Anaerobic metabolic threshold prediction system and anaerobic metabolic threshold prediction method

Abstract

To forecast an AT easily and accurately while eliminating subjectivity.SOLUTION: An anaerobic metabolism threshold forecasting system includes: a first AT forecasting model creation unit that determines at least one sixth criterion on other temporal data and non-temporal data for forecasting an AT by inputting a group of teacher data obtained by cardiopulmonary exercise testing to a first learning model and executing the first learning model, and creates a first AT forecasting model for creating a first AT forecasting model for forecasting an AT on the basis of first to sixth criteria, at least one sixth criterion being different from the first to fifth criteria; and a first AT forecasting unit for forecasting an AT of object data on the basis of first to sixth criteria of the object data by inputting object data including temporal data and non-temporal data of the same kinds as temporal data and non-temporal data included in the teacher data to the first AT forecasting model, and executing the first AT forecasting model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to systems and methods for predicting anaerobic threshold (AT) during cardiopulmonary exercise testing. [Background technology]

[0002] Heart disease is the leading cause of death worldwide, ranking second in Japan after malignant neoplasms. Furthermore, along with cerebrovascular disease, which ranks third, cardiovascular disease is a serious threat to the lives of the general public. However, according to the WHO, appropriate exercise, diet, and smoking cessation can prevent 80% of heart disease and type 2 diabetes, and 40% of cancer. Therefore, lifestyle modification is essential for preventing serious non-communicable diseases. Meanwhile, exercise therapy (especially cardiac rehabilitation) for ischemic heart disease, such as angina pectoris and myocardial infarction, and heart failure, the end-stage of all heart disease, is a treatment with strong evidence base. The Basic Act on Cardiovascular Disease Control, enacted in 2018, calls for the prompt and appropriate provision of high-quality cardiac rehabilitation from the acute phase through the recovery and normal life stages. Appropriate exercise therapy and cardiac rehabilitation are known to reduce mortality rates for myocardial infarction by more than 20% and rehospitalizations for heart failure by approximately 30%.

[0003] To ensure high-quality and appropriate cardiac rehabilitation, cardiopulmonary exercise stress tests are conducted on cardiac disease patients, and a safe and effective exercise prescription (e.g., type, intensity, and frequency of exercise) is issued. Exercising based on this prescription is necessary. This prevents complications caused by excessive exercise stress in cardiac disease patients and enables the safest possible therapeutic benefits. The most important index in issuing this exercise prescription is the anaerobic threshold (AT) derived from exercise stress tests. AT, along with maximal oxygen uptake (Peak VO2), is used clinically as a strong prognostic predictor for disease severity and to assess treatment effectiveness.

[0004] Gradually increasing the intensity of exercise (exercise intensity) during exercise increases the energy required for exercise (adenosine triphosphate (ATP)). Energy metabolism, which produces ATP, typically involves aerobic metabolism, which uses oxygen. However, when aerobic metabolism becomes insufficient, anaerobic metabolism, which does not use oxygen, supplements aerobic metabolism. The exercise intensity just before this anaerobic metabolism is introduced is the AT. Exercise below the AT is aerobic, while anaerobic exercise exceeding the AT results in lactic acid accumulation, blood acidification (acidosis), and the resulting changes in ventilation and pulmonary gas exchange. In essence, the AT is the exercise intensity or oxygen intake just before anaerobic metabolism is introduced into aerobic metabolism and the associated changes in gas exchange. It is a comprehensive exercise performance index that combines external respiration (ventilation and gas exchange), circulation, and internal respiration (metabolism). Because it is determined by oxygen transport capacity and oxygen utilization, it is an important index for determining exercise intensity in exercise therapy.

[0005] Below the AT, oxygen is used to efficiently produce energy (aerobic metabolism), and lactic acid does not accumulate, making it possible to exercise for long periods of time. On the other hand, above the AT, energy is produced without using oxygen (anaerobic metabolism), lactic acid accumulates, causing fatigue and making it impossible to continue exercising for long periods of time. In other words, AT is an indicator of endurance and daily activity level. Furthermore, AT can be used as an indicator of life prognosis for healthy individuals and patients with various diseases, specifically, to evaluate the severity of the disease, determine whether surgery is necessary, and evaluate the effectiveness of various treatments. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2010-046477 Summary of the Invention [Problem to be solved by the invention]

[0007] In actual exercise stress tests, the technician or doctor in charge of the test determines the AT based on subjective judgment, taking into account multiple criteria. Determining the AT subjectively requires knowledge of exercise physiology, exercise cardiology, respiratory and cardiac function, and experience with exercise stress tests. The AT may vary significantly depending on the analyst's subjectivity, such as differences in analysis results between experts and non-experts. Furthermore, fluctuations in the biological data itself may prevent accurate determination of the AT.

[0008] Due to the difficulty of accurately assessing AT, only a few medical facilities provide exercise prescriptions based on AT and provide cardiac rehabilitation in accordance with these exercise prescriptions. However, as a heart failure pandemic (a significant increase in heart failure patients) is expected in the future, cardiac rehabilitation will become increasingly important in the prevention and treatment of cardiovascular disease.

[0009] According to Patent Document 1, the operations required to obtain the desired respiratory metabolic data analysis result (AT) are displayed in an appropriate order, and an operation instruction describing the meaning of each operation or a specific operation method is displayed. The examiner looks at the instruction and performs the operations in order to obtain the desired respiratory metabolic data analysis result (AT) (Abstract). Specifically, the examiner presses the execute button 6, determines the point on the screen 3 that is thought to be the AT point, and sets the analysis interval by looking at the V-Slope curves before and after it (paragraph 0034). When the execute button 6 is pressed, the V-Slope curve (V CO2 and V O2 Two regression lines are automatically created on the graph (paragraph 0036 and Figure 3 of Patent Document 1). Looking at the V-Slope curve and the two regression lines, the examiner determines a point near the intersection of the two regression lines that they deem appropriate as the AT point, and sets the AT point by placing the cursor there (paragraph 0037 of Patent Document 1).

[0010] In Patent Document 1, by displaying the necessary operations in the appropriate order and providing operation instructions, it is thought that the examiner can be presented with an appropriate path to determining the AT point, but the final determination of AT depends on the subjective visual judgment of each examiner. Therefore, there is a risk that the AT may vary depending on the examiner, and as a result, there is a risk that an appropriate exercise prescription based on the AT may not be provided.

[0011] In view of the above circumstances, an object of the present disclosure is to eliminate subjectivity and predict AT simply and accurately. [Means for solving the problem]

[0012] The anaerobic metabolic threshold prediction system according to one embodiment of the present disclosure uses primary indices directly measured by an exhaled gas analyzer, namely, TV (tidal volume), RR (respiratory rate), and exhaled oxygen concentration (F EO2 ), exhaled carbon dioxide concentration (F ECO2 ), and simultaneously measured electrocardiogram waveforms (ST deviation, etc.), HR (heart rate), BP (blood pressure), arterial blood oxygen saturation (S PO2 ), and the following secondary indicators calculated from them: V O2 (minute oxygen intake) data over time, V CO2 (minute carbon dioxide emissions) over time, R (gas exchange ratio: V CO2 / V O2 ) longitudinal data and V E (minute ventilation) data over time, V E / V O2 and longitudinal data of V E / V CO2 and longitudinal data of F(P) ETCO2 (End-tidal carbon dioxide fractional concentration (partial pressure)) data over time, F(P) ETO2 (End-tidal oxygen fraction concentration (partial pressure)) over time data, other time-series data and non-time-series data different from each of the time-series data; The rate of increase of R is V O2 The first criterion is that the growth rate of V CO2 The rate of increase is V O2 The second criterion is that the growth rate of V E / V CO2 Without increasing V E / V O2 The third criterion is the point at which F(P) ETCO2 Without change, F(P) ETO2 The fourth criterion is the point at which V E The rate of increase is V O2 The fifth criterion is that the growth rate of The AT (anaerobic metabolic threshold) determined based on a first AT prediction model generation unit that inputs a group of teacher data obtained by a cardiopulmonary exercise stress test into a first learning model and executes the first learning model to determine at least one sixth criterion for predicting AT regarding the other time-course data and the non-time-course data, the at least one sixth criterion being different from the first to fifth criteria, and generates a first AT prediction model that predicts AT based on the first to sixth criteria; a first AT prediction unit that inputs target data including each of the time-series data and the non-time-series data of the same type as the time-series data and the non-time-series data included in the training data into the first AT prediction model, and executes the first AT prediction model to predict the AT of the target data based on the first to sixth criteria of the target data; It is equipped with:

[0013] According to this embodiment, the first AT prediction model predicts the AT of the target data by adding a sixth criterion, which is a new indicator (i.e., a new indicator that humans have not known) that is different from the first to fifth criteria that are typically used by examiners to determine AT based on their subjective judgment, thereby eliminating subjectivity and the presence or absence of experience and making it possible to predict AT simply and accurately.

[0014] a second AT prediction model generation unit that inputs the group of teacher data into a second learning model different from the first learning model, and executes the second learning model to determine at least one seventh criterion related to the other time-course data and the non-time-course data for predicting AT, the at least one seventh criterion being different from the first to fifth criteria, and generates a second AT prediction model different from the first AT prediction model that predicts AT based on the first to fifth and seventh criteria; a second AT prediction unit that inputs the target data into the second AT prediction model and executes the second AT prediction model to predict the AT of the target data based on the first to fifth and seventh criteria of the target data; may further comprise:

[0015] According to this embodiment, a seventh criterion is determined by a second learning model different from the first learning model, and a second AT prediction model is generated, thereby making it possible to predict an AT as an alternative to the AT predicted by the first AT prediction model. Furthermore, this AT is also predicted by the second AT prediction model based on the seventh criterion, which is a new index (i.e., a new index not previously known to humans) different from the first to fifth criteria typically used by examiners to determine AT based on their subjective judgment. Therefore, subjectivity and the presence or absence of experience are eliminated, and AT can be predicted simply and accurately.

[0016] an optimal solution output AI prediction unit that compares the AT predicted by the first AT prediction unit and the AT predicted by the second AT prediction unit from the common target data, determines whether the first AT prediction unit or the second AT prediction unit has high accuracy, and outputs the AT predicted by the first AT prediction unit or the second AT prediction unit that is determined to have high accuracy; may further comprise:

[0017] According to this embodiment, it is possible to output a more accurate AT from among the two types of AT that are two options.

[0018] An optimal solution output AI prediction unit that outputs both the AT predicted by the first AT prediction unit and the AT predicted by the second AT prediction unit from the common target data. may further comprise:

[0019] According to this embodiment, it is possible to provide the examiner with two options, that is, two types of AT.

[0020] The first learning model and the second learning model different from the first learning model may each be a logistic regression model, a Lasso regression model, a decision tree model, an XG boost model, a deep learning model, or a new model that will become available in the future, which analyzes numerical values ​​or graph images of each of the time series data.

[0021] According to this embodiment, the first AT prediction model and the second AT prediction model are created from different types of first and second learning models, respectively, so the first AT prediction model and the second AT prediction model are also different. Two types of ATs can be created from different first AT prediction models and second AT prediction models, respectively, and compared, making it easy to confirm the reliability of each of the two types of AT. For example, if the two types of ATs have the same or very similar values, both the first AT prediction model and the second AT prediction model are considered to be highly reliable. On the other hand, if the two types of ATs have dissimilar values, it is determined that the reliability of at least one of the first AT prediction model and the second AT prediction model is low. Therefore, a group of training data can be prepared again and the less reliable first AT prediction model or the second AT prediction model can be recreated.

[0022] An exercise prescription creation unit that creates an exercise prescription based on the predicted AT may further comprise:

[0023] According to this embodiment, based on AT that is predicted simply and accurately without subjectivity, an exercise prescription can be prepared simply and accurately without subjectivity.

[0024] The other time-course data is time-course data of TV (tidal volume), and the sixth criterion is that the rate of increase of TV is low, The other time-course data is time-course data of RR (respiratory rate), and the sixth criterion is that the rate of increase of RR is high; The other time-course data is time-course data of HR (heart rate), and the sixth criterion is that the rate of increase of HR is high. the other time-varying data is time-varying data of HR, and the sixth criterion is that high frequency components of heart rate variability disappear; The other time-course data is time-course data of BPs (systolic blood pressure), and the sixth criterion is that the rate of increase of BPs is high. The other time-course data is time-course data of DP (double product), which is the product of BPs and HR, and the sixth criterion is that the rate of increase of DP is high. the cardiopulmonary exercise stress test is performed using an ergometer, the other time-dependent data is time-dependent data of the rotation speed of the ergometer, and the sixth criterion is a temporary increase in the rotation speed; The sixth criterion is V E / V CO2 and / or The sixth criterion may be that the rate of increase of TV / RR is low. The sixth criterion may be derived from new learning models that become available in the future.

[0025] The other time-course data is time-course data of TV (tidal volume), and the seventh criterion is that the rate of increase of TV is low, The other time-course data is time-course data of RR (respiratory rate), and the seventh criterion is that the rate of increase of RR is high; The other time-course data is time-course data of HR (heart rate), and the seventh criterion is that the rate of increase of HR is high. The other time-varying data is time-varying data of HR, and the seventh criterion is that high frequency components of heart rate fluctuations disappear; The other time-course data is time-course data of BPs (systolic blood pressure), and the seventh criterion is that the rate of increase of BPs is high. The other time-course data is time-course data of DP (double product), which is the product of BP and HR, and the seventh criterion is that the rate of increase of DP is high. the cardiopulmonary exercise stress test is performed using an ergometer, the other time-dependent data is time-dependent data of the rotation speed of the ergometer, and the seventh criterion is a temporary increase in the rotation speed; The seventh criterion is V E / V CO2 and / or The seventh criterion may be that the rate of increase of TV / RR is low. The seventh criterion may be derived from new learning models that become available in the future.

[0026] The anaerobic metabolic threshold prediction system an nth AT prediction model generation unit that inputs the group of teacher data into an nth learning model different from the first learning model and the second learning model, and executes the nth learning model to determine at least one nth criterion related to the other time-course data and the non-time-course data for predicting AT, the at least one nth criterion being different from the first to fifth criteria, and generates an nth AT prediction model different from the first AT prediction model and the second AT prediction model that predicts AT based on the first to fifth and nth criteria; an nth AT prediction unit that inputs the target data into the nth AT prediction model and executes the nth AT prediction model to predict the AT of the target data based on the first to fifth and nth criteria of the target data; Further comprising: n may be an integer of 3 or greater.

[0027] A method for predicting an anaerobic metabolic threshold according to an embodiment of the present disclosure includes: V O2 (minute oxygen intake) data over time, V CO2 (minute carbon dioxide emissions) over time, R (gas exchange ratio: V CO2 / V O2 ) longitudinal data and V E (minute ventilation) data over time, V E / V O2 and longitudinal data of V E / V CO2 and longitudinal data of F(P) ETCO2 (End-tidal carbon dioxide fractional concentration (partial pressure)) data over time, F(P) ETO2 (End-tidal oxygen fraction concentration (partial pressure)) over time data, other time-series data and non-time-series data different from each of the time-series data; The rate of increase of R is V O2 The first criterion is that the growth rate of V CO2 The rate of increase is V O2 The second criterion is that the growth rate of V E / V CO2 Without increasing V E / V O2 The third criterion is the point at which F(P) ETCO2 Without change, F(P) ETO2 The fourth criterion is the point at which V E The rate of increase is V O2 The fifth criterion is that the growth rate of The AT (anaerobic metabolic threshold) determined based on inputting a group of training data obtained by a cardiopulmonary exercise stress test, including the above, into a first learning model, and executing the first learning model to determine at least one sixth criterion for predicting AT regarding the other time-course data and the non-time-course data, the at least one sixth criterion being different from the first to fifth criteria, and generating a first AT prediction model that predicts AT based on the first to sixth criteria; Target data including each of the time-series data and non-time-series data of the same type as the time-series data and non-time-series data included in the training data is input into the first AT prediction model, and the first AT prediction model is executed to predict the first AT of the target data based on the first to sixth criteria of the target data. [Effects of the Invention]

[0028] According to the present disclosure, it is possible to eliminate subjectivity and predict AT simply and accurately.

[0029] The effects described here are not necessarily limited to those described herein, and may be any of the effects described in this disclosure. [Brief explanation of the drawings]

[0030] [Figure 1] 1 shows the configuration of an anaerobic metabolic threshold prediction system according to an embodiment of the present disclosure. [Figure 2] An example of training data is shown below. [Figure 3] 1 shows the operational flow of the anaerobic metabolic threshold prediction method by the anaerobic metabolic threshold prediction system. DETAILED DESCRIPTION OF THE INVENTION

[0031] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0032] 1. Configuration of the anaerobic metabolic threshold prediction system

[0033] FIG. 1 shows the configuration of an anaerobic metabolic threshold prediction system according to an embodiment of the present disclosure.

[0034] The anaerobic metabolic threshold prediction system 100 includes a first AT prediction model generation unit 110, a second AT prediction model generation unit 120, a first AT prediction unit 130, a second AT prediction unit 140, an optimal solution output AI prediction unit 170, an AT output unit 150, and an exercise prescription creation unit 160.

[0035] Typically, the first AT prediction unit 130, the second AT prediction unit 140, the optimal solution output AI prediction unit 170, the AT output unit 150, and the exercise prescription creation unit 160 are realized by the cooperation of hardware and software resources in the same information processing device 10. The first AT prediction model generation unit 110 and the second AT prediction model generation unit 120 may be realized by the cooperation of hardware and software resources in an information processing device different from the information processing device 10, and may be capable of communicating with the information processing device 10 via a network. Alternatively, the first AT prediction model generation unit 110 and the second AT prediction model generation unit 120 may be realized by the information processing device 10. The information processing device refers to a computer having a control circuit in which a CPU loads an information processing program recorded in a ROM into a RAM and executes it.

[0036] The first AT prediction model generation unit 110 inputs a group of training data 200 into the first learning model 111 and executes the first learning model 111 to generate the first AT prediction model 131.

[0037] The second AT prediction model generation unit 120 inputs the group of teacher data 200 into the second learning model 121 and executes the second learning model 121 to generate the second AT prediction model 141.

[0038] The first learning model 111 and the second learning model 121 are different. For example, the first learning model 111 and the second learning model 121 may each be a logistic regression model, a Lasso regression model, a decision tree model, an XG boost model, a deep learning model, or a new model that will become available in the future, which analyzes numerical values ​​or graph images of each time series data. For example, the first learning model 111 is a machine learning model, and the second learning model 121 is a deep learning model.

[0039] The first learning model 111 may be, for example, a prediction from various data using XG boost, a prediction from time series data using XG boost, an analysis using machine learning or deep learning of image analysis, or an analysis using machine learning or deep learning from time series data analysis.

[0040] The second learning model 121 uses an algorithm of deep learning, for example, a neural network, etc. The second learning model 121 may be, for example, a model that learns an image (graph) or a table (numerical time-series data).

[0041] The first AT prediction unit 130 inputs the target data 300 into the first AT prediction model 131 generated by the first AT prediction model generation unit 110, and predicts the AT 132 by executing the first AT prediction model 131. The target data 300 is data used for determining the AT, and includes all the expiratory gas data output from the expiratory gas analyzer, as well as blood pressure, heart rate, and the like.

[0042] The second AT prediction unit 140 inputs the target data 300 into the second AT prediction model 141 generated by the second AT prediction model generation unit 120, and predicts the AT 142 by executing the second AT prediction model 141.

[0043] The optimal solution output AI prediction unit 170 inputs one or more programs for a model that outputs a plausible optimal solution from among several patterns of created AI models, and estimates one or more optimal ATs 142 from the ATs 132 predicted by the first AT prediction unit 130 and the ATs 142 predicted by the second AT prediction unit 140. The optimal solution output AI prediction unit 170 outputs the estimated one or more optimal ATs 142 to the AT output unit 150.

[0044] The AT output unit 150 outputs (e.g., displays on a display) the AT 132 predicted by the first AT prediction unit 130 and one or more ATs 142 predicted by the second AT prediction unit 140, which are estimated by the optimal solution output AI prediction unit 170.

[0045] The exercise prescription creation unit 160 creates and outputs an exercise prescription 161 based on both or either of the AT 132 predicted by the first AT prediction unit 130 and the AT 142 predicted by the second AT prediction unit 140, as well as the blood pressure, heart rate, double product (systolic blood pressure × heart rate), presence or absence of arrhythmia, electrocardiogram waveform, and oxygen saturation (target data 301) measured during exercise. The exercise prescription creation unit 160 may create an exercise prescription 161 based on the AT determined by a doctor based on the predicted AT, as well as electrocardiogram changes during exercise (particularly ST deviation), the presence and type of arrhythmia, blood pressure, oxygen saturation, etc.

[0046] 2. Training data

[0047] FIG. 2 shows an example of the training data.

[0048] A group of teacher data 200 includes multiple teacher data 200. One teacher data 200 will be explained below. One teacher data 200 is data obtained from a cardiopulmonary exercise stress test of one subject. The subject wears a respiratory measurement mask, a blood pressure monitor terminal, an electrocardiogram terminal, and a pulse rate monitor using an oxygen saturation meter, and exercises on an ergometer (cycle ergometer, treadmill ergometer, etc.). Using an exhaled gas analyzer, an automatic blood pressure monitor, and a stress electrocardiogram, the subject's exhaled gas is analyzed over time during exercise, and blood pressure, electrocardiogram, heart rate, arrhythmia, pulse rate, etc. are measured.

[0049] An example of a cardiopulmonary exercise stress test using a cycle ergometer is described below. During the rest period, the subject remains seated on the cycle ergometer without pedaling and maintains normal breathing. Next, during the warm-up period, the subject pedals at a light load of approximately 0-20 W at a pedal speed of 50-60 rpm. If no abnormal signs are observed in the electrocardiogram or blood pressure during the warm-up period, the subject enters the ramp period (linearly increasing load). During the ramp period, the pedal load is automatically gradually increased (the work rate is gradually increased), and the subject is instructed to pedal using a metronome or other device, while the rotation speed is monitored. During the ramp period, the subject continues pedaling until they feel they have reached their limit (peak). After the load is completed, during the recovery period, the subject remains seated on the cycle ergometer without pedaling and remains at rest. When using a treadmill as a resistance device, after standing and resting, the warm-up is performed at a speed of 2 to 4 km / h, and the belt speed and incline are gradually increased instead of increasing the work rate. The examiner (doctor, clinical laboratory technician, etc.; the same applies below) determines AT by referring to the time-series data obtained during exercise, such as the analysis values ​​of exhaled gases, blood pressure, electrocardiogram, and heart rate.

[0050] Figure 2 (A) shows the data used to determine AT using the Time Trend method, and Figure 2 (B) shows the data used to determine AT using the V slope method. The Time Trend method represents the criteria of exhaled gas on a time axis, and determines AT from the inflection points of multiple criteria. The V slope method uses the V CO2 The rate of increase is V O2 This is a method of determining AT from the point at which the rate of increase of AT becomes steeper than the rate of increase of AT. One training data 200 includes both data.

[0051] Teacher data 200 is V O2 (minute oxygen uptake) data over time 201 and V CO2 Time course data of (minute carbon dioxide output) 202 and R (gas exchange ratio: V CO2 / V O2 ) longitudinal data203 and V E(minute ventilation) data over time 204 and V E / V O2 Longitudinal data of 205 and V E / V CO2 The time course data of 206 and F(P) ETCO2 (End-tidal carbon dioxide fractional concentration (partial pressure)) data over time 207 and F(P) ETO2 These data are the primary indicators directly measured by the exhaled gas analyzer, namely, TV (tidal volume), RR (respiratory rate), and exhaled oxygen concentration (F EO2 ), exhaled carbon dioxide concentration (F ECO2 ), and simultaneously measured electrocardiogram waveforms (ST deviation, etc.), HR (heart rate), BP (blood pressure), arterial blood oxygen saturation (S PO2 ), and the following secondary indicators calculated from them:

[0052] As shown in Figure 2, a dot "·" representing "minutes ( / min)" is normally written above the "V" of each of the time-series data 201 to 206. In other words, a "V" with a dot "·" written on it represents an amount within a certain period of time. However, in this specification, the dot "·" is omitted.

[0053] The examiner determines whether the rate of increase in R is V O2 The first criterion is the point where the growth rate of V is higher than that of V (see Figure 2(A)). CO2 The rate of increase is V O2 The second criterion is the point at which the growth rate of V becomes higher than that of V (see Figure 2(B)). E / V CO2 Without increasing V E / V O2 The third criterion is the point where F(P) increases (see Figure 2(A)). ETCO2 Without change, F(P) ETO2 The fourth criterion is the point where V increases (see Figure 2(A)). E The rate of increase is V O2The AT 209 is determined by subjective judgment based on the fifth criterion (see FIG. 2(A)), which is the point at which the rate of increase of the AT becomes higher than the rate of increase of the AT 209. The training data 200 further includes the AT 209 determined based on the first to fifth criteria. The AT 209 is preferably data that is considered to be highly reliable, such as an AT determined by an examiner who has performed 10,000 or more cardiopulmonary exercise tests or by multiple examiners.

[0054] The teacher data 200 further includes other time-series data different from the above-mentioned time-series data, such as time-series data of TV (tidal volume), time-series data of RR (respiratory rate), time-series data 210 of HR (heart rate) (see FIG. 2(A)), time-series data of BPs (systolic blood pressure), time-series data of DP (double product) which is the product of BPs and HR, time-series data of the rotation speed and power of an ergometer, and time-series data of the speed and incline of a treadmill.

[0055] The teacher data 200 further includes non-chronological data, such as personal information such as age and gender, recent physical condition, medical history, medication history, and test result history, history of past exercise time, heart rate, electrocardiogram, heart rate trend graph, pulse rate, and transcutaneous oxygen saturation, medical checkup data, and data from various medical records.

[0056] In short, one piece of teacher data 200 includes each time-series data 201-208 obtained by a cardiopulmonary exercise stress test of one subject, an AT209 determined based on the first to fifth criteria based on each time-series data 201-208, other time-series data different from the above-mentioned each time-series data, and non-time-series data.

[0057] The group of training data 200 may be data generated by k-fold cross-validation. k-fold cross-validation is effective when the number of samples is small. According to k-fold cross-validation, the entire data is divided into n groups (k1, k2, k3, ... Kn), and the data of one group (for example, k1) is used as validation data, and the data of the remaining groups is used as training data. After that, the analysis is repeated a total of n times, for example, using k2 for validation, k3 for validation, and so on, and the results are integrated.

[0058] 3. Anaerobic metabolic threshold prediction method using the anaerobic metabolic threshold prediction system

[0059] FIG. 3 shows the operational flow of the anaerobic metabolic threshold prediction method by the anaerobic metabolic threshold prediction system.

[0060] The first AT prediction model generation unit 110 inputs a group of teacher data 200 into the first learning model 111 (step S101). The first AT prediction model generation unit 110 determines at least one sixth criterion by executing the first learning model 111 (step S102). "At least one sixth criterion" means one or more different criteria, and the number may be any number greater than or equal to one, and means that multiple different types of sixth criteria are used in combination.

[0061] The sixth criterion, unlike the first to fifth criteria, is a criterion for predicting AT based on the other time-course data and non-time-course data. In other words, the sixth criterion is a new index that is not typically used by examiners to determine AT based on subjective judgment. As described above, the other time-course data includes, for example, time-course data of TV, time-course data of RR, time-course data of HR, time-course data of BPs, time-course data of DP, and time-course data of rotation speed. For example, the sixth criterion includes a point where the rate of increase of TV decreases, a point where the rate of increase of RR increases, a point where the rate of increase of HR increases, a point where the high frequency components of heart rate fluctuations disappear, a point where the rate of increase of BPs increases, a point where the rate of increase of DP increases, a point where the rotation speed temporarily increases, a point where V E / V CO2 The sixth criterion may be determined as a variable sixth criterion that takes into account non-time-series data in addition to the other time-series data. The sixth criterion may be derived from a new learning model that becomes available in the future.

[0062] The first AT prediction model generation unit 110 generates a first AT prediction model 131 (step S103). The first AT prediction model 131 is a model that predicts the AT based at least on at least one determined sixth criterion. The first AT prediction model 131 may predict the AT based on the first to sixth criteria.

[0063] The first AT prediction unit 130 inputs the target data 300 to the first AT prediction model 131 (step S104).

[0064] The target data 300 is data obtained by a cardiopulmonary exercise stress test of one patient whose AT is to be determined. The target data 300 includes time-course data and non-time-course data of the same type as the time-course data and non-time-course data included in the teacher data 200. That is, the target data 300 includes V O2 Longitudinal data of 201 and V CO2 Time-lapse data 202 of R, time-lapse data 203 of V E Longitudinal data of 204 and V E / V O2 Longitudinal data of 205 and V E / V CO2 The time course data of 206 and F(P) ETCO2 The time course data of 207 and F(P) ETO2 The target data 300 includes the time-series data 208, other time-series data (for example, time-series data of TV, time-series data of RR, time-series data of HR, time-series data of BPs, time-series data of DP, time-series data of RPM, etc.), and non-time-series data and the like. However, the target data 300 does not include the AT 209 determined based on the first to fifth criteria.

[0065] The first AT prediction unit 130 executes the first AT prediction model 131 generated by the first AT prediction model generation unit 110 to predict the AT132 of the target data 300 based on at least the sixth criterion of the target data 300, typically based on the first to sixth criteria (step S105).

[0066] Meanwhile, the second AT prediction model generation unit 120 inputs a group of teacher data 200 to the second learning model 121 (step S111). The group of teacher data 200 input to the first AT prediction model generation unit 110 and the group of teacher data 200 input to the second AT prediction model generation unit 120 may be the same as or different from each other. The second AT prediction model generation unit 120 determines at least one seventh criterion by executing the second learning model 121 (step S112). "At least one seventh criterion" means one or more different criteria, and the number may be any number greater than or equal to one, and means that multiple different types of seventh criteria are used in combination.

[0067] The seventh criterion, unlike the first to fifth criteria, is a criterion for predicting AT based on the other time-course data and non-time-course data. In other words, the seventh criterion is a new index that is not typically used by examiners to determine AT based on subjective judgment. As described above, the other time-course data includes, for example, time-course data of TV, time-course data of RR, time-course data of HR, time-course data of BPs, time-course data of DP, and time-course data of rotation speed. For example, the seventh criterion includes a point where the rate of increase of TV decreases, a point where the rate of increase of RR increases, a point where the rate of increase of HR increases, a point where the high frequency components of heart rate fluctuations disappear, a point where the rate of increase of BPs increases, a point where the rate of increase of DP increases, a point where the rotation speed temporarily increases, a point where V E / V CO2 Possible points include a point where the rate of increase of (TV / RR) is low and / or a point where the rate of increase of (TV / RR) is low, but these are merely predictions based on the findings of the inventors and are not limited to these, and may not include these at all. The seventh criterion may be a variable seventh criterion that takes into account non-time-series data in addition to the other time-series data. The seventh criterion may be derived from a new learning model that will become available in the future. The seventh criterion and the sixth criterion are typically different, but may also be the same. Typically, a combination of multiple sixth criteria is different from a combination of multiple seventh criteria.

[0068] The second AT prediction model generation unit 120 generates a second AT prediction model 141 (step S113). The second AT prediction model 141 is a model that predicts the AT based on at least one determined seventh criterion. The second AT prediction model 141 may predict the AT based on the first to fifth and seventh criteria.

[0069] The second AT prediction unit 140 inputs the target data 300 to the second AT prediction model 141 (step S114). The target data 300 input to the second AT prediction model 141 is the same as the target data 300 input to the first AT prediction model 131. In other words, the target data 300, which is data obtained by a cardiopulmonary exercise stress test of one patient, is input to the first AT prediction model 131 (step S104) and then to the second AT prediction model 141 (step S114).

[0070] The second AT prediction unit 140 executes the second AT prediction model 141 generated by the second AT prediction model generation unit 120 to predict the AT142 of the target data 300 based on at least the seventh criterion of the target data 300, typically based on the first to fifth and seventh criteria (step S115).

[0071] The optimal solution output AI prediction unit 170 outputs (for example, displays on a display) both or one of the AT132 predicted by the first AT prediction unit 130 and the AT142 predicted by the second AT prediction unit 140 (step S106). For example, the optimal solution output AI prediction unit 170 may compare the AT132 predicted by the first AT prediction unit 130 and the AT142 predicted by the second AT prediction unit 140 from the common target data 300 to determine which of the first AT prediction unit 130 or the second AT prediction unit 140 has higher accuracy, and output the AT132 predicted by the first AT prediction unit 130 or the AT142 predicted by the second AT prediction unit 140 that is determined to have higher accuracy. Alternatively, the optimal solution output AI prediction unit 170 may simply output both the AT 132 predicted by the first AT prediction unit 130 and the AT 142 predicted by the second AT prediction unit 140, or may output them in order of increasing accuracy. When outputting them in order of increasing accuracy, the accuracy (probability of success) of each may be displayed. A doctor can make a diagnosis by referring to the AT 132 predicted by the first AT prediction unit 130 and / or the AT 142 predicted by the second AT prediction unit 140.

[0072] The exercise prescription creation unit 160 creates and outputs an exercise prescription 161 based on both or either of the AT 132 predicted by the first AT prediction unit 130 and the AT 142 predicted by the second AT prediction unit 140, as well as the blood pressure, heart rate, double product (systolic blood pressure × heart rate), presence or absence of arrhythmia, electrocardiogram waveform, and oxygen saturation (target data 301) measured during exercise (step S107). The exercise prescription is different for each user and includes various numerical values ​​that are targets when exercising using an ergometer. The exercise prescription includes target values ​​for the user's biological data (blood pressure, transcutaneous oxygen saturation, heart rate, pulse rate, etc.). The exercise prescription may further include target values ​​for the ergometer's power, torque (Nm: Newton meters), rotation speed, and exercise duration.

[0073] In this example, the first AT prediction model generation unit 110 determines at least one sixth criterion to generate a first AT prediction model 131, the first AT prediction unit 130 executes the first AT prediction model 131 to predict AT 132, and the second AT prediction model generation unit 120 determines at least one seventh criterion to generate a second AT prediction model 141, and the second AT prediction unit 140 executes the second AT prediction model 141 to predict AT 142. In addition, the nth AT prediction model generation unit may determine at least one nth criterion to generate an nth AT prediction model, and the nth AT prediction unit may execute the nth AT prediction model to predict AT (n is an integer of 3 or more).

[0074] In this case, the optimal solution output AI prediction unit 170 outputs (for example, displays on a display) all or one of the AT 132 predicted by the first AT prediction unit 130, the AT 142 predicted by the second AT prediction unit 140, and the AT predicted by the nth AT prediction unit (step S106). For example, the optimal solution output AI prediction unit 170 compares the AT 132 predicted by the first AT prediction unit 130, the AT 142 predicted by the second AT prediction unit 140, and the AT predicted by the nth AT prediction unit from the common target data 300 to determine which of the first AT prediction unit 130, the second AT prediction unit 140, or the nth AT prediction unit has higher accuracy, and may output the AT 132 predicted by the first AT prediction unit 130, the AT 142 predicted by the second AT prediction unit 140, or the AT predicted by the nth AT prediction unit that is determined to have higher accuracy. Alternatively, the optimal solution output AI prediction unit 170 may simply output all of the AT 132 predicted by the first AT prediction unit 130, the AT 142 predicted by the second AT prediction unit 140, and the AT predicted by the nth AT prediction unit, or may output them in order of increasing accuracy. When outputting them in order of increasing accuracy, the accuracy (probability of success) of each may be displayed. A doctor can make a diagnosis by referring to the AT 132 predicted by the first AT prediction unit 130, the AT 142 predicted by the second AT prediction unit 140, and / or the AT predicted by the nth AT prediction unit.

[0075] The exercise prescription creation unit 160 creates and outputs an exercise prescription 161 based on all or one of the AT 132 predicted by the first AT prediction unit 130, the AT 142 predicted by the second AT prediction unit 140, and the AT predicted by the nth AT prediction unit, as well as the blood pressure, heart rate, double product (systolic blood pressure x heart rate), presence or absence of arrhythmia, electrocardiogram waveform, and oxygen saturation (target data 301) measured during exercise (step S107).

[0076] When the exercise prescription creation unit 160 creates the exercise prescription 161, the following physiological characteristics of the AT (1) to (6) serve as guidelines for determining the intensity of exercise (exercise intensity). (1) The increase in heart rate up to the AT is mainly due to a decrease in parasympathetic nerve activity, resulting in low secretion of stress hormones. (2) Above the AT, sympathetic nerve activity (stress hormone secretion) increases, blood pressure rises, and arrhythmia becomes more likely. For this reason, exercise at the AT level (i.e., the level just before anaerobic exercise is added to aerobic exercise) is recommended for exercise therapy for hypertension and diabetes. (3) Above the AT, even healthy individuals do not increase their stroke volume. In heart disease, stroke volume decreases and the left ventricular ejection fraction decreases (deteriorating cardiac function). For this reason, exercise at the AT level is recommended for rehabilitation of heart disease and heart failure. (4) Above the AT, ventilation increases. For this reason, exercise at the AT level is recommended for exercise therapy for COPD (chronic obstructive pulmonary disease). (5) When the AT is exceeded, blood lactate levels increase. For this reason, exercise at the AT level is recommended for exercise therapy for patients with chronic kidney disease or hemodialysis. (6) Sugar is used as an energy source rather than fat. For this reason, exercise at the AT level is recommended for improving lipid metabolism.

[0077] 4. Conclusion

[0078] According to this embodiment, the first AT prediction model generation unit 110 determines a sixth criterion by executing the first learning model 111 (step S102) and generates a first AT prediction model 131 that predicts AT based at least on the sixth criterion (step S103). The first AT prediction unit 130 executes the first AT prediction model 131 to predict the AT 132 of the target data 300 based at least on the sixth criterion, typically based on the first to sixth criteria (step S105). According to this embodiment, the first AT prediction model 131 predicts the AT 132 of the target data 300 based on the sixth criterion, which is a new index (i.e., a new index not previously known to humans) that differs from the first to fifth criteria typically used by examiners to determine AT based on subjective judgment. This eliminates subjectivity and experience, allowing for simple and accurate prediction of AT 132.

[0079] According to this embodiment, the second AT prediction model generation unit 120 determines a seventh criterion by executing a second learning model 121 different from the first learning model 111 (step S112), and generates a second AT prediction model 141 that predicts AT based at least on the seventh criterion (step S113). The second AT prediction unit 140 predicts AT 142 of the target data 300 based at least on the seventh criterion, typically based on the first to fifth and seventh criteria, by executing the second AT prediction model 141 (step S115). According to this embodiment, by determining the seventh criterion using the second learning model 121 different from the first learning model 111 and generating the second AT prediction model 141, it is possible to predict AT 142 as an option different from AT 132 predicted by the first AT prediction model 131. Furthermore, this AT142 is also predicted by the second AT prediction model 141 based on the seventh criterion, which is a new indicator (i.e., a new indicator that humans have not been able to know) that is different from the first to fifth criteria that are typically used by examiners to determine AT based on their subjective judgment, thereby eliminating subjectivity and the presence or absence of experience, and making it possible to predict AT142 simply and accurately.

[0080] According to this embodiment, the optimal solution output AI prediction unit 170 may compare the AT132 predicted by the first AT prediction unit 130 and the AT142 predicted by the second AT prediction unit 140 from the common target data 300 to determine which of the first AT prediction unit 130 or the second AT prediction unit 140 has higher accuracy, and output the AT132 predicted by the first AT prediction unit 130 or the AT142 predicted by the second AT prediction unit 140 that is determined to have higher accuracy (step S106). According to this embodiment, of the two options, AT132 and AT142, the AT with higher accuracy can be output.

[0081] According to this embodiment, the optimal solution output AI prediction unit 170 may output both the AT132 predicted by the first AT prediction unit 130 and the AT142 predicted by the second AT prediction unit 140 (step S106). According to this embodiment, both the AT132 and AT142, which are two options, can be provided to the examiner.

[0082] According to this embodiment, the first learning model 111 and the second learning model 121 are each a logistic regression model, a Lasso regression model, a decision tree model, an XG boost model, or a deep learning model that analyzes the numerical values ​​or graph images of each time series data. According to this embodiment, the first AT prediction model 131 and the second AT prediction model 141 are created from the different types of the first learning model 111 and the second learning model 121, respectively, so that the first AT prediction model 131 and the second AT prediction model 141 are also different. Since AT132 and AT142 can be created from the different first AT prediction model 131 and the different second AT prediction model 141 and compared, the reliability of each AT132 and AT142 can be easily confirmed. For example, if AT132 and AT142 are the same or very similar values, both the first AT prediction model 131 and the second AT prediction model 141 are considered to be highly reliable. On the other hand, if AT132 and AT142 are not approximate values, it is determined that the reliability of at least one of the first AT prediction model 131 and the second AT prediction model 141 is low, and a group of training data 200 can be prepared again to recreate the first AT prediction model 131 or the second AT prediction model 141, whichever is less reliable.

[0083] According to this embodiment, the exercise prescription creation unit 160 creates and outputs an exercise prescription 161 based on both or either of the AT 132 predicted by the first AT prediction unit 130 and the AT 142 predicted by the second AT prediction unit 140, as well as the blood pressure, heart rate, double product (systolic blood pressure × heart rate), presence or absence of arrhythmia, electrocardiogram waveform, and oxygen saturation measured during exercise (step S107). According to this embodiment, the exercise prescription 161 can be created simply and accurately, eliminating subjectivity, based on the ATs 132 and 142 predicted simply and accurately, eliminating subjectivity.

[0084] Although the embodiments and modified examples of the present technology have been described above, the present technology is not limited to the above-described embodiments, and it goes without saying that various modifications can be made within the scope of the gist of the present technology. [Explanation of symbols]

[0085] 10. Information processing equipment 100 Anaerobic metabolic threshold prediction system 110 First AT prediction model generation unit 111 First Learning Model 120 Second AT prediction model generation unit 121 Second Learning Model 130 First AT Prediction Unit 131 First AT prediction model 132 AT 140 Second AT Prediction Unit 141 Second AT prediction model 142 AT 150 AT output section 160 Exercise Prescription Department 161 Exercise prescription 170 Optimal Solution Output AI Prediction Unit 200 training data 300 Target Data

Claims

1. V Ｏ２ (minute oxygen intake) over time, V ＣＯ２ (minute carbon dioxide emissions) over time, R (gas exchange ratio: V ＣＯ２ / V Ｏ２ ) longitudinal data and V Ｅ (minute ventilation) data over time, V Ｅ / V Ｏ２ and longitudinal data of V Ｅ / V ＣＯ２ and longitudinal data of F(P) ＥＴＣＯ２ (End-tidal carbon dioxide fractional concentration (partial pressure)) over time data, F(P) ＥＴＯ２ (End-tidal oxygen fraction concentration (partial pressure)) over time data, other time-series data and non-time-series data different from each of the time-series data; The rate of increase of R is V Ｏ２ The first criterion is the point at which the growth rate of V ＣＯ２ The rate of increase is V Ｏ２ The second criterion is that the growth rate of V Ｅ / V ＣＯ２ Without increasing V Ｅ / V Ｏ２ The third criterion is the point at which F(P) ＥＴＣＯ２ Without change, F(P) ＥＴＯ２ The fourth criterion is the point at which V Ｅ The rate of increase is V Ｏ２ The fifth criterion is that the growth rate of AT (anaerobic metabolic threshold) determined based on a first AT prediction model generation unit that inputs a group of teacher data obtained by a cardiopulmonary exercise stress test into a first learning model and executes the first learning model to determine at least one sixth criterion for predicting AT regarding the other time-course data and the non-time-course data, the at least one sixth criterion being different from the first to fifth criteria, and generates a first AT prediction model that predicts AT based on the first to sixth criteria; a first AT prediction unit that inputs target data including each of the time-series data and the non-time-series data of the same type as the time-series data and the non-time-series data included in the training data into the first AT prediction model and executes the first AT prediction model to predict the AT of the target data based on the first to sixth criteria of the target data; An anaerobic metabolic threshold prediction system comprising:

2. The anaerobic metabolic threshold prediction system according to claim 1, a second AT prediction model generation unit that inputs the group of teacher data into a second learning model different from the first learning model, and executes the second learning model to determine at least one seventh criterion related to the other time-course data and the non-time-course data for predicting AT, the at least one seventh criterion being different from the first to fifth criteria, and generates a second AT prediction model different from the first AT prediction model that predicts AT based on the first to fifth and seventh criteria; a second AT prediction unit that inputs the target data into the second AT prediction model and executes the second AT prediction model to predict the AT of the target data based on the first to fifth and seventh criteria of the target data; The anaerobic metabolic threshold prediction system further comprises:

3. The anaerobic metabolic threshold prediction system according to claim 2, an optimal solution output AI prediction unit that compares the AT predicted by the first AT prediction unit and the AT predicted by the second AT prediction unit from the common target data, determines whether the first AT prediction unit or the second AT prediction unit has high accuracy, and outputs the AT predicted by the first AT prediction unit or the second AT prediction unit that is determined to have high accuracy; The anaerobic metabolic threshold prediction system further comprises:

4. The anaerobic metabolic threshold prediction system according to claim 2, An optimal solution output AI prediction unit that outputs both the AT predicted by the first AT prediction unit and the AT predicted by the second AT prediction unit from the common target data. The anaerobic metabolic threshold prediction system further comprises:

5. The anaerobic metabolic threshold prediction system according to any one of claims 2 to 4, The first learning model and the second learning model different from the first learning model are a logistic regression model, a Lasso regression model, a decision tree model, an XG boost model, or a deep learning model that analyzes numerical values ​​or graph images of time-series data, respectively. Anaerobic metabolic threshold prediction system.

6. The anaerobic metabolic threshold prediction system according to any one of claims 1 to 5, An exercise prescription creation unit that creates an exercise prescription based on the predicted AT The anaerobic metabolic threshold prediction system further comprises:

7. The anaerobic metabolic threshold prediction system according to any one of claims 1 to 6, The other time-course data is time-course data of TV (tidal volume), and the sixth criterion is that the rate of increase of TV is low. The other time-course data is time-course data of RR (respiratory rate), and the sixth criterion is that the rate of increase of RR is high. The other time-varying data is time-varying data of HR (heart rate), and the sixth criterion is that the rate of increase of HR becomes high. the other time-varying data is time-varying data of HR, and the sixth criterion is that high frequency components of heart rate fluctuations disappear; The other time-course data is time-course data of BPs (systolic blood pressure), and the sixth criterion is that the rate of increase of BPs is high. The other time-course data is time-course data of DP (double product), which is the product of BPs and HR, and the sixth criterion is that the rate of increase of DP is high. the cardiopulmonary exercise stress test is performed using an ergometer, the other time-dependent data is time-dependent data of the rotation speed of the ergometer, and the sixth criterion is that the rotation speed temporarily increases; The sixth criterion is V Ｅ / V ＣＯ２ and / or The sixth criterion is that the rate of increase of TV / RR is low. Anaerobic metabolic threshold prediction system.

8. The anaerobic metabolic threshold prediction system according to any one of claims 2 to 5, The other time-dependent data is time-dependent data of TV (tidal volume), and the seventh criterion is that the rate of increase of TV is low and the rate of increase of TV is low; The other time-course data is time-course data of RR (respiratory rate), and the seventh criterion is that the rate of increase of RR is high. The other time-varying data is time-varying data of HR (heart rate), and the seventh criterion is that the rate of increase of HR becomes high. the other time-varying data is time-varying data of HR, and the seventh criterion is that high frequency components of heart rate fluctuations disappear; The other time-course data is time-course data of BPs (systolic blood pressure), and the seventh criterion is that the rate of increase of BPs is high. The other time-course data is time-course data of DP (double product), which is the product of BPs and HR, and the seventh criterion is that the rate of increase of DP is high. the cardiopulmonary exercise stress test is performed using an ergometer, the other time-dependent data is time-dependent data of the rotation speed of the ergometer, and the seventh criterion is that the rotation speed temporarily increases; The seventh criterion is V Ｅ / V ＣＯ２ and / or The seventh criterion is that the rate of increase of TV / RR is low. Anaerobic metabolic threshold prediction system.

9. The anaerobic metabolic threshold prediction system according to claim 6, The exercise prescription creation unit creates an exercise prescription based on the AT determined based on the predicted AT, as well as electrocardiogram changes during exercise, the presence and type of arrhythmia, blood pressure, and oxygen saturation. Anaerobic metabolic threshold prediction system.

10. The anaerobic metabolic threshold prediction system according to any one of claims 2 to 5, an nth AT prediction model generation unit that inputs the group of teacher data into an nth learning model different from the first learning model and the second learning model, and executes the nth learning model to determine at least one nth criterion related to the other time-course data and the non-time-course data for predicting AT, the at least one nth criterion being different from the first to fifth criteria, and generates an nth AT prediction model different from the first AT prediction model and the second AT prediction model that predicts AT based on the first to fifth and nth criteria; an nth AT prediction unit that inputs the target data into the nth AT prediction model and executes the nth AT prediction model to predict the AT of the target data based on the first to fifth criteria and the nth criteria of the target data; Further comprising: n is an integer of 3 or more Anaerobic metabolic threshold prediction system.

11. V Ｏ２ (minute oxygen intake) over time, V ＣＯ２ (minute carbon dioxide emissions) over time, R (gas exchange ratio: V ＣＯ２ / V Ｏ２ ) longitudinal data and V Ｅ (minute ventilation) data over time, V Ｅ / V Ｏ２ and longitudinal data of V Ｅ / V ＣＯ２ and longitudinal data of F(P) ＥＴＣＯ２ (End-tidal carbon dioxide fractional concentration (partial pressure)) over time data, F(P) ＥＴＯ２ (End-tidal oxygen fraction concentration (partial pressure)) over time data, other time-series data and non-time-series data different from each of the time-series data; The rate of increase of R is V Ｏ２ The first criterion is the point at which the growth rate of V ＣＯ２ The rate of increase is V Ｏ２ The second criterion is that the growth rate of V Ｅ / V ＣＯ２ Without increasing V Ｅ / V Ｏ２ The third criterion is the point at which F(P) ＥＴＣＯ２ Without change, F(P) ＥＴＯ２ The fourth criterion is the point at which V Ｅ The rate of increase is V Ｏ２ The fifth criterion is that the growth rate of AT (anaerobic metabolic threshold) determined based on inputting a group of training data obtained by a cardiopulmonary exercise stress test, including the above, into a first learning model, and executing the first learning model to determine at least one sixth criterion for predicting AT regarding the other time-course data and the non-time-course data, the at least one sixth criterion being different from the first to fifth criteria, and generating a first AT prediction model that predicts AT based on the first to sixth criteria; Target data including each of the time-series data and the non-time-series data of the same type as the time-series data and the non-time-series data included in the training data is input to the first AT prediction model, and the first AT prediction model is executed to predict a first AT of the target data based on the first to sixth criteria of the target data. Anaerobic metabolic threshold prediction method.

Images