Estimating the concentration of a respiratory gas in the blood of a patient

Abstract

To provide a method for estimating the concentration of a respiratory gas in the blood of a patient, a processing device, a program, a computer-readable medium, and medical engineering equipment.SOLUTION: Receiving measurement data indicative of a volume-dependency curve of a concentration of a respiratory gas in a respiratory airflow expired by a patient depending on a respiratory volume expired by the patient, generating input data (14) from the measurement data, the input data (14) comprising a matrix of values of various parameters associated with the volume-dependency curve, and inputting the input data (14) into a machine learning module (15) trained to transform the input data (14) into output data (17) indicative of the concentration of the respiratory gas in the blood of the patient; And outputting the output data (17) by the machine learning module (15).SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to a method for estimating the concentration of respiratory gases in the blood of a patient, and further to a method for training a machine learning module for use in such a method, as well as to a data processing device, a computer program, a computer readable medium, and medical engineering equipment for performing at least one of these methods. [Background technology]

[0002] Mechanical ventilation of anesthetized and critically ill patients requires the continuous maintenance of regular vital gas exchange. The gold standard for determining gas exchange in the clinical field is the analysis of vital gases in arterial blood samples. In this case, the partial pressures of carbon dioxide and oxygen in the blood are compared with the rate of inspired oxygen and alveolar ventilation to determine the presence or absence of respiratory failure. However, such blood gas analysis is invasive and time-consuming.

[0003] In addition, capnography can be used to measure alveolar carbon dioxide partial pressure (p A It is possible to measure CO2 non-invasively, but such measurements cannot yet replace conventional blood gas analysis, especially in anesthetized and critically ill patients. Summary of the Invention [Problem to be solved by the invention]

[0004] The object of the present invention can be seen as providing a method which allows determining the concentration of respiratory gases in a patient's blood non-invasively and with sufficient accuracy. Another object of the present invention can be seen as providing a method for training a corresponding machine learning module, a corresponding data processing device, a corresponding computer program, a corresponding computer-readable medium, and a corresponding medical engineering device. [Means for solving the problem]

[0005] These problems are solved by the subject matter of the independent claims. Advantageous embodiments of the invention are set out in the dependent claims, the following description and the accompanying drawings.

[0006] A first aspect of the present invention relates to a computer-implemented method for estimating the concentration of a respiratory gas in a patient's blood, the method comprising: receiving measurement data indicative of a volume-dependence curve of the concentration of a respiratory gas in a respiratory airflow exhaled by a patient depending on the respiratory volume exhaled by the patient; generating from the measurement data input data comprising a matrix of values ​​of various parameters related to the volume-dependence curve; inputting the input data to a machine learning module trained to convert the input data into output data indicative of the concentration of the respiratory gas in the patient's blood; and outputting the output data by the machine learning module.

[0007] A second aspect of the present invention relates to a computer-implemented method for training a machine learning module for a medical engineering device, the method comprising: receiving a plurality of measurement datasets each comprising measurement data indicative of a volume-dependence curve of a concentration of a respiratory gas in a respiratory airflow exhaled by a patient depending on the volume of respiratory gas exhaled by the patient, wherein the measurement data of the different measurement datasets are at least partially assigned to different patients; generating from the measurement datasets a plurality of training datasets, each training dataset being assigned to one of the patients and comprising a matrix of values ​​of different parameters related to the volume-dependence curve; inputting each training dataset as input data to a machine learning module configured to convert the input data into output data indicative of the concentration of the respiratory gas in the blood of a respective patient; outputting the output data by the machine learning module; determining a deviation of the output data from target data assigned to the respective training dataset (based on the output data); and adapting weights of the machine learning module in an optimization method to reduce the deviation.

[0008] These methods may be performed automatically by a processor, for example by a processor of a medical engineering device. The machine learning module used in the method according to the first aspect of the invention may have been trained using the method according to the second aspect of the invention. The method according to the first aspect of the invention may further comprise the steps of the method according to the second aspect of the invention.

[0009] The approach presented here is based on the finding that volumetric capnogram or oxigram curves can be analyzed using machine learning algorithms to estimate the concentration of respiratory gases in a patient's blood, particularly arterial blood. These curves change depending on the patient's pulmonary ventilation and perfusion fluctuations. These fluctuations can further increase inaccuracies in traditional estimations.

[0010] In contrast, the methods described above and below allow for highly accurate estimation of the concentration of respiratory gases in a patient's blood, even across diverse patients and / or with wide variations in lung function, with the advantage that invasive blood gas analysis may need to be performed less frequently or even be eliminated.

[0011] The methods described above and below have the further advantage that, compared to embodiments in which the concentration of respiratory gases in the patient's blood is estimated directly based on (raw) measurement data or in which (raw) measurement data is used as input data, they require significantly fewer calculation operations and allow for sufficiently accurate estimation even in cases where the measurement data is not very good. Furthermore, the risk of misinterpretation is reduced because the input data, unlike the (raw) measurement data, are predefined.

[0012] It is particularly advantageous if multiple parameters related to the curve are analyzed: such a multivariate approach provides more information for the estimation than if only a single parameter, such as the alveolar partial pressure of the respiratory gas in question, were analyzed, and in combination with a correspondingly configured machine learning algorithm allows for much more accurate and robust estimation.

[0013] Next, some terms will be explained in more detail.

[0014] A "patient" may be understood as a ventilated patient, ie a person or animal subject who is or is to be ventilated by a ventilator.

[0015] "Breathing gas" may be understood as, for example, one of the following gases: carbon dioxide, oxygen, nitrogen, water vapor, anaesthetic gas.

[0016] "Respiratory air" as in "respiratory volume" or "respiratory airflow" may be understood to mean a respiratory gas mixture including respiratory gas.

[0017] "Concentration" can generally be understood as a proportion or quantity, particularly a partial pressure.

[0018] The concentration of a respiratory gas in the patient's blood may be, for example, the arterial and / or venous partial pressure of the respiratory gas.

[0019] The measurement data may have been generated at least in part using a corresponding sensor system, for example a sensor system of a medical engineering device. For example, the measurement data may have been generated non-invasively by capnography and / or oxigraphy. In other words, the measurement data may be, for example, data from a capnogram and / or an oxigram.

[0020] The measurement data underlying the training data set can include real (i.e., obtained from actual measurements) data and / or simulated data, which, unlike real data, can be generated using a simulation environment in which various patient pulmonary conditions are simulated by a computer.

[0021] "Input data" can be understood as data that is different from and / or in contrast to the measured data and that is specifically adapted to the machine learning module. In particular, the input data can be compressed data compared to the measured data. The measured data can be input to another machine learning module that is trained to convert the measured data into input data.

[0022] A "machine learning module" may be understood as a hardware and / or software module for transforming input data into output data with an algorithm parameterized, i.e., trained, by machine learning. Such an algorithm may be, for example, an artificial neural network, a decision tree, a random forest, a k-nearest neighbor algorithm, a support vector machine, a Bayesian classifier, a k-means algorithm, a genetic algorithm, a kernel regression algorithm, a discriminant analysis algorithm, or a combination of at least two of these examples.

[0023] Each training data set input to the machine learning module can be assigned a set of predefined target data. The target data can include, for example, target values ​​of respiratory gas concentrations in the blood of each patient assigned to the respective training data set. In particular, the target data can represent the results of measurements (e.g., blood gas analysis) performed simultaneously and / or immediately before and / or after the measurement data underlying each training data set was generated. The target data can include real (i.e., obtained from actual measurements) data and / or simulated data. In contrast to real data, the simulated data can be generated using a simulation environment in which various patient pulmonary conditions are computer-simulated together with the respective training data set.

[0024] To determine the deviation of the output data from the target data, the output data and the target data can be input into a suitable loss function to calculate a score that quantifies the deviation, which can be calculated, for example, using the least squares method.

[0025] An "optimization method" can be understood as an iterative method for minimizing a loss function, such as a gradient method with backpropagation, in particular a stochastic gradient method.

[0026] A third aspect of the invention relates to a data processing apparatus, comprising a processor configured to perform at least one of the methods described above and below.

[0027] A "data processing device" can generally be understood as a computer. A data processing device can include hardware and / or software components. A data processing device can be, for example, a control device, a PC, a server, a laptop, a tablet, a smartphone, or a combination of at least two of these examples. Alternatively, a "data processing device" can be understood as at least one hardware and / or software component of at least one of these examples.

[0028] A "processor" may be understood as, for example, a central processing unit (CPU), a graphics processor, a tensor processing unit (TPU), or a combination of at least two of these examples.

[0029] In addition to the processor, the data processing device may include at least one of the following components: a storage device, a bus system for data communication between the processor and the storage device, and a data communication interface for wireless and / or wired data communication with peripheral devices.

[0030] It is noted that the features of the methods described above and below may also be features of a data processing device (and vice versa).

[0031] A fourth aspect of the invention relates to a medical engineering device comprising a sensor system for generating measurement data indicative of a volume-dependence curve of the concentration of respiratory gas in the respiratory airflow exhaled by the patient depending on the volume of respiratory air exhaled by the patient, and a data processing device as described above and below.

[0032] The medical engineering device may be, for example, a ventilator for invasive and / or non-invasive ventilation of a patient and / or a surveillance monitor for monitoring vital parameters of a patient.

[0033] The sensor system may include one or more gas sensors. By "gas sensor" we mean, for example, galvanic, paramagnetic, or optical sensors. Such gas sensors may be located in the main and / or side streams of exhaled breath and / or may be designed as pressure and / or flow sensors.

[0034] Another aspect of the invention relates to a computer program and a computer-readable medium on which the computer program is stored.

[0035] The computer program comprises instructions that, when executed by a processor, cause the processor (for example a processor of a data processing apparatus as described above and below) to perform at least one of the methods described above and below.

[0036] The computer-readable medium may be a volatile or non-volatile data storage device. For example, the computer-readable medium may be a hard disk, a universal serial bus (USB) storage device, a random-access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, or a combination of at least two of these examples. The computer-readable medium may also be a data communication network that allows the program code to be downloaded (e.g., via the Internet) or a cloud.

[0037] It is noted that features of the methods described above and below may also be features of a computer program and / or a computer readable medium (and vice versa).

[0038] Various embodiments of the present invention are described below, which should not be construed as limiting the scope of the present invention.

[0039] According to one embodiment, the measurement data may represent a volume-dependence curve associated with a single breath of the patient. In other words, each point on the volume-dependence curve may be assigned a specific percentage of the total volume of respiratory air, including respiratory gas, exhaled by the patient in a single breath. Thus, a volume of zero may be assigned to the beginning of the volume-dependence curve, and a volume equal to the total volume may be assigned to the end of the volume-dependence curve. Such a total volume may also be referred to as tidal volume or tidal volume. The measurement data may be generated during a single breath and / or may be newly generated and / or newly received for each breath.

[0040] In other words, the measurement data may include a series of concentration values ​​for the concentration of respiratory gas in the respiratory airflow and a series of quantity values ​​for the respiratory volume, each quantity value being from a range of values ​​bounded by a lower and upper limit value, the lower limit being zero and the upper limit indicating the total volume exhaled by the patient in a single breath, with each concentration value being assigned a different quantity value.

[0041] According to one embodiment, the measurement data may further indicate a positive end-expiratory pressure (PEP) assigned to the volume-dependent curve for ventilating the patient. In this case, the input data may include the PEP and / or may be generated taking the PEP into account. This allows for a more accurate estimation compared to embodiments that do not take the PEP into account.

[0042] According to one embodiment, the measured data can be used to determine a mathematical function that approximately defines at least a portion of the dose-dependence curve. At least one of the values ​​in the matrix can then be calculated using the mathematical function. The mathematical function can be used to determine the appropriate parameters in a predictable and transparent manner. The mathematical function can be a single mathematical function or a combination of multiple individual mathematical functions.

[0043] According to one embodiment, the input data may include a matrix of values ​​for 2-20, 10-20, or 10-15 different parameters associated with the dose-dependence curve. In this way, the consumption of computational resources may be significantly reduced compared to embodiments using larger input matrices.

[0044] Above and below, "parameter" can be understood as a ventilation parameter related to the ventilation of a patient. In this case, each value in the matrix can be assigned to one of the various parameters. Thus, depending on the number of various parameters, the matrix can include, for example, 2-20, 10-20, or 10-15 input values. The matrix can be understood as a one-, two-, or three-dimensional vector. For example, a machine learning module can be configured to convert these input values ​​into a single output value indicative of the concentration of respiratory gas in the patient's blood.

[0045] According to one embodiment, the measurement data may be generated in multiple successive time steps. In this case, a mathematical function may be determined using the measurement data from the various time steps, e.g., by regression. For example, a single time step may last as long as a single patient breath. However, the duration of each of the time steps may also be fixed and / or may be, for example, on the order of 1 ms, 10 ms, 100 ms, 1 s, or 10 s.

[0046] According to one embodiment, the mathematical function can be determined according to the Levenberg-Marquardt algorithm. The "Levenberg-Marquardt algorithm" can be understood as a special numerical optimization algorithm for solving nonlinear balance problems using least-squares techniques. This algorithm can be understood as a combination of the Gauss-Newton method and a regularization technique that enforces a decrease in function value. This allows for a more accurate and computationally efficient approximation than the classical Fowler method implementation, even when there is a large variation in the volume-dependence curve between consecutive breaths and / or between different patients.

[0047] According to one embodiment, the various parameters may include at least one of the following parameters: total volume of respiratory gas exhaled by the patient in a single breath; total volume of respiratory air including respiratory gas exhaled by the patient in a single breath (also called tidal volume or tidal volume); minute ventilation; alveolar ventilation; airway dead space; mixed expiratory pressure of respiratory gas; end-expiratory pressure of respiratory gas; positive end-expiratory pressure for ventilating the patient.

[0048] Minute ventilation can be interpreted as the product of tidal volume and respiratory rate.

[0049] Alveolar ventilation (

number

[0050] "Airway dead space" can generally be understood as the portion of the tidal volume that remains within the airways that deliver air with each breath and therefore does not reach the alveoli.

[0051] End-tidal pressure of a respiratory gas can be understood as the concentration of the respiratory gas in the respiratory airstream exhaled by the patient at the end of a single breath.

[0052] The mixed expiratory partial pressure of the respiratory gas may be understood as a particular fraction of the end-tidal partial pressure and / or may correspond to the average concentration of the respiratory gas in the respiratory airstream exhaled in a single breath, for example, by the patient.

[0053] According to one embodiment, the volume-dependence curve can be divided into at least three consecutive characteristic expiratory phases in a single breath of the patient, wherein the various parameters can include at least one of the following parameters: a first respiratory gas volume as the volume of respiratory gas exhaled by the patient in a first (e.g., earliest) of the expiratory phases, a second respiratory gas volume as the volume of respiratory gas exhaled by the patient in a second (e.g., middle) of the expiratory phases, a third respiratory gas volume as the volume of respiratory gas exhaled by the patient in a third (e.g., last or penultimate) of the expiratory phases, and a (e.g., average) slope of the volume-dependence curve in at least one of the expiratory phases, in particular in the last (or penultimate) of the expiratory phases.

[0054] An intermediate expiratory phase can be understood as an expiratory phase located between the earliest expiratory phase and the last (or penultimate) expiratory phase, where the first expiratory phase can transition directly into the second expiratory phase and / or the second expiratory phase can transition directly into the third expiratory phase.

[0055] The "slope" can be understood as the (e.g., average) rate of change of the concentration of the respiratory gas in the respiratory gas flow for each expiratory phase(s). Furthermore, the slope can be normalized in an appropriate manner to obtain a normalized slope, which can then be used as one of the parameters.

[0056] The expiratory phases can be determined, for example, using the mathematical functions described above and / or according to the Fowler method. The expiratory phases can be the typical (three or four) phases of a volumetric capnogram or oxigram. The expiratory phases can vary significantly in their length and / or the (e.g., average) slope of the volume-dependence curve.

[0057] For example, in the case of a capnogram, the first expiratory phase can extend from the beginning of expiration to a first point where the rate of change of the second derivative of the dose-dependence curve reaches a maximum or where the third derivative of the dose-dependence curve reaches a maximum on the left side. The second expiratory phase can extend from the first point to a second point where the third derivative of the dose-dependence curve reaches a maximum on the right side. The third expiratory phase can extend from the second point to the end of expiration. The term "dose-dependence curve" can also be understood here as an approximation of, for example, the shape of the aforementioned mathematical function.

[0058] The first expiratory phase may be the earliest phase of exhalation (10%-12% of the total breath) during which little or no carbon dioxide is present in the exhaled air. The second expiratory phase may be the phase of greatest (average) increase in carbon dioxide concentration in the respiratory airstream (15%-18% of the total breath). The third expiratory phase, in contrast to the preceding expiratory phases, may be the phase during which the carbon dioxide concentration in the respiratory airstream is determined primarily by gas leaving the alveoli (70%-75% of the total breath).

[0059] According to one embodiment, at least one normalized respiratory gas volume can be determined by dividing one of the three respiratory gas volumes by the total volume (of respiratory gas or of respiratory air containing respiratory gas) exhaled by the patient in a single breath, in which case the parameters can include at least one normalized respiratory gas volume in addition to or instead of the respective (non-normalized) first, second, or third respiratory gas volumes.

[0060] According to one embodiment, the machine learning module may include an artificial neural network. Thus, the weights of the machine learning module may be the weights of the artificial neural network. The artificial neural network may include an input layer for receiving input data, an output layer for receiving output data, and at least one intermediate layer between the input layer and the output layer. The artificial neural network may include, for example, at least one of the following network types: multi-layer perceptron, deep neural network (DNN), convolutional neural network (CNN), recurrent neural network (RNN), and long short-term memory (LSTM). For example, the artificial neural network may include up to 30, up to 15, or up to 5 trainable intermediate layers. Such a network architecture is particularly computationally efficient while still allowing for sufficiently accurate estimation.

[0061] According to one embodiment, the artificial neural network can be implemented as an adaptive neuro-fuzzy inference system (ANFIS). Such an inference system can be based on a Takagi-Sugeno controller and / or a Tsukamoto controller and / or can include a set of fuzzy if-then rules that can be trained to approximate a nonlinear function. For example, the architecture of an ANFIS can include a so-called fuzzification layer as the first of five layers. This embodiment allows combining the advantages of artificial neural networks and fuzzy logic in a single framework.

[0062] According to one embodiment, the measurement data can include at least first and second measurement data. In this case, the first measurement data can represent a volume-dependence curve of the concentration of a first respiratory gas in the respiratory airflow as a function of the respiratory volume, and the second measurement data can represent a volume-dependence curve of the concentration of a second respiratory gas different from the first respiratory gas in the respiratory airflow as a function of the respiratory volume. The output data can thus represent the concentrations of the first and second respiratory gases in the patient's blood. This embodiment allows for simultaneous estimation of the concentrations of various respiratory gases, such as carbon dioxide and oxygen, in the patient's blood.

[0063] According to one embodiment, the input data can include first input data generated from the first measurement data and second input data generated from the second measurement data. The first input data can be input to a first artificial neural network, and first output data indicative of the concentration of a first respiratory gas in the patient's blood can be output by the first artificial neural network. Similarly, the second input data can be input to a second artificial neural network, and second output data indicative of the concentration of a second respiratory gas in the patient's blood can be output by the second artificial neural network. Thus, the output data can include the first output data and the second output data. The weights of the machine learning module can be, for example, weights of the first or second artificial neural network. The two artificial neural networks can have different or identical architectures and / or weights and / or be trained separately. For example, at least one, or each, of the two artificial neural networks can have an architecture that is identical to the aforementioned network and / or can be implemented as an adaptive neuro-fuzzy inference system. Alternatively, the first input data and the second input data may be input to a common artificial neural network, such as the network described above.

[0064] For example, the first input data can include a matrix of first values ​​of various first parameters associated with the amount-dependence curve of the first measurement data, and / or the second input data can include a matrix of second values ​​of various second parameters associated with the amount-dependence curve of the second measurement data. The first parameters can be identical in number and / or type to the second parameters and / or can be different from the second parameters. At least one of the first values ​​can be determined using a first mathematical function that approximately defines at least a portion of the amount-dependence curve of the first measurement data (e.g., from various time steps), and / or at least one of the second values ​​can be determined using a second mathematical function that approximately defines at least a portion of the amount-dependence curve of the second measurement data (e.g., from various time steps).

[0065] Embodiments of the present invention will now be described with reference to the accompanying drawings, in which neither the description nor the drawings should be construed as limiting the scope of the invention. [Brief explanation of the drawings]

[0066] [Figure 1] FIG. 1 illustrates a ventilator according to an embodiment of the present invention. [Figure 2] 1 is a flow chart illustrating an embodiment of a method for estimating the concentration of respiratory gas in a patient's blood. [Figure 3] 1 is a flow chart illustrating an embodiment of a method for training a machine learning module for a ventilator. [Figure 4] FIG. 1 illustrates a capnogram for use in a method according to an embodiment of the present invention. [Figure 5] FIG. 1 illustrates a machine learning module for use in a method according to an embodiment of the present invention. [Figure 6] 1 is a flow chart illustrating the generation of a training data set in a method according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0067] The figures are purely schematic and not to scale. Where the same reference signs are used in different figures, these signify the same or identical functional features.

[0068] 1 shows a ventilator 1 for invasive and / or non-invasive ventilation of a patient. The ventilator 1 comprises a sensor system 3 for generating measurement data 5, a storage device 9 and a data processing device 7 having a processor 11 configured to execute a computer program stored in the storage device 9 for processing the measurement data 5 in at least one of the ways described below.

[0069] The measurement data 5 define a volume dependence curve 13 (see FIG. 4) of the concentration of respiratory gas (eg pCO2) in the respiratory airflow exhaled by the patient in dependence on the respiratory volume exhaled by the patient.

[0070] In particular, the measurement data 5 may show a volume-dependence curve 13 associated with a single breath of the patient. Furthermore, the measurement data 5 may show a positive end-expiratory pressure (abbreviated as PEEP) for ventilating the patient, which is assigned to the volume-dependence curve 13.

[0071] For example, as shown in Figure 4, the measurement data 5 may code a capnogram showing a dose-dependence curve 13 of the concentration of carbon dioxide as a respiratory gas, generated non-invasively by volumetric capnography. Additionally or alternatively, the measurement data 5 may code an oxigram showing a dose-dependence curve 13 of the concentration of oxygen as a respiratory gas, generated non-invasively by volumetric oxigraphy.

[0072] 2 shows an example of a method for estimating the concentration of respiratory gases in the blood of a patient. The method is carried out by a data processing device 7 and comprises the following steps:

[0073] In step 200, measurement data 5 is received.

[0074] In step 202, input data 14 is generated from the measurement data 5 (see also FIG. 5). The input data 14 comprises a matrix of values ​​of various parameters related to the dose dependency curve 13.

[0075] In step 204, the input data 14 is input to a machine learning module 15 ("module" above and below can be understood as a software and / or hardware module) that is trained to convert the input data 14 into output data 17 indicative of (estimated) concentrations of respiratory gases in the patient's blood. For example, the output data 17 can be an estimate of the partial pressure of arterial carbon dioxide (p a CO2) and / or arterial oxygen partial pressure (p a This may include an estimate of the oxygen concentration (O2).

[0076] In step 206, output data 17 is output by machine learning module 15.

[0077] The output data 17 may then be additionally further processed, for example for displaying it in an appropriate manner on a display of the ventilator 1 and / or for transmission via a wireless and / or wired data communication connection to an external device, such as a server, PC, laptop, tablet, smartphone or smartwatch.

[0078] 3 shows an example of a method for training the machine learning module 15. This method can also be performed by the data processing device 7. Additionally or alternatively, this method can be performed by a separate data processing device outside the ventilator 1. This method comprises the following steps:

[0079] In step 300, a plurality of measurement data sets 19 are received, each comprising measurement data 5 (see also FIG. 6 ). The measurement data 5 of the different measurement data sets 19 are at least partially assigned to different patients 21. For example, if the patient 21 is mechanically ventilated, the measurement data sets 19 can be generated in a plurality of controlled ventilation steps 23 of a specific length, for example 2 minutes, 5 minutes or 10 minutes. Each measurement data set 19 then comprises the measurement data 5 of exactly one breath of the respective patient 21.

[0080] In step 302, a plurality of training data sets 25 are generated from the measurement data sets 19 of various patients 21, each training data set 25 being assigned to one of the plurality of patients 21 and including a matrix of values ​​of various parameters associated with a respective volume-dependence curve 13. The training data sets 25 can be generated by a suitably configured selection module 27, which can be implemented, for example, as a separate machine learning module. As shown in FIG. 6, to generate the training data sets 25, it is possible to use only the measurement data sets 19 generated during the ventilation phase 23, for example, during the final moments. Thus, for a controlled breathing duration of 2 seconds, 30 measurement data sets are generated per ventilation phase 23 for each patient 21. Then, for example, a training data set 25 can be generated from each of these measurement data sets 19.

[0081] In step 304, each training data set 25 is input as input data 14 to a machine learning module 15. The machine learning module 15 converts the input data 14 into output data 17 indicative of the (estimated) concentration of respiratory gases in the blood of each patient 21.

[0082] In step 306, output data 17 is output by machine learning module 15.

[0083] In step 308, the deviation of the output data 17 from the target data 29 assigned to each training data set 25 or each input data 14 is determined.

[0084] Finally, in step 310, the weights of the machine learning module 15 are adapted in an optimization manner to reduce the deviation, for example, by gradient approximation with backpropagation.

[0085] Steps 308, 310 may be performed, for example, by a suitably configured optimization module 31.

[0086] Preferably, machine learning module 15 is trained over multiple successive time steps until the deviation reaches an acceptable value, in which case steps 304 to 310 can be performed, for example, at each time step.

[0087] As shown in Figures 5 and 6, the matrix can be a one-dimensional vector and / or can contain values ​​of at least 2, at least 5, at least 10 different parameters, in particular 11 or 12 different parameters. In order to keep the consumption of computational resources low, the number of parameters should not be too large, for example not more than 15, 20 or 30. The number of values ​​in the matrix can match the number of different parameters, i.e. each value in the matrix can define exactly one of the different parameters.

[0088] Depending on the embodiment, the various parameters may include the following parameters associated with each patient: the total volume of respiratory gas exhaled by the patient in a single breath; the total volume of respiratory gas, including respiratory gas exhaled by the patient in a single breath, also called tidal volume or tidal ventilation (V); T ), minute ventilation, alveolar ventilation, airway dead space (V Daw ), the mixed expiratory pressure of the respiratory gas (e.g., p E CO2), end-tidal pressure of respiratory gases (e.g., p etand positive end-expiratory pressure to ventilate the patient (see also Figure 4).

[0089] Furthermore, the volume-dependence curve 13 of each measurement data 5 can be divided into at least three consecutive characteristic expiratory phases for a single breath of each patient, wherein the various parameters can comprise at least one of the following parameters: a first volume of respiratory gas as the volume of respiratory gas exhaled by the patient in a first expiratory phase I, a second volume of respiratory gas as the volume of respiratory gas exhaled by the patient in a second expiratory phase II, a third volume of respiratory gas as the volume of respiratory gas exhaled by the patient in a third expiratory phase III, and an average slope of the volume-dependence curve 13 for at least one of the expiratory phases I, II, III, in particular for the third expiratory phase III.

[0090] Optionally, the three respiratory gas volumes can be normalized by dividing each by the tidal volume, in which case the various parameters can include each normalized respiratory gas volume.

[0091] The values ​​of these parameters can be calculated, at least in part, from the measurement data 5 using special mathematical functions 33. The mathematical functions 33 can also be used to detect the three expiratory phases I, II, III. Alternatively, the various parameters can be determined, at least in part, using additional machine learning modules.

[0092] The mathematical function 33 may be an approximation of at least a part of the volume dependence curve 13. This approximation may, for example, be determined by regression analysis of measurement data 5 generated for the patient at several successive time steps, for example over several breaths. A particularly accurate and computationally efficient approximation can be achieved if the mathematical function 33 is determined according to the Levenberg-Marquardt algorithm.

[0093] The machine learning module 15 may be implemented as an artificial neural network 35 (see FIG. 5 ), for example, having an input layer 37 for receiving input data 14, an output layer 39 for outputting output data 17, and at least one trainable hidden layer 41 for transforming the input data 14 into the output data 17. Alternatively, the machine learning module 15 may be implemented as a combination of multiple artificial neural networks. Other machine learning algorithms are also possible, such as decision trees, random forests, k-nearest neighbor algorithms, support vector machines, Bayesian classifiers, k-means algorithms, genetic algorithms, kernel regression algorithms, or discriminant analysis algorithms.

[0094] For example, to allow for rapid training and cost-effective hardware use in the ventilator 1, the network 35 should not be overly complex. Experiments have shown that architectures with up to 10 trainable hidden layers 41 are a good compromise between efficiency and accuracy. Networks 35 in the form of adaptive neuro-fuzzy inference systems (ANFIS) are particularly advantageous. However, much more complex DNN architectures are also possible.

[0095] The machine learning module 15 can also be trained to simultaneously estimate the concentrations of various respiratory gases in the patient's blood.

[0096] In this case, the measurement data 5 may comprise, for example, first and second measurement data, the first measurement data representing a dose-dependence curve 13 of the concentration of a first respiratory gas, e.g., carbon dioxide, in the respiratory airflow as a function of the respiratory volume, and the second measurement data representing a dose-dependence curve of the concentration of a second respiratory gas, e.g., oxygen, in the respiratory airflow as a function of the respiratory volume. Thus, a training data set 25 or input data 14 may be generated from the first and second measurement data, and the output data 17 may represent the (estimated) concentrations of the first and second respiratory gases in the blood of the respective patient.

[0097] For this purpose, the input data 14 can be input to a common artificial neural network 35 or to two separately trainable or trained artificial neural networks. In the second case, for example, first input data for the first network can be generated from first measurement data, and second input data for the second network can be generated from second measurement data. Thus, first output data indicative of the concentration of a first respiratory gas in the patient's blood is output by the first network, and second output data indicative of the concentration of a second respiratory gas in the patient's blood is output by the second network.

[0098] Training the machine learning module 15 may include, for example, the following steps.

[0099] First, blood samples are taken from various patients and the time of collection is recorded for each blood sample taken.

[0100] Blood samples are analyzed to determine each patient's actual arterial carbon dioxide and / or oxygen partial pressures.

[0101] Simultaneously with and / or immediately before and / or after taking the blood sample, the partial pressure of carbon dioxide or the partial pressure of oxygen in the respiratory airflow exhaled by each patient in several breaths is detected by a sensor at least during exhalation. Further, the flow of the respiratory airflow at least during exhalation is detected by a sensor.

[0102] By integrating the flow, the respiratory volume exhaled by the patient with each breath is determined.

[0103] A corresponding volumetric capnogram or oxigram is then generated from the time-based curve of the carbon dioxide or oxygen partial pressure in the respiratory airflow versus the respiratory volume.

[0104] Subsequently, in an optimization method, for example using regression analysis, a mathematical function is determined that approximately defines the curve in the capnogram or oxigram.

[0105] Then, based on mathematical functions, various parameters are determined that describe particular characteristics of the curve.

[0106] These parameters, together with the actual arterial blood carbon dioxide partial pressure or oxygen partial pressure values, respectively, are used to train the machine learning module 15 .

[0107] The practicality of the method was confirmed, inter alia, by the following experiments.

[0108] In 14 experimental animals that underwent lung lavage, the arterial carbon dioxide partial pressure (p a The CO2 was continuously measured. At the same time, a capnogram was measured for each breath. The animals were mechanically ventilated at fixed settings and with varying positive end-expiratory pressure (PEP) from 0 to 22 cmH2O. The resulting 8599 data points were analyzed as p a CO2 values ​​and each set of 12 parameters derived from respiratory capnograms were input into the ANFIS model in pairs. The data points were divided into a first set of 7370 data points (85%) for training the model and a second set of 1229 data points (15%) for testing the trained model. The ANFIS analysis was repeated in 10 independent steps, each with input data points selected randomly.

[0109] Bland-Altman plots of 10 ANFIS models tested independently of each other show the estimated and actual p aThe mean deviation between CO2 values ​​was 0.03 ± 0.03 mmHg, the range of agreement was 2.25 ± 0.42 mmHg, and the root mean square deviation was 1.15 ± 0.06 mmHg. Thus, the estimation was sufficiently accurate. The agreement index was calculated to be 95.5% (quadrant diagram) or 94.3% (polar diagram), indicating a good trend.

[0110] The advantages of this method are due, inter alia, to the following:

[0111] First, volume-based measurements are used instead of time-based measurements. In particular, capnography or oxigraphy can be used to monitor a number of compelling parameters related to gas exchange. These parameters provide a more robust data base for determining gas exchange than purely time-based measurements of exhaled respiratory gas flow.

[0112] Furthermore, it could be shown that assessing multiple parameters in parallel provides a more accurate estimate than assessing a single parameter (e.g., end-tidal carbon dioxide partial pressure).

[0113] Furthermore, the use of a machine learning module offers the advantage of being able to take into account non-linear hidden information related to breath-by-breath pulmonary gas exchange.

[0114] Finally, it is noted that the words "having", "comprising", "including", "with" and the like do not exclude other elements or steps, and that the indefinite article "a" or "an" does not exclude a plurality.

[0115] Furthermore, it is noted that features or steps described with reference to one of the above embodiments may also be used in combination with features or steps described with reference to other of the above embodiments.

[0116] Any reference signs in the claims should not be construed as limiting the scope of the subject matter defined by the claims. [Explanation of symbols]

[0117] 1 ventilator 3 Sensor system 5. Measurement data 7 Data Processing Device 9 Storage device 11 processors 13 Quantity dependence curve 14 Input Data 15 Machine Learning Module 17 Output Data 19 Measurement Datasets 21 patients 23 Ventilation process 25 training datasets 27 Selection Module 29 Target Data 31 Optimization Module 33 Mathematical Functions 35 Artificial Neural Networks 37 Input layer 39 Output layer 41 Middle Class 200 Receiving measurement data 202 Input Data Generation 204 Input data entry 206 Output data 300 Receiving measurement data sets 302 Generating a training dataset 304 Input data entry 306 Output data 308 Determination of deviation 310 Weight Adaptation Aw-alv inflection point (boundary between airways and alveoli) I. First expiratory phase II. Second expiratory phase III Third expiratory phase pCO2 partial pressure of carbon dioxide p a CO2 arterial carbon dioxide partial pressure p A CO2 alveolar carbon dioxide partial pressure p et CO2 end-tidal carbon dioxide partial pressure p E CO2 mixed exhaled carbon dioxide partial pressure V Respiratory volume V Daw Airway dead space V T tidal volume V Talv Alveolar tidal volume

Claims

1. The concentration of respiratory gases in the patient's (21) blood (p ａ CO ２ 1. A computer-implemented method for estimating a the concentration of respiratory gases (pCO2) in the respiratory airflow exhaled by the patient (21) depending on the respiratory volume (V) exhaled by the patient (21); ２ receiving (200) measurement data (5) indicative of a dose-dependence curve (13) of generating (202) from said measurement data (5) input data (14) comprising a matrix of values ​​of various parameters related to said dose dependency curve (13); The input data (14) is used to calculate the concentration of respiratory gases in the blood of the patient (21) (p ａ CO ２ inputting (204) the input data (14) into a machine learning module (15) trained to convert the input data (14) into output data (17) indicative of outputting (206) the output data (17) by the machine learning module (15); A method that encompasses

2. A computer-implemented method for training a machine learning module (15) for a medical engineering device (1), comprising: The concentration of respiratory gases (pCO ) in the respiratory airflow exhaled by the patient (21) depending on the respiratory volume (V) exhaled by said patient (21). ２ receiving (300) a plurality of measurement data sets (19) each including measurement data (5) indicative of a dose-dependence curve (13) of the serotonin-dependent serotonin level (S), the measurement data (5) of the different measurement data sets (19) being at least partially assigned to different patients (21); generating (302) from said measurement dataset (19) a plurality of training datasets (25), each training dataset (25) being assigned to one of a plurality of patients (21) and comprising a matrix of values ​​of various parameters related to said dose-dependence curve (13); The input data (14) is used to calculate the concentration of respiratory gases in the blood (p ａ CO ２ inputting (304) each training dataset (25) as input data (14) to a machine learning module (15) configured to convert each training dataset (25) into output data (17) indicative of the outputting (306) the output data (17) by the machine learning module (15); determining (308) deviations of the output data (17) from target data (29) assigned to each of the training data sets (25); Adapting (310) the weights of the machine learning module (15) in an optimization manner to reduce the deviation; A computer-implemented method comprising:

3. the measurement data (5) represent the volume-dependence curve (13) associated with a single breath of the patient (21); and / or The measurement data (5) further indicates a positive end-expiratory pressure (PEP) for ventilating the patient (21) assigned to the volume-dependence curve (13).

10. A method according to any one of the preceding claims.

4. a mathematical function (33) that approximately defines at least a portion of the dose-dependence curve (13) is determined using the measurement data (5), and at least one of the values ​​in the matrix is ​​calculated using the mathematical function (33); 10. A method according to any one of the preceding claims.

5. the measurement data (5) is generated at multiple successive time steps, and the mathematical function (33) is determined using the measurement data (5) from the various time steps; and / or said mathematical function (33) being determined according to the Levenberg-Marquardt algorithm; The method of claim 4.

6. The various parameters include the following parameters: total volume of respiratory gas exhaled in a single breath by the patient (21); total volume of respiratory air including the respiratory gas exhaled in a single breath by the patient (21) (V Ｔ ), minute ventilation, alveolar ventilation, airway dead space (V Ｄａｗ ), the mixed expiratory pressure of the respiratory gas (p Ｅ CO ２ ), the end-tidal pressure of the respiratory gas (p ｅｔ CO ２ ), positive end-expiratory pressure for ventilating the patient (21), and / or The volume-dependence curve (13) is divided into at least three consecutive characteristic expiratory phases (I, II, III) in a single breath of the patient (21), and the various parameters include at least one of the following parameters: a first volume of respiratory gas as the volume of respiratory gas exhaled by the patient (21) in a first phase (I) of the expiratory phases (I, II, III); a second volume of respiratory gas as the volume of respiratory gas exhaled by the patient (21) in a second phase (II) of the expiratory phases (I, II, III); a third volume of respiratory gas as the volume of respiratory gas exhaled by the patient (21) in a third phase (III) of the expiratory phases (I, II, III); a slope of the volume-dependence curve (13) in at least one of the expiratory phases (I, II, III), in particular in the last phase (III) of the expiratory phases (I, II, III).

10. A method according to any one of the preceding claims.

7. One of the three respiratory gas volumes is the total volume of respiratory air (V) including respiratory gas exhaled by the patient (21) in a single breath. Ｔ ) to determine at least one normalized respiratory gas volume, and the parameter comprises the at least one normalized respiratory gas volume. The method of claim 6.

8. the machine learning module (15) includes an artificial neural network (35); 10. A method according to any one of the preceding claims.

9. the artificial neural network (35) is implemented as an adaptive neuro-fuzzy inference system; The method of claim 8.

10. The measurement data (5) includes at least first measurement data and second measurement data, and the first measurement data is a concentration (pCO ２ ) and the second measurement data shows a dose-dependence curve (13) of the concentration of a second respiratory gas in the respiratory airflow depending on the respiratory volume (V), The output data (17) is the concentration (p ａ CO ２ ) and a concentration of the second respiratory gas; 10. A method according to any one of the preceding claims.

11. The input data (14) includes first input data generated from the first measurement data and second input data generated from the second measurement data; The first input data is input to a first artificial neural network to generate a concentration (p ａ CO ２ ) is output by the first artificial neural network; and the second input data is input to a second artificial neural network, and second output data indicative of the concentration of the second respiratory gas in the blood of the patient (21) is output by the second artificial neural network; The output data (17) includes the first output data and the second output data. The method of claim 10.

12. A data processing device (7), comprising a processor (11) configured to perform at least one of the methods according to any one of the preceding claims.

13. A medical engineering device (1), comprising: The concentration of respiratory gases (pCO ) in the respiratory airflow exhaled by the patient (21) depending on the respiratory volume (V) exhaled by said patient (21). ２ a sensor system (3) for generating measurement data (5) showing a dose-dependence curve (13) of A data processing device (7) according to claim 12, A medical engineering device comprising:

14. A computer program comprising instructions which, when executed by a processor (11), cause the processor (11) to carry out the method according to any one of claims 1 to 11.

15. A computer readable medium having stored thereon a computer program according to claim 14.

Images