratively combined to form a plurality of patient feature categories.

[0031] Exemplarily, the acute myocardial infarction patients are divided into a 'young patient without underlying disease group', an 'elderly patient with hypertension group', a 'postoperative recovery period group', and the like.

[0032] Meanwhile, in the clustering process, by presetting a range of the number of patient characteristic categories (such as 3-8 categories), and combining the clinical diagnosis and treatment logic to filter the optimal classification result, that is, when the average similarity of the patient characteristics in the patient characteristic category is higher than 85% and the average similarity between categories is lower than 40%, it is determined that the clustering result is reasonable, and a plurality of stable patient characteristic categories can be obtained.

[0033] On this basis, the individual characteristic information of the target patient is obtained, including information such as age 65, combined with hypertension, and in the second day after operation, and the characteristic information is matched with the obtained plurality of patient characteristic categories.

[0034] Specifically, the characteristic similarity is calculated by the cosine similarity algorithm in the prior art. For example, if the similarity of the target patient to the 'elderly patient with hypertension group' reaches 0.85, which is higher than the similarity to other patient characteristic categories, it is determined that this group is the matching patient characteristic category to which the target patient belongs.

[0035] Further, after determining the matching patient characteristic category, the individual monitoring baseline is constructed in combination with the individual monitoring record of the target patient.

[0036] The method provided in the embodiments of the present application includes the following steps: Obtaining the historical physiological monitoring data of the same type of patients in the matching patient characteristic category, extracting the statistical distribution range of each physiological parameter as the category reference monitoring range; Extracting the individual numerical characteristics of each physiological parameter in the individual monitoring record of the target patient; Comparing and analyzing the individual numerical characteristics of the target patient with the category reference monitoring range, and calculating the individual deviation coefficient; According to the individual deviation coefficient, the category reference monitoring range is individualized and corrected to obtain the normal value interval of each physiological parameter of the target patient, and the individualized monitoring baseline is formed.

[0037] In the embodiments of the present application, in order to construct a monitoring baseline that fits the individual differences of the target patient, the hierarchical analysis of the same type of patient group monitoring data and individual monitoring data is required, and the universal category reference monitoring range is converted into exclusive monitoring standards in combination with the quantitative correction method, so as to improve the accuracy of intensive care monitoring.

[0038] Firstly, historical physiological monitoring data of patients with the same characteristics are retrieved from the matching patient characteristic category. For example, if the target patient belongs to the category of "elderly patients with hypertension", the historical physiological monitoring data of 12 core physiological parameters such as heart rate, systolic pressure, diastolic pressure, and oxygen saturation of 300 patients in this category are collected, covering a monitoring period of 1-3 months after the disease, with a total data point of not less than 5000 to ensure the representativeness and integrity of the sample.

[0039] On this basis, the statistical distribution range of each physiological parameter is extracted through statistical analysis.

[0040] Specifically, the statistical analysis is performed on the historical physiological monitoring data of 12 core physiological parameters of 300 patients in the "elderly patients with hypertension" category. That is, the 2.5% quantile and the 97.5% quantile of each physiological parameter are calculated to determine the statistical distribution range covering 95% of the patient data in the category as the category reference monitoring range.

[0041] For example, for the heart rate parameter, after statistical analysis of all monitoring data of the patients in this group within 1-3 months, the 2.5% quantile is 62 beats per minute and the 97.5% quantile is 88 beats per minute, and the category reference monitoring range of heart rate is 62-88 beats per minute (different from the commonly used 60-100 beats per minute in textbooks, which is more suitable for the characteristics of this group).

[0042] In addition, the systolic pressure parameter is calculated, with the 2.5% quantile being 130 mmHg and the 97.5% quantile being 155 mmHg, so the category reference monitoring range of systolic pressure is 130-155 mmHg; considering the possible characteristics of heart and lung function of this group, the 2.5% quantile of oxygen saturation is 92% and the 97.5% quantile is 97%, so the category reference monitoring range is 92-97%.

[0043] At the same time, the mean value (such as the average heart rate of this group of patients is 75 beats per minute, the average systolic pressure is 142 mmHg, and the average oxygen saturation is 94.5%) and the standard deviation (such as the standard deviation of heart rate is 6 beats per minute, the standard deviation of systolic pressure is 8 mmHg, and the standard deviation of oxygen saturation is 1.2%) of each physiological parameter are recorded to provide quantitative reference for subsequent comparative analysis with the individual characteristics of the target patient.

[0044] Further, the individual numerical characteristics of the target patient are extracted. The individual numerical characteristics include the mean value, variation range and variation trend pattern of each physiological parameter, which can accurately reflect the individual physiological state of the patient and provide the core basis for subsequent comparison with the category reference monitoring range.

[0045] Specifically, 12 physiological parameters corresponding to the category benchmark monitoring range are screened from the health records of the target patient 3 months before admission, preoperative examination records, and 24-hour postoperative monitoring data, and the mean value, variation range, and change trend mode of each physiological parameter are calculated.

[0046] For example, a certain target patient belongs to the "elderly patients with hypertension group", and the individual numerical characteristic extraction is as follows: the mean value of resting heart rate in the health records 3 months before admission is 68 times / min (forming the individual heart rate baseline), the variation range of the monitoring data within 24 hours after surgery is 65-72 times / min, and the change trend mode shows that the heart rate at night (22:00-6:00) is 8 times / min lower than the average during the day (reflecting the parameter change habit).

[0047] At the same time, the mean systolic pressure in the preoperative examination records is 135 mmHg (forming the individual blood pressure baseline), the variation range within 24 hours after surgery is 130-140 mmHg, and the change trend mode is that the systolic pressure in the morning (6:00-8:00) is 5-8 mmHg higher than that in other periods (consistent with its historical monitoring mode).

[0048] Secondly, the oxygen saturation parameter is 96% (forming the individualized oxygen saturation baseline considering the history of lung disease) in each preoperative examination record due to the target patient's history of mild chronic obstructive pulmonary disease (individual characteristic parameter), and the variation range is stable at 95%-97% without obvious time-dependent change trend.

[0049] Finally, these individual numerical characteristics obtained fully present the difference details of the target patient from the "elderly patients with hypertension group" category benchmark monitoring range, and fully reflect the individual normal value range, individual characteristic parameter, and historical monitoring mode, laying a data foundation for calculating the individual deviation coefficient.

[0050] Further, the extracted individual numerical characteristics of the target patient are compared and analyzed with the category benchmark monitoring range, and the individual deviation coefficient is calculated to quantify the difference degree of the target patient from the same group in the physiological parameters, providing accurate basis for the individualization of the category benchmark monitoring range in the future.

[0051] Specifically, for each physiological parameter, the individual deviation coefficient is calculated using the formula "individual deviation coefficient = (individual numerical characteristic mean value of the target patient - category benchmark monitoring range mean value) / category benchmark monitoring range standard deviation".

[0052] For example, also taking the "elderly patients with hypertension" as an example, if the average heart rate of the target patient is 68 beats per minute, the average heart rate category reference monitoring range is 75 beats per minute, and the standard deviation of the heart rate category reference monitoring range is 6 beats per minute, then the individual deviation coefficient of the heart rate is equal to (68-75) / 6≈-1.17.

[0053] Secondly, in terms of systolic pressure, the average systolic pressure of the target patient is 135 mmHg, the average systolic pressure category reference monitoring range is 142 mmHg, and the standard deviation of the systolic pressure category reference monitoring range is 8 mmHg. The individual deviation coefficient of the systolic pressure is equal to (135-142) / 8≈-0.88.

[0054] In addition, in terms of blood oxygen saturation, the average blood oxygen saturation of the target patient is 96%, the average blood oxygen saturation category reference monitoring range is 94.5%, and the standard deviation of the blood oxygen saturation category reference monitoring range is 1.2%. The individual deviation coefficient of the blood oxygen saturation is equal to (96-94.5) / 1.2≈1.25.

[0055] Further, for features such as variation range and change trend pattern, they are first converted into quantifiable mean features (such as the average value of the night heart rate drop, the average value of the morning systolic pressure rise, etc.), and then the calculation formula of the above individual deviation coefficient is calculated.

[0056] For example, also taking the "elderly patients with hypertension" as an example, for the variation range and change trend pattern of the heart rate, if the average night heart rate drop of the target patient is 8 beats per minute, the average night drop of the category reference monitoring range is 10 beats per minute, and the standard deviation of the heart rate category reference monitoring range is 2 beats per minute, then the individual deviation coefficient corresponding to the variation range and change trend pattern of the heart rate is (8-10) / 2=-1.

[0057] Secondly, for the variation range and change trend pattern of the systolic pressure, if the average morning systolic pressure rise of the target patient is 6.5 mmHg, the average morning rise of the category reference monitoring range is 4.5 mmHg, and the standard deviation of the systolic pressure category reference monitoring range is 2 beats per minute, then the individual deviation coefficient corresponding to the variation range and change trend pattern of the systolic pressure is (6.5-4.5) / 2=1.

[0058] In addition, for the variation range and change trend pattern of the blood oxygen saturation, if the average variation range of the target patient's blood oxygen saturation is 96%, the average variation range of the category reference monitoring range is 94.5%, and the standard deviation of the blood oxygen saturation category reference monitoring range is 1.5%, then the individual deviation coefficient corresponding to the variation range and change trend pattern of the blood oxygen saturation is (96-94.5) / 1.5=1.

[0059] On this basis, for heart rate, systolic blood pressure, blood oxygen saturation and other physiological parameters, the individual deviation coefficient calculated is used to adjust the category reference monitoring range.

[0060] Specifically, for the heart rate parameter, the category reference monitoring range is 62-88 times / min, the individual deviation coefficient of the target patient is -1.17, and when correcting, the upper and lower limits are adjusted by the product of the individual deviation coefficient and the standard deviation of the category reference monitoring range, i.e. the lower limit = 62 + (-1.17 x 6) ≈ 55 times / min, and the upper limit = 88 + (-1.17 x 6) ≈ 81 times / min. The normal value interval of the corrected heart rate is 55-81 times / min.

[0061] Similarly, for the systolic blood pressure parameter, the category reference monitoring range is 130-155 mmHg, and the individual deviation coefficient is -0.88. After correction, the lower limit = 130 + (-0.88 x 8) ≈ 123 mmHg, and the upper limit = 155 + (-0.88 x 8) ≈ 148 mmHg, forming a normal value interval of systolic blood pressure of 123-148 mmHg.

[0062] Similarly, for the blood oxygen saturation parameter, the category reference monitoring range is 92-97%, and the individual deviation coefficient is 1.25. After correction, the lower limit = 92 + (1.25 x 1.2) ≈ 93.5%, and the upper limit = 97 + (1.25 x 1.2) ≈ 98.5%, resulting in a normal value interval of blood oxygen saturation of 93.5-98.5%.

[0063] At the same time, the individual deviation coefficient combined with the variation range and trend mode is fine-tuned, i.e. the deviation coefficient of the heart rate change trend is -1, which further narrows the interval by 5%, resulting in a final corrected heart rate normal value interval of 52-77 times / min (55 x 95% ≈ 52, 81 x 95% ≈ 77).

[0064] At the same time, the individual deviation coefficient of the systolic blood pressure change trend is 1, and the interval is widened by 5%, resulting in a final positive systolic blood pressure normal value interval of 117-155 mmHg (123 x 95% ≈ 117, 148 x 105% ≈ 155).

[0065] In addition, the blood oxygen saturation related individual deviation coefficient is 1, and the interval is fine-tuned by 0.5% proportionally, resulting in a final blood oxygen saturation normal value interval of 93-99% (93.5 x 99% ≈ 93, 98.5 x 100.5% ≈ 99).

[0066] Further, the normal value intervals of the heart rate, systolic blood pressure, blood oxygen saturation and other physiological parameters after the individual deviation coefficient correction and variation trend fine-tuning are integrated to form an individualized monitoring baseline covering multiple core physiological indicators of the target patient.

[0067] Wherein, in the formed individualized monitoring baseline, the normal value interval of heart rate is 52-77 times / min, the normal value interval of systolic pressure is 117-155 mmHg, and the normal value interval of blood oxygen saturation is 93-99%, and the normal value interval of each physiological parameter is accurately adapted to the individual physiological characteristics and change law of the patient, thereby providing a dedicated and accurate reference standard for the subsequent judgment of the monitoring state.

[0068] Further, based on the obtained individualized monitoring baseline, a reference patient group with a similar individualized detection baseline is searched in a historical intensive care monitoring database of patients with the same disease.

[0069] Specifically, the normal value interval of each physiological parameter in the individualized monitoring baseline of the target patient is used as a search condition to screen a patient group with an individualized monitoring baseline similarity higher than 90% from the intensive care monitoring database as the reference patient group.

[0070] Exemplarily, for the individualized monitoring baseline of the target patient with a heart rate of 52-77 times / min, a systolic pressure of 117-155 mmHg, and a blood oxygen saturation of 93-99%, 200 patients with “old age combined with hypertension and postoperative recovery period” are matched in the intensive care monitoring database, and the coincidence degree of the individualized monitoring baseline interval of each physiological parameter of these patients with the target patient is more than 90%, such as the heart rate monitoring baseline of these patients is mostly 50-75 times / min, the systolic pressure monitoring baseline is 115-150 mmHg, and the blood oxygen saturation monitoring baseline is 92-98%, thereby forming the reference patient group.

[0071] Further, based on the reference patient group, a plurality of historical physiological monitoring trajectories corresponding thereto are obtained.

[0072] Wherein, these historical physiological monitoring trajectories contain multidimensional physiological parameter data of each patient at different intensive care monitoring periods, such as heart rate, blood pressure, blood oxygen, etc. recorded every 5 minutes, and the time span covers 1-3 months after the disease, so as to form a continuous monitoring data chain.

[0073] Meanwhile, each historical physiological monitoring point is marked with a stable mark or an abnormal mark by professional medical staff according to clinical judgment, for example, the monitoring point of a patient at 10:00 on the 5th day after surgery, each physiological parameter is within the individualized baseline range, and is marked with a “stable” mark; while at 16:00, the heart rate rises suddenly and the blood pressure drops, which is marked with an “abnormal” mark.

[0074] Further, the physiological parameter state combination extraction is performed on the historical monitoring points marked with the "stable" identifier. Through statistical analysis, it is found that more than 80% of the stable points in the reference patient population present a physiological parameter combination mode of "heart rate 55-70 beats / min + systolic blood pressure 120-140 mmHg + blood oxygen saturation 95-98%". These high-frequency appearing combinations are sorted to form a stable state combination set.

[0075] At the same time, the physiological parameter state combination extraction is performed on the historical monitoring points marked with the "abnormal" identifier. It is found that 70% of the abnormal points in the reference patient population present a combination mode of "heart rate >75 beats / min and systolic blood pressure <120 mmHg" "blood oxygen saturation <94% and respiratory rate >25 times / min", and the like. Accordingly, an abnormal state combination set is formed.

[0076] Finally, based on the stable state combination set, the stable monitoring state of the target patient is established. When the real-time monitoring data of the target patient matches any parameter combination in the combination set, it is determined to be in a stable monitoring state. Based on the abnormal state combination set, the key monitoring state is established. When the real-time data matches any abnormal combination, it is determined to be in a key monitoring state, which provides a basis for subsequent monitoring frequency adjustment.

[0077] S120: Collecting multi-dimensional real-time physiological monitoring data of the target patient, and determining a current monitoring state of the target patient according to the multi-dimensional real-time physiological monitoring data and the individualized monitoring baseline, the current monitoring state being a stable monitoring state or a key monitoring state; In the embodiments of the present application, in order to grasp the physiological state changes of the target patient in real time and accurately determine the current monitoring state thereof, multi-dimensional physiological monitoring data needs to be dynamically collected and compared with the individualized monitoring baseline and the preset state combination set, so as to accurately determine the monitoring state.

[0078] Specifically, first, the multi-dimensional physiological parameters of the target patient are collected in real time by the intensive care equipment, covering 12 core indicators such as heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, respiratory rate, and body temperature. The collection frequency is initially set to be once every 3 minutes according to the previous state of the patient, so as to ensure the timeliness and continuity of the data.

[0079] For example, the real-time physiological monitoring data of a certain target patient on the third day after surgery is: heart rate 65 beats / min, systolic blood pressure 130 mmHg, blood oxygen saturation 97%, and respiratory rate 20 times / min, and each physiological parameter data is accompanied by an accurate collection time stamp (such as May 20, 2024 08:30:00).

[0080] Further, the collected multi-dimensional real-time physiological monitoring data is preliminarily compared with the individualized monitoring baseline of the patient, and physiological parameters within the individualized monitoring baseline interval and physiological parameters exceeding the interval are screened out.

[0081] For example, taking the individualized monitoring baseline of the target patient (heart rate 52-77 beats / min, systolic pressure 117-155 mmHg, blood oxygen saturation 93-99%) as an example, if the real-time heart rate is 65 beats / min, which is within the interval of 52-77 beats / min, the systolic pressure is 130 mmHg, which is within the interval of 117-155 mmHg, and the blood oxygen saturation is 97%, which is within the interval of 93-99%, it is preliminarily determined that these physiological parameters meet the range of the individualized monitoring baseline; if the heart rate rises to 80 beats / min at a certain moment, which exceeds the upper limit of the individualized monitoring baseline of 77 beats / min, it is marked as a baseline deviating parameter.

[0082] On this basis, further determination is made in combination with the stable state combination set and the abnormal state combination set.

[0083] Specifically, the real-time physiological parameter combination is matched with the stable state combination set (such as "heart rate 55-70 beats / min + systolic pressure 120-140 mmHg + blood oxygen saturation 95-98%"), if the real-time physiological parameter combination (65 beats / min, 130 mmHg, 97%) completely falls within a certain stable state combination interval, and there is no baseline deviating parameter, it is determined that the current monitoring state is a stable monitoring state.

[0084] For example, if the real-time monitoring data of the target patient for three consecutive times all match the pattern in the stable state combination set, and the fluctuation amplitude of each parameter is less than 5%, it is confirmed that it is in a stable state.

[0085] On the contrary, if the real-time physiological parameter combination matches the abnormal state combination set (such as "heart rate >75 beats / min and systolic pressure <120 mmHg"), or there are two or more real-time physiological parameters deviating from the individualized monitoring baseline and the fluctuation amplitude exceeds 10%, it is determined that the current monitoring state is a key monitoring state.

[0086] For example, the real-time heart rate of the target patient is 82 beats / min (>75 beats / min), and the systolic pressure is 115 mmHg (<120 mmHg), which matches the abnormal state combination set, and the blood oxygen saturation drops to 92% (lower than the lower limit of the individualized monitoring baseline of 93%), which triggers the determination of the key monitoring state.

[0087] In addition, a dynamic determination threshold is set to cope with complex clinical scenarios. That is, when the real-time physiological parameters do not completely match the abnormal state combination, but there is a single physiological parameter that changes rapidly (such as an increase of 15 beats / min within 10 minutes), and the change trend is similar to the abnormal trajectory of the reference patient group, the determination sensitivity is automatically increased, and it is classified as a key monitoring state.

[0088] For example, the heart rate of a certain patient increases from 68 beats per minute to 80 beats per minute within 8 minutes, although it does not exceed the upper limit of the baseline too much, but the change rate meets the "pre-alarm before abnormality" in the reference patient population, so it is determined as the key monitoring state.

[0089] Through the multi-dimensional comparison and dynamic determination of the above steps, the stable monitoring state and the key monitoring state of the target patient can be accurately distinguished, providing a reliable basis for subsequent adjustment of the monitoring frequency, avoiding excessive monitoring of stable state patients, and ensuring close attention to key state patients.

[0090] S130: Based on the current monitoring state, determine the multi-dimensional parameter acquisition frequency of the multi-dimensional physiological parameters of the target patient, and acquire and monitor the multi-dimensional physiological parameters of the target patient according to the multi-dimensional parameter acquisition frequency.

[0091] In the embodiments of the present application, in order to realize dynamic and reasonable monitoring of the physiological parameters of the target patient, the corresponding multi-dimensional parameter acquisition frequency needs to be determined according to the current monitoring state, so as to ensure the accuracy of monitoring while avoiding redundant consumption of resources.

[0092] Specifically, when the current monitoring state of the target patient is the stable monitoring state, the mapping relationship model bound thereto will be automatically invoked to determine the acquisition frequency of each physiological parameter according to the multi-dimensional real-time physiological monitoring data of the target patient.

[0093] The mapping relationship model is constructed based on the historical monitoring data of the same disease patient population with similar individualized monitoring baseline in the stable monitoring state, and can output the adaptive acquisition frequency according to the subtle changes of the real-time physiological parameters.

[0094] Further, when the current monitoring state is the key monitoring state, the high-frequency acquisition strategy matrix is invoked to configure the acquisition frequency of each physiological parameter, so as to realize intensive monitoring of the physiological parameters of the target patient.

[0095] The high-frequency acquisition strategy matrix is calculated by analyzing the historical monitoring data of the same patient in the key monitoring state, combining the key degree and occurrence frequency of the physiological parameters. For physiological parameters with high key degree and easy fluctuation, such as heart rate and systolic pressure, higher acquisition frequency will be used to capture the trend of parameter changes in time.

[0096] On this basis, after determining the multi-dimensional physiological parameter acquisition frequency, the physiological parameters of the target patient will be continuously acquired and monitored according to the acquisition frequency, and the monitoring data will be stored in real time.

[0097] Meanwhile, the acquisition frequency is dynamically adjusted according to the change of the patient state during the monitoring process, and if the stable monitoring state lasts for more than 24 hours, the acquisition frequency can be appropriately reduced; if the physiological parameters tend to be stable in the intensive monitoring state, the acquisition frequency is gradually reduced to the acquisition frequency in the stable monitoring state, so as to realize intelligent and personalized intensive care monitoring.

[0098] The step S130 in the method provided in the embodiment of the application comprises: When the current monitoring state is the stable monitoring state, a mapping relationship model bound with the stable monitoring state is called, and the multi-dimensional parameter acquisition frequency of the target patient is determined according to the multi-dimensional real-time physiological monitoring data and the mapping relationship model; When the current monitoring state is the intensive monitoring state, a high-frequency acquisition strategy matrix bound with the intensive monitoring state is called, and the multi-dimensional parameter acquisition frequency of the target patient is determined according to the multi-dimensional real-time physiological monitoring data and the high-frequency acquisition strategy matrix.

[0099] In the embodiment of the application, in order to realize dynamic and accurate monitoring of physiological parameters of a critical patient, the acquisition frequency of each physiological parameter is determined in a differentiated manner according to the current monitoring state of the patient, which can ensure timely capture of key data and avoid unnecessary waste of resources, thereby improving the efficiency and accuracy of intensive care monitoring.

[0100] Specifically, when the current monitoring state of the target patient is the stable monitoring state, a mapping relationship model bound therewith is automatically called, and the adaptive parameter acquisition frequency is determined in combination with the multi-dimensional real-time physiological monitoring data.

[0101] In the method provided in the embodiment of the application, the construction step of the mapping relationship model comprises: Based on the individualized monitoring baseline of the target patient, a patient population with similar individualized monitoring baseline is retrieved; Historical monitoring data of the patient population with similar diseases in the stable monitoring state is acquired to construct a stable state sample set; The acquisition frequency of each physiological parameter in the stable state sample set is labeled to form an acquisition frequency label set; A machine learning algorithm is used to train the stable state sample set as input and the acquisition frequency label set as a supervised label to establish a mapping relationship model in the stable monitoring state.

[0102] In the embodiment of the application, in order to construct a mapping relationship model that can accurately output the acquisition frequency of physiological parameters in the stable monitoring state, data screening, labeling and model training are required in multiple steps to ensure that the model output is highly adapted to the actual clinical monitoring needs, and to provide a scientific basis for individualized monitoring in the stable monitoring state.

[0103] Specifically, first, based on the individualized monitoring baseline of the target patient, a patient population with similar baseline characteristics is retrieved from the intensive care monitoring database.

[0104] Illustratively, taking the "elderly patients with hypertension group" as an example, the individualized monitoring baseline of the target patient is heart rate 52-77 times / min, systolic pressure 117-155 mmHg, and blood oxygen saturation 93-99%, which is the retrieval condition. A patient population with a high degree of coincidence of more than 90% in the individualized monitoring baseline of each physiological parameter (such as 200 elderly patients with hypertension and in the stable stage) is screened out to form a patient population with similar diseases for model construction.

[0105] Further, the historical monitoring data of the patient population with similar diseases in the stable monitoring state is obtained, and a stable state sample set is constructed accordingly.

[0106] Among them, the historical monitoring data includes 12 core physiological parameters such as heart rate, systolic pressure, diastolic pressure, blood oxygen saturation, and respiratory rate, and the time span covers the continuous monitoring period of the patient in the stable state (such as the 3rd-10th day after surgery). The monitoring data collection time interval includes multiple types such as 1 minute, 5 minutes, and 10 minutes, and each monitoring data is attached with corresponding patient basic information (such as age, disease stage) and physiological state description at the monitoring time (such as "resting state" "after mild activity").

[0107] Illustratively, the historical monitoring data record of a certain patient in the stable monitoring state is: on the 5th day after surgery at 8:00, heart rate 65 times / min, systolic pressure 130 mmHg, blood oxygen saturation 97%, collection interval 5 minutes, and physiological state description "resting state".

[0108] On this basis, experienced professional doctors annotate the collection frequency of each physiological parameter in the stable state sample set to form a collection frequency annotation set.

[0109] Specifically, the collection frequency annotation process needs to combine clinical diagnosis and treatment specifications and patient individual characteristics, and follow the principles of "dynamic adaptation and precise efficiency". That is, for physiological parameters such as heart rate and systolic pressure that fluctuate relatively frequently, the resting state of the patient is annotated as collected every 10 minutes, and adjusted to every 8 minutes after mild activity.

[0110] Secondly, for relatively stable physiological parameters such as blood oxygen saturation and body temperature, the collection frequency is annotated as every 15-20 minutes; for physiological parameters such as respiratory rate that are less affected by external factors, the collection frequency is annotated as every 12 minutes.

[0111] Exemplarily, for the historical monitoring data of the stable state sample set "heart rate 62-70 times / min, systolic pressure 120-135 mmHg, resting state", the doctor will label the heart rate collection frequency as "10 minutes / time", the systolic pressure collection frequency as "10 minutes / time", and the blood oxygen saturation collection frequency as "15 minutes / time".

[0112] After the labeling is completed, consistency verification of the collection frequency labeling set is also required, that is, the same batch of historical monitoring data is independently labeled by 3 or more doctors, and when the consistency rate of the labeling results exceeds 90%, it is determined that the collection frequency labeling set is valid.

[0113] On this basis, a random forest algorithm is used as the basic framework, and a mapping relationship model in a stable monitoring state is established by using the stable state sample set as input and the collection frequency labeling set as supervised labels for model training.

[0114] Specifically, first, the physiological parameter data (such as heart rate value, systolic pressure value, and numerical coding of physiological state description) in the stable state sample set are taken as input features, and the frequency value (converted into a numerical form, such as "10 minutes / time" coded as 10) in the collection frequency labeling set is taken as an output label.

[0115] Further, the stable state sample set is divided into a training set and a test set in a ratio of 8:2, the training set is used to train the random forest model, a plurality of decision trees (such as 100) are constructed, each tree is independently trained based on randomly selected samples and features, and finally the average value of the prediction of the plurality of trees is taken as the model output.

[0116] At the same time, a cross-validation mechanism is introduced in the training process, and the prediction accuracy of the test set is taken as the evaluation index, when the accuracy rate reaches 95% and there is no obvious improvement for 5 consecutive iterations, the training is stopped, and the model is determined to be converged.

[0117] Finally, the obtained mapping relationship model can automatically output the optimal collection frequency of each physiological parameter according to the input real-time physiological parameter data, so as to realize the precise matching of the collection frequency in a stable monitoring state.

[0118] Exemplarily, when the target patient is in a stable monitoring state, the real-time physiological parameters are heart rate 68 times / min, systolic pressure 135 mmHg, blood oxygen saturation 96%, and the physiological state is "resting state", these data are input into the mapping relationship model, the model will match the similar sample features in the training set, and output the heart rate collection frequency as 10 minutes / time, the systolic pressure collection frequency as 10 minutes / time, and the blood oxygen saturation collection frequency as 15 minutes / time, which is completely consistent with the reasonable frequency determined by the clinician according to experience, verifying the effectiveness of the mapping relationship model.

[0119] Conversely, if the current monitoring state is the key monitoring state, the high-frequency acquisition strategy matrix bound with the key monitoring state is called, and the parameter acquisition frequency of the target patient is determined in combination with the multi-dimensional real-time physiological monitoring data.

[0120] The method provided in the embodiments of the application includes the following steps: Based on the individualized monitoring baseline of the target patient, a patient population with similar individualized monitoring baseline is retrieved; The historical monitoring data of the patient population with similar diseases in the key monitoring state is obtained to construct a key state sample set. The state criticality index and the state occurrence frequency index of the key state sample set are calculated, wherein the state criticality index is determined based on the physiological parameter change amplitude and the monitoring accuracy requirement, and the state occurrence frequency index is determined based on the occurrence proportion of the key monitoring state in the monitoring process of the similar patients. The dynamic acquisition density control parameter is obtained by weight fusion calculation based on the state criticality index and the state occurrence frequency index, the high-frequency acquisition frequency configuration of each physiological parameter is determined according to the dynamic acquisition density control parameter, and the high-frequency acquisition strategy matrix is formed.

[0121] In the embodiments of the application, in order to construct the high-frequency acquisition strategy matrix that can accurately adapt to the key monitoring state, targeted data retrieval, index calculation and weight fusion are required to ensure that the acquisition frequency of each physiological parameter in the matrix can accurately capture the condition change and avoid redundant consumption of resources.

[0122] Specifically, first, based on the individualized monitoring baseline of the target patient, a patient population with similar baseline characteristics is retrieved from the intensive care monitoring database.

[0123] Similarly, taking the target patient of the "elderly patients with hypertension" group as an example, the individualized monitoring baseline is heart rate 52-77 times / min, systolic pressure 117-155 mmHg, and blood oxygen saturation 93-99%. Similarly, the baseline physiological parameter interval coincidence degree is higher than 90% and the patient population has been in the key monitoring state, and the patient population is screened out to form a patient population with similar diseases for matrix construction.

[0124] Further, the historical monitoring data of the patient population with similar diseases in the key monitoring state is obtained to construct a key state sample set.

[0125] Among them, these historical monitoring data also cover 12 core physiological parameters such as heart rate, systolic blood pressure, and blood oxygen saturation, and the time span covers the entire monitoring period of the patient in the key monitoring state (such as 24-72 hours of postoperative condition fluctuation), and the monitoring data collection interval includes high-frequency types such as 1 minute, 2 minutes, and 5 minutes, and each data is attached with the corresponding condition change description (such as "heart rate rises sharply" "blood pressure drops") and clinical intervention record (such as "use of antihypertensive drugs" "adjust oxygen flow").

[0126] Exemplarily, the historical monitoring data record of a certain patient in the key monitoring state is: on the 2nd day after the operation at 14:00, the heart rate is 85 times / min, the systolic blood pressure is 110 mmHg, and the blood oxygen saturation is 92%, the collection interval is 2 minutes, and the postoperative condition description is "abnormal increase of heart rate".

[0127] On this basis, the state key degree index and the state occurrence frequency index of the key state sample set are calculated to quantify the importance and occurrence probability of each physiological parameter in the key monitoring state, and to provide accurate quantitative basis for the subsequent configuration of high-frequency collection frequency.

[0128] Firstly, for the state key degree index, the parameter change amplitude and the monitoring accuracy requirement are quantified.

[0129] Specifically, the larger the parameter change amplitude (such as an increase of 20 times / min in heart rate within 10 minutes), the higher the clinical intervention requirement (such as emergency treatment for sudden drop of systolic blood pressure), and the higher the corresponding state key degree index value (value range 0-1).

[0130] Exemplarily, the key degree index of heart rate is 0.9 (large change amplitude and close attention required), the key degree index of blood oxygen saturation is 0.8 (high accuracy requirement), and the key degree index of body temperature is 0.5 (relatively stable).

[0131] Secondly, for the state occurrence frequency index, the proportion of abnormal physiological parameters in the same type of patient in the key monitoring process is determined.

[0132] Specifically, the higher the proportion of physiological parameter abnormalities (such as 80% of patients in the key monitoring state have abnormal heart rate), the higher the corresponding state occurrence frequency index value (the value range is also 0-1).

[0133] Exemplarily, the occurrence frequency index of heart rate abnormality is 0.8, the occurrence frequency index of systolic blood pressure abnormality is 0.7, and the occurrence frequency index of respiratory rate abnormality is 0.4.

[0134] Then, the state key degree index and the state occurrence frequency index are fused to obtain the dynamic collection density control parameter.

[0135] Specifically, the weight fusion calculation formula can be expressed as "dynamic acquisition density control parameter = (state key degree index * 0.6 + state occurrence frequency index * 0.4)", by giving the state key degree index a higher weight, to ensure that the core physiological parameters are preferentially monitored at a high frequency.

[0136] Exemplarily, the dynamic acquisition density control parameter of heart rate = (0.9 * 0.6 + 0.8 * 0.4) = 0.86, the dynamic acquisition density control parameter of systolic pressure = (0.8 * 0.6 + 0.7 * 0.4) = 0.76, and the dynamic acquisition density control parameter of blood oxygen saturation = (0.8 * 0.6 + 0.6 * 0.4) = 0.72.

[0137] Finally, the high-frequency acquisition frequency configuration of each physiological parameter is determined according to the calculated dynamic acquisition density control parameter, to construct a high-frequency acquisition strategy matrix.

[0138] Wherein, the higher the dynamic acquisition density control parameter value, the higher the corresponding acquisition frequency. When the dynamic acquisition density control parameter is greater than or equal to 0.8, the acquisition frequency is once every 1 minute; when 0.6 is less than or equal to the dynamic acquisition density control parameter and is less than 0.8, the acquisition frequency is once every 2 minutes; when 0.4 is less than or equal to the dynamic acquisition density control parameter and is less than 0.6, the acquisition frequency is once every 3 minutes.

[0139] Exemplarily, the acquisition frequency corresponding to heart rate (the dynamic acquisition density control parameter is 0.86) is 1 minute / time, the acquisition frequency corresponding to systolic pressure (the dynamic acquisition density control parameter is 0.76) is 2 minutes / time, the acquisition frequency corresponding to blood oxygen saturation (the dynamic acquisition density control parameter is 0.72) is 2 minutes / time, and the acquisition frequency corresponding to body temperature (the dynamic acquisition density control parameter is 0.5) is 3 minutes / time. The high-frequency acquisition strategy matrix finally formed clearly presents the high-frequency acquisition scheme of each physiological parameter, and provides clear guidance for real-time monitoring data acquisition in the key monitoring state.

[0140] This step realizes the dynamic adaptation of physiological parameter acquisition frequency in different monitoring states by constructing the mapping relationship model and the high-frequency acquisition strategy matrix. In the stable monitoring state, the adaptive acquisition frequency is output according to the real-time monitoring data by means of the mapping relationship model, to balance the monitoring accuracy and efficiency; in the key monitoring state, the high-frequency acquisition monitoring is implemented on the key parameters by means of the high-frequency acquisition strategy matrix, to ensure that the condition changes can be captured in time, thereby forming a personalized and accurate acquisition scheme covering the whole monitoring period, and improving the scientificity and pertinence of intensive care monitoring.

[0141] The embodiments of the present application achieve the following technical effects through the specific implementation manners described above. This application proposes an artificial intelligence-based intensive care monitoring method. First, it acquires the target patient's disease type information and retrieves a set of individual characteristic information from historical patients with similar diseases for cluster analysis, resulting in multiple patient characteristic categories. The target patient's individual characteristic information is then matched with these categories to determine the corresponding matched patient characteristic category. Next, combined with historical physiological monitoring data from patients in the same matched category and the target patient's individual monitoring records, an individualized monitoring baseline is constructed. Based on this individualized monitoring baseline, a similar reference patient group is retrieved, and combinations of stable and abnormal physiological parameter states are extracted from their historical physiological monitoring trajectories to establish the target patient's stable and key monitoring states. Subsequently, multidimensional real-time physiological data of the target patient is collected, and the current monitoring state is determined by combining the individualized monitoring baseline and the state combination set. Finally, based on the stable or key monitoring state, a mapping relationship model or a high-frequency acquisition strategy matrix is ​​invoked to dynamically adjust the acquisition frequency of various physiological parameters and implement monitoring.

[0142] The method provided in this application, through the technical solution of "individualized baseline construction - monitoring status division - dynamic acquisition frequency adjustment", solves the problems of insufficient accuracy and waste of resources caused by the generalization of monitoring standards and the fixing of frequencies in traditional intensive care. It realizes personalized monitoring from individualized monitoring baseline construction to monitoring status determination and acquisition frequency adjustment, improves the accuracy and efficiency of intensive care, and provides reliable technical support for the scientific monitoring of critically ill patients.

[0143] Example 2, as shown in the appendix Figure 2 As shown, based on the inventive concept of the AI-based intensive care monitoring method provided in Embodiment 1, this application also provides an AI-based intensive care monitoring system, specifically including: The individualized monitoring baseline construction module 01 is used to acquire individual characteristic information and individual monitoring records of the target patient, construct an individualized monitoring baseline, and determine the stable monitoring status and key monitoring status of the target patient based on the individualized monitoring baseline; The real-time monitoring status determination module 02 is used to collect multidimensional real-time physiological monitoring data of the target patient and determine the current monitoring status of the target patient based on the multidimensional real-time physiological monitoring data and the individualized monitoring baseline. The current monitoring status is either a stable monitoring status or a key monitoring status. The parameter acquisition frequency adaptation module 03 is used to determine the multidimensional parameter acquisition frequency of the target patient's multidimensional physiological parameters based on the current monitoring status, and to collect and monitor the multidimensional physiological parameters of the target patient according to the multidimensional parameter acquisition frequency.

[0144] In one embodiment, the individualized monitoring baseline construction module 01 is further configured to: collecting disease type information of the target patient, and obtaining a historical individual feature information set of a historical patient with a same disease type according to the disease type information; performing cluster analysis on the historical individual feature information set to obtain a plurality of patient feature categories; obtaining individual feature information of the target patient, matching the individual feature information of the target patient with the plurality of patient feature categories, and determining a matching patient feature category to which the target patient belongs; constructing an individual monitoring baseline for the target patient based on the matching patient feature category and the individual monitoring record of the target patient.

[0145] retrieving a reference patient population with a similar individual monitoring baseline based on the individual monitoring baseline of the target patient; obtaining a historical physiological monitoring trajectory of the reference patient population, the historical physiological monitoring trajectory having a plurality of historical monitoring points, each of the historical monitoring points having a stable identifier and an abnormal identifier; performing physiological parameter state combination extraction on the historical monitoring points marked as stable identifiers to obtain a stable state combination set; performing physiological parameter state combination extraction on the historical monitoring points marked as abnormal identifiers to obtain an abnormal state combination set; establishing a stable monitoring state of the target patient based on the stable state combination set, and establishing a key monitoring state of the target patient based on the abnormal state combination set.

[0146] In one embodiment, the parameter collection frequency adaptation module 03 is further configured to: when the current monitoring state is a stable monitoring state, calling a mapping relationship model bound to the stable monitoring state, and determining the multi-dimensional parameter collection frequency of the target patient according to the multi-dimensional real-time physiological monitoring data and the mapping relationship model; when the current monitoring state is a key monitoring state, calling a high-frequency collection strategy matrix bound to the key monitoring state, and determining the multi-dimensional parameter collection frequency of the target patient according to the multi-dimensional real-time physiological monitoring data and the high-frequency collection strategy matrix.

[0147] It should be noted that the above sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. Moreover, the above describes a specific embodiment of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0148] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0149] The specification and drawings are only exemplary and illustrative of the present application and are to be considered within the scope of the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and equivalent technology, the present application is intended to include these modifications and variations.

Claims

1. An artificial intelligence-based intensive care monitoring method, characterized by, The method comprises: obtaining individual characteristic information and individual monitoring records of a target patient, constructing an individualized monitoring baseline, and determining a stable monitoring state and a key monitoring state of the target patient according to the individualized monitoring baseline; collecting multi-dimensional real-time physiological monitoring data of the target patient, determining a current monitoring state of the target patient according to the multi-dimensional real-time physiological monitoring data and the individualized monitoring baseline, and the current monitoring state being a stable monitoring state or a key monitoring state; based on the current monitoring state, determining a multi-dimensional parameter collection frequency of multi-dimensional physiological parameters of the target patient, and collecting and monitoring the multi-dimensional physiological parameters of the target patient according to the multi-dimensional parameter collection frequency.

2. The method of claim 1, wherein, Obtaining individual characteristic information and individual monitoring records of a target patient, constructing an individualized monitoring baseline, comprises: collecting disease type information of the target patient, and obtaining a set of historical individual characteristic information of patients with the same disease type according to the disease type information; performing cluster analysis on the set of historical individual characteristic information to obtain a plurality of patient characteristic categories; obtaining individual characteristic information of the target patient, matching the individual characteristic information of the target patient with the plurality of patient characteristic categories, and determining a matching patient characteristic category to which the target patient belongs; based on the matching patient characteristic category and the individual monitoring records of the target patient, constructing an individualized monitoring baseline for the target patient.

3. The method of claim 2, wherein, Based on the matching patient characteristic category and the individual monitoring records of the target patient, constructing an individualized monitoring baseline for the target patient, comprises: obtaining historical physiological monitoring data of patients with the same disease type in the matching patient characteristic category, and extracting statistical distribution ranges of each physiological parameter as a category reference monitoring range; extracting individual numerical characteristics of each physiological parameter in the individual monitoring records of the target patient; comparing and analyzing the individual numerical characteristics of the target patient with the category reference monitoring range, and calculating an individual deviation coefficient; according to the individual deviation coefficient, individualizing the category reference monitoring range to obtain a normal value interval of each physiological parameter of the target patient, and forming the individualized monitoring baseline.

4. The method of claim 3, wherein, According to the individualized monitoring baseline, determining a stable monitoring state and a key monitoring state of the target patient, comprises: based on the individualized monitoring baseline of the target patient, retrieving a reference patient population with a similar individualized monitoring baseline; obtaining a historical physiological monitoring trajectory of the reference patient population, the historical physiological monitoring trajectory having a plurality of historical monitoring points, each historical monitoring point having a stable identifier and an abnormal identifier; performing physiological parameter state combination extraction on the historical monitoring points marked with the stable identifier to obtain a stable state combination set; performing physiological parameter state combination extraction on the historical monitoring points marked with the abnormal identifier to obtain an abnormal state combination set; based on the stable state combination set, establishing a stable monitoring state of the target patient, and based on the abnormal state combination set, establishing a key monitoring state of the target patient.

5. The method of claim 1, wherein, Based on the current monitoring state, determining a multi-dimensional parameter collection frequency of multi-dimensional physiological parameters of the target patient, comprises: When the current monitoring state is a stable monitoring state, a mapping relationship model bound with the stable monitoring state is called, and a multi-dimensional parameter acquisition frequency of the target patient is determined according to the multi-dimensional real-time physiological monitoring data and the mapping relationship model; When the current monitoring state is a key monitoring state, a high-frequency acquisition strategy matrix bound with the key monitoring state is called, and the multi-dimensional parameter acquisition frequency of the target patient is determined according to the multi-dimensional real-time physiological monitoring data and the high-frequency acquisition strategy matrix.

6. The method of claim 5, wherein, The construction steps of the mapping relationship model include: Based on the individualized monitoring baseline of the target patient, a patient population with similar individualized monitoring baseline is retrieved; The historical monitoring data of the patient population with similar diseases in the stable monitoring state is obtained, and a stable state sample set is constructed; The acquisition frequency of each physiological parameter in the stable state sample set is labeled to form an acquisition frequency labeled set; A machine learning algorithm is used to train the stable state sample set as input and the acquisition frequency labeled set as a supervised label to establish a mapping relationship model in the stable monitoring state.

7. The method of claim 5, wherein, The construction steps of the high-frequency acquisition strategy matrix include: Based on the individualized monitoring baseline of the target patient, a patient population with similar individualized monitoring baseline is retrieved; The historical monitoring data of the patient population with similar diseases in the key monitoring state is obtained, and a key state sample set is constructed; The state criticality index and the state occurrence frequency index of the key state sample set are calculated, wherein the state criticality index is determined based on the physiological parameter change amplitude and the monitoring accuracy requirement, and the state occurrence frequency index is determined based on the occurrence proportion of the key monitoring state in the monitoring process of the similar patients; The dynamic acquisition density control parameters are obtained by weight fusion calculation based on the state criticality index and the state occurrence frequency index, and the high-frequency acquisition frequency configuration of each physiological parameter is determined according to the dynamic acquisition density control parameters to form a high-frequency acquisition strategy matrix.

8. Artificial intelligence based intensive care monitoring system characterized in that, The system is used to execute the artificial intelligence-based intensive care monitoring method of any one of claims 1-7, and the system includes: An individualized monitoring baseline construction module is configured to obtain individual characteristic information and individual monitoring records of a target patient, construct an individualized monitoring baseline, and determine a stable monitoring state and a key monitoring state of the target patient according to the individualized monitoring baseline; A real-time monitoring state determination module is configured to collect multi-dimensional real-time physiological monitoring data of the target patient, and determine a current monitoring state of the target patient according to the multi-dimensional real-time physiological monitoring data and the individualized monitoring baseline, wherein the current monitoring state is a stable monitoring state or a key monitoring state; A parameter acquisition frequency adaptation module is configured to determine a multi-dimensional parameter acquisition frequency of multi-dimensional physiological parameters of the target patient based on the current monitoring state, and collect and monitor the multi-dimensional physiological parameters of the target patient according to the multi-dimensional parameter acquisition frequency.

Images