Respiratory disease risk prediction device, respiratory disease risk prediction system, respiratory disease risk prediction method and low respiratory disease risk prediction program

Abstract

To provide a low-cost respiratory disease risk prediction device allowing even inexperienced medical personnel to recognize that a patient is at high risk of respiratory disease.SOLUTION: A respiratory disease risk prediction device 10 for predicting a respiratory disease risk of a patient on the basis of measurement information of a ventilator comprises: an image receiving unit that receives input of a graphic monitor image displayed on a graphic monitor 30 of the ventilator that displays the measurement information; a calculation unit that calculates information regarding the respiratory disease risk of the patient; and an output unit that outputs the calculated information regarding the respiratory disease risk, where the calculation unit has a learned calculation model that has been subjected to machine learning using teacher data including a graphic monitor image of a model lung and a subject as input and information regarding the presence or absence of respiratory disease onset in both the model lung and the subject as output such that the information regarding the respiratory disease risk is calculated when the graphic monitor image is input.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a respiratory disease risk prediction device, a respiratory disease risk prediction system, a respiratory disease risk prediction method, and a low respiratory disease risk prediction program. [Background technology]

[0002] In conventional artificial ventilation management in intensive care units, intensivists manage breathing based on the time curves of airway pressure, airflow, and ventilation volume displayed in real time on the graphic monitor of the ventilator. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Fernandez-Perez, ER, Hubmayr, RD Interpretation of airway pressure waveforms. Intensive Care Med 32, 658-659 (2006). Summary of the Invention [Problem to be solved by the invention]

[0004] On the other hand, intensive care specialists are mainly employed in large hospitals such as university hospitals, and in medium-sized or small general hospitals, general doctors manage respiratory management based on the airway pressure, flow rate, and ventilation volume values ​​displayed on the ventilator's graphic monitor.

[0005] However, numerically based respiratory management has variable diagnostic accuracy compared to time-course based respiratory management. Analyzing the time-course curve information of airway pressure, flow, and ventilation displayed in real time on a graphic monitor and determining appropriate ventilator settings requires advanced expertise and experience, making it difficult for anyone other than intensive care specialists. The lung condition of ventilated patients can improve or worsen over time. If the lung condition worsens, initially appropriate ventilator settings may no longer be appropriate. Leaving the settings unchanged can further worsen the lung condition, increasing the risk of developing serious respiratory diseases such as ventilator-associated lung injury (VILI) and spontaneous breathing-induced lung injury (P-SILI).

[0006] Therefore, there is a need for a low-cost respiratory disease risk prediction device that allows even inexperienced medical professionals, rather than experienced doctors, to recognize that patients are at high risk of respiratory disease. [Means for solving the problem]

[0007] The gist of the present invention is as follows. (1) A respiratory disease risk prediction device that predicts a patient's respiratory disease risk based on measurement information of a ventilator used by the patient, an image receiving unit that receives an input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; a calculation unit that calculates information about the patient's respiratory disease risk when the received graphic monitor image is input; and an output unit that outputs information related to the calculated respiratory disease risk Equipped with the calculation unit has a trained calculation model that has been subjected to machine learning using training data that includes, as an input, a graphic monitor image of the model lung, the subject, or both of them, and, as an output, information regarding the presence or absence of respiratory disease onset of the model lung, the subject, or both of them, so that information regarding the respiratory disease risk is calculated when the graphic monitor image is input; Respiratory disease risk prediction device. (2) A respiratory disease risk prediction device as described in (1) above, wherein the information regarding the respiratory disease risk includes information represented by images, numbers, letters, symbols, sounds, or combinations thereof. (3) The respiratory disease risk prediction device described in (1) or (2) above, wherein the information regarding the respiratory disease risk includes whether or not the patient has a respiratory disease, the type of respiratory disease, whether or not the ventilator settings need to be changed, whether or not the ventilator can be extubated, the level of respiratory disease risk, a respiratory disease risk score, a heat map showing the areas among the partial areas included in the graphic monitor image that affect the calculation of the information regarding the respiratory disease risk, or a combination thereof. (4) A respiratory disease risk prediction device described in any of (1) to (3) above, wherein the graphic monitor image is a captured image of a graphic monitor of the ventilator, a captured image of a graphic monitor displayed on the display of the respiratory disease risk prediction device connected to the ventilator, or a captured image of a graphic monitor displayed on the display of a computer connected to the ventilator. (5) The respiratory disease risk prediction device according to any one of (1) to (4) above, wherein the graphic monitor image includes a pressure waveform, a flow waveform, a ventilation volume waveform, or a combination thereof. (6) A respiratory disease risk prediction device according to any one of (1) to (5) above; the ventilator; A respiratory disease risk prediction system comprising: (7) A respiratory disease risk prediction method for predicting a patient's respiratory disease risk based on measurement information of a ventilator used by the patient, comprising: Accepting input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; inputting the received graphic monitor image to calculate information regarding the patient's respiratory disease risk; and outputting information relating to the calculated respiratory disease risk; Including, The calculation uses a trained calculation model that has been subjected to machine learning using training data including, as input, a graphic monitor image of the model lung, the subject, or both of them, and, as output, information regarding the presence or absence of respiratory disease onset, such that information regarding the respiratory disease risk is calculated when the graphic monitor image is input. Respiratory disease risk prediction methods. (8) A respiratory disease risk prediction program for predicting a patient's respiratory disease risk based on measurement information of a ventilator used by the patient, An image receiving function that receives an input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; and a calculation function that calculates information about the patient's respiratory disease risk when the received graphic monitor image is input; Including, The calculation function has a trained calculation model that has been subjected to machine learning using training data including, as input, a graphic monitor image of the model lung, the subject, or both of them, and, as output, information regarding the presence or absence of respiratory disease onset of the model lung, the subject, or both of them, so that information regarding the respiratory disease risk is calculated when the graphic monitor image is input. Respiratory disease risk prediction program. [Effects of the Invention]

[0008] According to the present invention, it is possible to provide a low-cost respiratory disease risk prediction device that enables even inexperienced medical personnel to recognize that patients who have previously been judged to be at high risk of respiratory disease by skilled doctors looking at the graphic monitor of a ventilator are at high risk of respiratory disease. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram showing an example of the configuration of a respiratory disease risk prediction system including the respiratory disease risk prediction device. [Figure 2]FIG. 2 is a diagram showing an example of the configuration of the respiratory disease risk prediction device. [Figure 3] FIG. 3 shows an example of a graphic monitor image measured on a patient. [Figure 4] FIG. 4 shows the graphic monitor image of FIG. 3 after the periphery has been trimmed to display only the measurement information waveform. [Figure 5] FIG. 5 is a configuration diagram of an example of a training data collection system when collecting training data TD used in machine learning. [Figure 6] FIG. 6 is a configuration diagram of an example of a machine learning system when machine learning is performed. [Figure 7] FIG. 7 is a schematic diagram showing an example of the calculation model M1. [Figure 8] FIG. 8 is a schematic diagram showing another example of the calculation model M1. [Figure 9] FIG. 9 is a configuration diagram of an example of respiratory disease risk prediction when predicting a patient's respiratory disease risk using the respiratory disease risk prediction device. [Figure 10] Figure 10 shows the learning curve and loss function when classifying the acquired graphic monitor images into four categories. [Figure 11] Figure 11 shows a heat map calculated based on a graphic monitor image that was determined to be normal using this respiratory disease risk prediction device. [Figure 12] Figure 12 shows a heat map calculated based on a graphic monitor image that was identified as asthma using this respiratory disease risk prediction device. [Figure 13] Figure 13 shows a heat map calculated based on graphic monitor images that were identified as COPD using this respiratory disease risk prediction device. [Figure 14] Figure 14 shows a heat map calculated based on graphic monitor images that were identified as ARDS using this respiratory disease risk prediction device. [Figure 15]Figure 15 shows a photograph of a model lung for which the risk of respiratory disease was predicted using this respiratory disease risk prediction device, and the appearance of an artificial respirator attached to the model lung. [Figure 16] Figure 16 shows the learning curve and loss function when classifying acquired graphic monitor images into four categories. DETAILED DESCRIPTION OF THE INVENTION

[0010] The present disclosure relates to a respiratory disease risk prediction device that predicts a patient's risk of respiratory disease based on measurement information from a ventilator used for the patient, the respiratory disease risk prediction device comprising: an image receiving unit that receives input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; a calculation unit that calculates information related to the patient's respiratory disease risk when the received graphic monitor image is input; and an output unit that outputs the calculated information related to the respiratory disease risk, wherein the calculation unit has a trained calculation model that has been subjected to machine learning using training data that includes, as input, a graphic monitor image of a model lung, a subject, or both, and, as output, information related to the presence or absence of respiratory disease on the model lung, the subject, or both, so that the information related to the respiratory disease risk is calculated when the graphic monitor image is input.

[0011] Conventional deep learning models use algorithms that refer to numerical values ​​obtained from waveform electrical signals.The respiratory disease risk prediction device (hereinafter referred to as this device) uses graphic monitor images obtained from the graphic monitor of the ventilator that medical professionals actually use as reference, and can generate a trained calculation model (prediction model) using the same data as that seen by intensive care specialists as training data.

[0012] This device can predict the risk of respiratory disease using the graphic monitor images displayed on the ventilator's graphic monitor. This device allows even inexperienced medical personnel to detect dangerous signs that appear on the ventilator's graphic monitor, enabling appropriate management of the ventilator. This device is also inexpensive because it can be used with conventional ventilators without the need for expensive dedicated equipment and software.

[0013] This device allows even inexperienced medical professionals, such as inexperienced doctors and nurses, to recognize that patients are at high risk of respiratory disease, allowing them to appropriately configure ventilators for patients at high risk of respiratory disease and prevent the onset of serious respiratory diseases.

[0014] This device can predict respiratory disease risk, enabling rapid and accurate decision-making regarding ventilator setting changes and minimizing the risk of ventilator-associated lung injury (VILI) and spontaneous breathing-induced lung injury (P-SILI). Predicting respiratory disease risk refers to predicting the risk of developing serious respiratory diseases such as VILI and P-SILI. This device can issue an alert regarding the need for ventilator setting changes if it determines a patient is at high risk of developing serious respiratory diseases. This allows medical professionals to change ventilator settings before serious respiratory diseases such as VILI and P-SILI develop. Ventilator settings include ventilation volume, oxygen concentration, positive end-expiratory pressure (PEEP), respiratory rate, breathing time, and airway pressure.

[0015] This device also allows non-specialists to easily interpret complex ventilator data, enabling rapid response at any time of the day or night, thereby reducing the burden on medical staff and improving the quality of patient care. Therefore, this device can also be applied to remote ICU and remote evaluation systems.

[0016] This device can also identify the type of respiratory disease. Respiratory diseases that can be identified are preferably normal, asthma, COPD (chronic obstructive pulmonary disease), ARDS (acute respiratory distress syndrome), etc. This device can also determine whether extubation is appropriate. Determining whether extubation is appropriate means determining whether the ventilator tube can be removed from the patient. This device's functions for identifying the type of respiratory disease and determining whether extubation is appropriate enable more accurate patient management and the formulation of treatment plans.

[0017] Furthermore, since this device uses a graphic monitor image as input, it can be used easily and at low cost across the boundaries of ventilator manufacturers with different measurement information data formats, etc. The graphic monitor image used as input can be a still image.

[0018] The respiratory disease risk prediction accuracy of this device was defined as having sufficient diagnostic capability when the trained calculation model showed a Kappa coefficient of 0.600 or higher. The Kappa coefficient is an index that evaluates the degree of agreement between the correct label and the value predicted by the trained calculation model. By achieving this desirable prediction accuracy, this device can be used in clinical respiratory disease risk prediction.

[0019] FIG. 1 is a schematic diagram showing an example of the configuration of a respiratory disease prediction system 100 (hereinafter also referred to as the present system) including the present device 10. The present device 10 can be connected to a patient, a model lung, or a ventilator used for a subject. A patient is someone whose respiratory disease risk is predicted using the present device 10, and a subject is someone from whom training data is obtained in the present device 10.

[0020] A model lung is a lung simulator that can reproduce the behavior of virtual lungs, from healthy lungs to lungs with various respiratory diseases such as restrictive and obstructive diseases. A model lung can be any lung that can reproduce the behavior of lungs with respiratory diseases such as normal, asthma, COPD, and ARDS, and examples include the Michigan Instruments Training Test Lung (TTL Model Lung).

[0021] Ventilators are commonly used in medical settings and can display measurement information waveforms measured on a patient, a model lung, or a test subject on a graphic monitor. The measurement information waveforms preferably include a pressure waveform, a flow waveform, a ventilation volume waveform, or a combination thereof. The pressure waveform is an airway pressure waveform that represents the time-dependent change in the circuit pressure measured inside the ventilator. The flow waveform is a waveform that represents the direction and speed (flow rate) of inhaled and exhaled air over time. The ventilation volume waveform is a waveform that represents the amount of gas entering the lungs and the amount of gas exhaled over time. Any ventilator can be used as long as it is capable of displaying the above-mentioned preferred measurement information waveforms on the graphic monitor 30, such as the Puritan Bennett (registered trademark) 980 series manufactured by Medtronic. The horizontal axis of the measurement information waveform graph displayed on the graphic monitor 30 represents time, and the graph is a moving image that moves from right to left on the screen in real time. A still image of the graphic monitor image obtained from the moving image displayed on the graphic monitor 30 can be used as input.

[0022] The graphic monitor image to be trained contains one or more types of measurement information waveforms, preferably two or more types, and more preferably three or more types. The measurement information waveforms contained in the graphic monitor image to be trained are preferably pressure waveforms, flow waveforms, ventilation volume waveforms, or combinations thereof. The measurement information waveforms contained in the graphic monitor image may be waveforms for multiple cycles contained in the entire graphic monitor image, or a waveform for one cycle out of multiple waveform cycles.

[0023] Mechanical ventilators typically display three waveforms on a graphic monitor: pressure waveform, flow waveform, and ventilation volume waveform. Experienced medical professionals determine treatment plans by taking into account the patient's condition based on the information from these three waveforms. However, inexperienced medical professionals may only pay attention to the numerical values ​​displayed on the graphic monitor and miss important information obtained from the waveforms, potentially leading to the development of serious respiratory diseases. Depending on the patient's current respiratory disease, the importance of certain measurement information waveforms may vary. By applying deep learning to one or more, preferably two or more, and more preferably three or more of the above waveforms, it is possible to enable mechanical ventilator setting management similar to that of experienced medical professionals.

[0024] The device 10 and the ventilator may be connected via a wired or wireless connection, such as Bluetooth (registered trademark), which allows for the transmission and reception of data for a graphic monitor image displayed on the graphic monitor 30 of the ventilator, or via a network 20 which allows for the transmission and reception of data for the graphic monitor image. The network 20 is a communications network such as the Internet. When the device 10 and the ventilator are connected as described above, a graphic monitor image can be displayed on the display of the device 10 or a display connected to the device 10, and the graphic monitor image displayed on the display can be captured (screenshot).

[0025] Preferably, the graphic monitor image is a captured image of the graphic monitor 30 of the ventilator, a captured image of the graphic monitor displayed on the display of the device connected to the ventilator, or a captured image of the graphic monitor displayed on the display of a computer separate from the device connected to the ventilator.

[0026] The device 10 does not need to be connected to a ventilator. If the device 10 is not connected to a ventilator, the graphic monitor image displayed on the graphic monitor 30 may be captured by a camera of the device 10, or may be captured by a camera such as a video camera, digital camera, tablet, or smartphone, and the captured graphic monitor image may be read into the device 10.

[0027] The device 10 is an information processing device such as a computer or server. The device 10 can be a personal computer, tablet, smartphone, or the like. The device 10 may also be an electronic medical record terminal equipped with a display. The device 10 has a calculation unit that calculates information regarding the patient's respiratory disease risk based on a graphic monitor image received by an image reception unit.

[0028] The device 10 may be configured as a single information processing device, or may be a collection of multiple physically separate information processing devices. In this case, each of the multiple information processing devices may have the same functions, or may have the functions of the single device 10 in a distributed manner.

[0029] FIG. 2 is a diagram showing an example of the configuration of the device 10. The device 10 has an image receiving unit with a receiving (receiving) function for acquiring a graphic monitor image of a patient measured with a ventilator, or a model lung or a test subject. The device 10 also has a calculation unit with a learning function for training a calculation model and a calculation function for calculating information related to the patient's respiratory disease risk using the trained calculation model, and an output unit with an output function for outputting information related to the calculated respiratory disease risk. The device 10 may include an image receiving unit 11, a memory unit 12, a display unit 13, an operation unit 14, and a processing unit 15.

[0030] The device 10 can acquire a graphic monitor image from the ventilator via the image receiving unit 11. The image receiving unit 11 can be a communication device implemented as hardware, firmware, communication software such as a TCP / IP driver or a PPP driver, or a combination of these. Communication via the communication device can be wireless or wired. The device 10 can acquire graphic monitor image data from the ventilator via the communication device. The communication device may receive graphic monitor image data from the ventilator via serial communication via a USB cable. The communication device may also have an interface circuit for short-range wireless communication according to a communication method such as Bluetooth (registered trademark) and receive radio waves from the ventilator. The communication device may also have a receiving circuit for receiving various signals corresponding to the graphic monitor image data via infrared communication or the like. The communication device may also have a communication interface circuit for a wired LAN.

[0031] Image receiving unit 11 may include an input / output device that detachably holds a portable storage medium instead of or in addition to the communication device. In this case, the input / output device acquires graphic monitor image data stored on the portable storage medium and supplies the acquired graphic monitor image data to processing unit 15.

[0032] The storage unit 12 is, for example, a semiconductor memory device such as a ROM or RAM. The storage unit 12 may be, for example, a magnetic disk, an optical disk, a magneto-optical disk, a nonvolatile semiconductor memory, or any other storage device capable of storing data. The storage unit 12 stores an operating system program, a driver program, an application program, data, and the like used for processing in the processing unit 15. The computer programs stored in the storage unit 12 may be downloaded online and installed in the storage unit 12, or may be installed in the storage unit 12 from a computer-readable portable recording medium such as a CD-ROM or DVD-ROM using a known setup program. The device 10 may be connected to the storage unit 12 and processing unit 15 implemented in the cloud to perform the functions of the storage unit 12 and processing unit 15.

[0033] The data stored in the storage unit 12 includes a calculation model M1, training data TD, etc., which will be described later. The storage unit 12 may also store data of a graphic monitor image acquired from the ventilator or data of a processed image of the graphic monitor image. The storage unit 12 may also temporarily store data related to a predetermined process.

[0034] The display unit 13 is a display. The display unit 13 may be a liquid crystal display, an organic EL display, etc. The display unit 13 displays the graphic monitor image data supplied from the processing unit 15, the results of calculation and processing of information related to the patient's respiratory disease risk, etc.

[0035] The information related to the respiratory disease risk calculated by the device is preferably an image, a numerical value, a character, a symbol, a sound, or a combination thereof. More preferably, the information related to the respiratory disease risk is the presence or absence of a respiratory disease, the type of respiratory disease, whether or not the ventilator settings need to be changed, whether or not the ventilator can be extubated, the level of the respiratory disease risk, a respiratory disease risk score, a heat map (saliency map) showing the regions of each partial region included in the graphic monitor image that affect the calculation of the information related to the respiratory disease risk, or a combination thereof.

[0036] Information relating to the respiratory disease risk calculated by the present device can be stored in the memory unit 12. The information relating to the respiratory disease risk stored in the memory unit 12 can be stored in the memory unit 12 together with the date and time when the information relating to the respiratory disease risk was calculated.

[0037] The presence or absence of respiratory disease refers to the current (real-time) presence or absence of respiratory disease in patients on ventilators.

[0038] The type of respiratory disease refers to the type of respiratory disease of a patient wearing a ventilator, including whether they are healthy (Normal). Examples of types of respiratory disease include Normal, Asthma, COPD, and ARDS. This device can distinguish the type of respiratory disease, making it possible to determine whether the current ventilator settings are appropriate, i.e., whether a change in settings is necessary.

[0039] The device can store the type of respiratory disease of the patient before or immediately after the application of a ventilator in the memory unit 12. Information about the patient's respiratory disease risk before or immediately after the application of a ventilator stored in the memory unit 12 can be stored in the memory unit 12 together with the date and time when the information about the respiratory disease risk was calculated.

[0040] The type of respiratory disease of the patient before wearing the device can be input from the operation unit 14 of the device or an external terminal connected to the device and stored together with the date and time in the memory unit 12. The type of respiratory disease of the patient immediately after wearing the device can be input from the operation unit 14 of the device or an external terminal connected to the device and stored together with the date and time in the memory unit 12, or the type of respiratory disease determined by the device can be stored together with the date and time in the memory unit 12 by the processing unit 15 of the device.

[0041] The type of respiratory disorder stored in the memory unit 12 can be compared with the type of respiratory disorder subsequently determined by the device to determine that a change in ventilator settings is necessary. Alternatively, the device can compare the respiratory status stored in the memory unit 12 with the respiratory status subsequently measured to determine that a change in ventilator settings is necessary.

[0042] The need for ventilator setting changes refers to whether or not ventilator settings should be changed. The lung condition of a ventilated patient can improve or worsen over time. For example, if the lung condition worsens, such as with obstructive, restrictive, or mixed lung disease, initially appropriate ventilator settings may no longer be appropriate. Leaving the settings unchanged could further worsen the lung condition, necessitating a change in ventilator settings. For example, for respiratory diseases that cause lung stiffness, such as ARDS, the airway pressure setting should be lowered. High PaCO2 requires setting changes such as increasing the tidal volume or respiratory rate. Poor oxygenation requires setting changes such as adjusting the oxygen concentration or positive end-expiratory pressure (PEEP).

[0043] The term "extubation of a ventilator" refers to whether or not it is OK to remove the ventilator from the patient. Once the patient's lung condition improves, the ventilator can be extubated.

[0044] High or low respiratory disease risk refers to the high or low risk of a patient on a ventilator developing a serious respiratory disease if the current ventilator settings are maintained. Respiratory disease risk score refers to the score of the risk of a patient on a ventilator developing a serious respiratory disease if the current ventilator settings are maintained.

[0045] The heat map can visualize areas that affect the calculation of information related to respiratory disease risk, with red representing a high impact, yellow representing a medium impact, green representing a low impact, and blue representing no impact, for example. When red or yellow is displayed on the heat map, medical professionals can recognize that the area has a high impact on the calculation of information related to respiratory disease risk. The heat map may be displayed separately from the graphic monitor image, or may be displayed overlaid or semi-transparently on the graphic monitor image. The heat map may be calculated by any method, such as Grad-CAM (Gradient-weighted Class Activation Mapping).

[0046] The operation unit 14 can be a keyboard, a mouse, and / or a pointing device such as a touch panel. A user of the device 10 can operate the device 10 using the operation unit 14. When operated by the user of the device 10, the operation unit 14 generates a signal corresponding to the operation. The generated signal is then supplied to the processing unit 15 as an instruction from the user. The graphic monitor image displayed on the display unit 13 can be captured automatically at a predetermined timing, preferably every 12 seconds, or at any timing by the operation unit 14.

[0047] The processing unit 15 is a processing device that loads the operating system program, driver program, application program, control program, etc. stored in the storage unit 12 into memory and executes instructions included in the loaded programs. The processing unit 15 is, for example, an electronic circuit such as a CPU (Central Processing Unit), MPU (Micro Processing Unit), DSP (Digital Signal Processor), or GPU (Graphics Processing Unit), or a combination of various electronic circuits.

[0048] Processing unit 15 may be realized by integrated circuits such as ASICs (Application Specific Integrated Circuits), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), MCUs (Micro Controller Units), etc. Although processing unit 15 is illustrated as a single component in Figure 2, processing unit 15 may also be a collection of multiple physically separate processors. For example, multiple processors operating cooperatively in parallel to execute instructions may be implemented.

[0049] The processing unit 15 can function as a reception processing unit 151, a conversion processing unit 152, a learning processing unit 153, a calculation processing unit 154, and an output processing unit 155 by executing various commands included in an application program (control program) stored in the storage unit 12. Examples of the functions of the reception processing unit 151, the conversion processing unit 152, the learning processing unit 153, the calculation processing unit 154, and the output processing unit 155 will be described with reference to FIGS. 3 to 9.

[0050] The receiving processor 151 receives the graphic monitor image data transmitted from the artificial ventilator via the image receiver 11 and stores the received graphic monitor image data in the memory 12.

[0051] The graphic monitor image, which is input data for the training data TD, is received and processed by the receiving processing unit 151. The receiving processing unit 151 receives data of the graphic monitor image of the model lung or the subject as input data, and generates training data TD in which the type of respiratory disease (including healthy) of the model lung or the subject is used as the correct answer label, for example, and stores this in the memory unit 12. Alternatively, it generates training data TD in which the correct answer label is a binary value indicating whether or not the model lung or the subject has a respiratory disease, and stores this in the memory unit 12. The binary value indicating whether or not a respiratory disease has developed can be a binary value in which 1 indicates the presence of a respiratory disease and 0 indicates the absence of a respiratory disease, or vice versa.

[0052] Figure 5 shows a configuration diagram of an example of a training data collection system for collecting training data TD used in machine learning. A ventilator equipped with a graphic monitor 30 measures the measurement information waveforms of a model lung, a subject, or both, to obtain a graphic monitor image. Additionally, for example, the respiratory disease type (including healthy individuals) of the model lung or the subject is also obtained, and deep learning is performed using the respiratory disease type as a correct label to generate a trained calculation model.

[0053] Data 121 of the graphic monitor image of the measured model lung, the subject, or both, and data 125 related to respiratory diseases of the model lung, the subject, or both, can be stored in the memory unit 12. The stored data can be displayed on the display unit 13.

[0054] As described above, data on the graphic monitor image and data 125 related to respiratory diseases are collected as training data TD. The data 125 related to respiratory diseases can be respiratory disease types (including healthy individuals) of the model lung, the subject, or both.

[0055] In the present device 10, the input data can be only the data of the graphic monitor image, but other data may also be included as input data. The other input data may be, for example, the age, sex, height, weight, Ppeak (peak airway pressure), Pmean (mean airway pressure), PEEP, f TOT (respiratory rate / min), I:E (inhalation to exhalation time ratio), V TE (expiratory tidal volume), V E TOT (minute ventilation), O2 (oxygen concentration), f (respiratory rate), V T (tidal volume) or other character data or numerical data.

[0056] The conversion processing unit 152 can process the data of the graphic monitor image stored in the memory unit 12 so as to display only the measurement information waveform portion to be measured, and store the processed data in the memory unit 12. The process of displaying only the measurement information waveform portion includes, for example, a process of trimming the surrounding area other than the measurement information waveform within a predetermined range, a process of converting colors other than those displaying the measurement information waveform to black, etc. By including only the measurement information waveform portion in the graphic monitor image, attention is focused on only the measurement waveform of the graphic monitor image when calculating information related to the patient's respiratory disease risk, and bias dependent on external indicators such as numbers and alarm marks can be avoided.

[0057] An example of a graphic monitor image measured on a patient is shown in Figure 3. The graphic monitor image in Figure 3 displays a graph with the horizontal axis representing time and the vertical axis representing the intensity of the measurement information waveform. The measurement information waveforms included in the graphic monitor image shown in Figure 3 are, from top to bottom, a pressure waveform, a flow waveform, and a ventilation volume waveform.

[0058] FIG. 4 shows a graphic monitor image in FIG. 3 after trimming the periphery so that only the measurement information waveform is displayed. FIG. 4 shows the graphic monitor image in FIG. 3 after trimming the periphery of the waveform display portion so that only the measurement information waveform portion is displayed. This trimming process prevents erroneous detection of components other than the waveform, further improving prediction accuracy. The device 10 can be equipped with a program for such trimming. Instead of or in addition to the trimming process, an extraction process may be performed to extract only the measurement information waveform portion. The extraction process can be achieved, for example, by a program that converts the color space of the image from RGB to HSV, specifies red, green, and blue wavelengths, and retains only those portions. The device 10 can be equipped with such a program.

[0059] Respiratory status is an important factor that affects the measurement information waveforms displayed on the graphic monitor 30 of a ventilator. To capture changes on the graphic monitor 30 that occur depending on the respiratory status, both inhalation and exhalation phases must be displayed on the same graphic monitor 30. In this specification, a phase refers to a time stage or moment when a specific physiological event occurs in the context of a ventilator. The inhalation and exhalation phases refer to their respective stages in the respiratory cycle. In other words, displaying both inhalation and exhalation phases on the same graphic monitor 30 means that a single graphic monitor image contains both the timing of inhalation and exhalation.

[0060] Spontaneous breathing is reported to occur at approximately 12–20 breaths per minute (Charilaos Chourpiliadis et al., Physiology, Respiratory Rate, StatPearls, Treasure Island, 2023), with each cycle of inspiration and expiration taking 3–5 seconds. In other words, a time axis of at least 6 seconds is required to capture one breathing cycle. The time span (span) of the graphic monitor image can be, for example, 6–16 seconds, 7–15 seconds, or 8–14 seconds. Graphic monitor images displayed at approximately 12-second spans may be acquired to capture the waveforms of three breaths at approximately 12–20 breaths per minute, or an average of approximately 16 breaths per minute. By acquiring graphic monitor images with a span similar to that of the graphic monitors typically viewed by physicians and other medical professionals, processing such as cropping the screen for a specified span is unnecessary, and the same amount of information can be acquired as input data as when a physician makes a judgment using only a screen capture.

[0061] The following describes the learning process performed by learning processing unit 153. Learning processing unit 153 generates a trained calculation model M1 trained using teacher data TD stored in storage unit 12, and stores the generated trained calculation model M1 in storage unit 12.

[0062] 6 shows a configuration diagram of an example of a machine learning system when machine learning is performed. During machine learning, data 121 of measured graphic monitor images stored in memory unit 12 and data 125 related to respiratory diseases are used as training data TD. Data related to respiratory diseases is, for example, data on the type of respiratory disease or data on the presence or absence of respiratory diseases.

[0063] In a learning device composed of a conversion processing unit 152 and a learning processing unit 153, graphic monitor image data 121 is preprocessed as desired, and the preprocessed graphic monitor image data is used as input data, and a calculation model is trained using training data TD with respiratory disease-related data 125 as output labels. The preprocessing is a trimming process and / or extraction process that displays only the waveform components of the graphic monitor image described above.

[0064] The calculation model M1 can be generated by updating the weighting variables of the learning model so that the error in the information on the risk of developing respiratory disease calculated (estimated) based on the input graphic monitor image data is reduced or consistent with the respiratory disease-related data 125, which is the output label. Deep learning is used to generate the calculation model M1 by machine learning, and learning methods such as the error backpropagation method, the feedback alignment method, the direct feedback alignment method, the synthetic gradient method, the target prop method, the difference target prop method, and the bootstrap method can be used.

[0065] In machine learning for generating a trained computational model, transfer learning may be used. Transfer learning is a deep learning technique that improves the accuracy of a trained computational model (AI model) by incorporating a trained model created using a large dataset into a model to be generated. For example, VGG16, a CNN model, can be used for transfer learning.

[0066] When the respiratory disease data 125 is a respiratory disease type corresponding to the graphic monitor image data, the weighting variables of the learning model can be updated to generate a calculation model M1 so that information about the respiratory disease risk calculated (estimated) based on the input graphic monitor image data matches the output label for the respiratory disease type. When the respiratory disease data 125 is a binary value of high or low respiratory disease risk corresponding to the graphic monitor image data, the weighting variables of the learning model can be updated to generate a calculation model M1 so that information about the respiratory disease risk calculated (estimated) based on the input graphic monitor image data matches the output label for an output label where 1 indicates high and 0 indicates low.

[0067] Graphic monitor images of the model lungs, subjects, and patients are acquired as time series, preferably once every 12 seconds, as data for each time point. Graphic monitor images of the model lungs and subjects are measured for the number of model lungs and subjects.

[0068] The vertical axis of the pressure waveform is in cmH2O, and normal values ​​for the pressure waveform are generally 10 to 25 cmH2O, so the range of the vertical axis may be, for example, -20 to 40 cmH2O, which is the range normally used in clinical practice.

[0069] The vertical axis of the flow waveform is in L / min, and since normal values ​​for the flow waveform are generally 20 to 80 L / min, the range of the vertical axis may be, for example, INSP 120 L / min to EXP 120 L / min, which is the range commonly used in clinical practice.

[0070] The vertical axis of the ventilation volume waveform is in mL, and since normal values ​​for the ventilation volume waveform are generally 200 to 800 mL, the range of the vertical axis may be, for example, 0 to 1000 mL, which is the range normally used in clinical practice.

[0071] The average resolution (pixels) of the graphic monitor image used for input in this device is, for example, 512 x 512 for a graphic monitor image that displays, for example, 1 to 3 or 1 to 4 types of measurement information waveforms, and for a graphic monitor image that has been cropped to display only the measurement information waveforms.

[0072] Fig. 7 is a schematic diagram showing an example of a calculation model M1. The calculation model M1 shown in Fig. 7 is a neural network model having an input layer M1-1, a hidden layer M1-2, and an output layer M1-3. A graphic monitor image of training data TD is input to each of the so-called "neurons" of the input layer M1-1. The number of neurons in the hidden layer M1-2 may be more or less than the number of neurons in the input layer M1-1.

[0073] Fig. 8 is a schematic diagram showing another example of the calculation model M1. The calculation model M1 shown in Fig. 8 is a deep neural network model including a convolutional neural network having an input layer M1-1, a convolutional layer M1-2, a pooling layer M1-3, and an output layer M1-4. There may be two or more convolutional layers M1-2 and two or more pooling layers M1-3. A graphic monitor image of the training data TD is input to each neuron of the input layer M1-1.

[0074] The learning processing unit 153 generates or updates a calculation model in which the weights of each neuron in the neural network are learned by performing known machine learning using the training data TD. The learning processing unit 153 may also generate or update a calculation model in which the weights of each neuron in the multi-layered neural network are learned by performing known deep learning using the training data TD.

[0075] 9 shows an example configuration diagram of the present system 100 when predicting information related to a patient's risk of respiratory disease using the present device 10. A predictor made up of a conversion processing unit 152, a learning processing unit 153, and an output processing unit 155 preprocesses graphic monitor image data measured by the ventilator as desired, and calculates information related to the patient's risk of respiratory disease using a trained calculation model M1 with the preprocessed graphic monitor image data as input data.

[0076] The calculation processing unit 154 executes calculation processing. The calculation processing unit 154 uses the received graphic monitor image data as input data and the trained calculation model M1 to calculate information about the patient's respiratory disease risk.

[0077] The calculation processing unit 154 transmits the calculation result to the output processing unit 155. The output processing unit 155 outputs information on the calculation result. The output of the information is displayed on the display unit 13 and / or transmitted to another device, etc.

[0078] The calculation process may be performed as desired by the conversion processing unit 152, the calculation processing unit 154, and the output processing unit 155. In the calculation process, when data of a graphic monitor image of a patient is acquired from a ventilator, information on the patient's respiratory disease risk corresponding to the acquired graphic monitor image is calculated.

[0079] The present disclosure also relates to a respiratory disease risk prediction method for predicting a patient's risk of respiratory disease based on measurement information from a ventilator used for the patient, the method including: accepting input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; inputting the accepted graphic monitor image to calculate information related to the patient's respiratory disease risk; and outputting the calculated information related to the respiratory disease risk, wherein the calculating uses a trained calculation model that has been subjected to machine learning using training data that includes, as input, a graphic monitor image of a model lung, a subject, or both, and, as output, information related to the presence or absence of respiratory disease on the model lung, the subject, or both, so that the information related to the respiratory disease risk is calculated when the graphic monitor image is input.

[0080] The present disclosure also relates to a respiratory disease risk prediction program for predicting a patient's risk of respiratory disease based on measurement information from a ventilator used by the patient, the program including: an image receiving function for receiving an input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; and a calculation function for calculating, upon input of the received graphic monitor image, information about the patient's respiratory disease risk, wherein the calculation function has a trained calculation model that has been subjected to machine learning using training data that includes, as input, a graphic monitor image of a model lung, a subject, or both, and, as output, information about the presence or absence of a respiratory disease on the model lung, the subject, or both, so that the information about the respiratory disease risk is calculated upon input of the graphic monitor image. The same description of the respective components included in the respiratory disease risk prediction device described above may be applied to the respective components included in the respiratory disease risk prediction method and the respiratory disease risk prediction program. [Example]

[0081] Example 1 The respiratory disease risk of the model lung was estimated using this device 10, using a neural network as a calculation model. The Puritan Bennett® 980 series ventilator manufactured by Medtronic was used as the ventilator connected to this device 10. The model lung used was a training test lung (TTL model lung) manufactured by Michigan Instruments. Figure 15 shows photographs of the TTL model lung, whose respiratory disease risk was predicted using this respiratory disease risk prediction device, and the ventilator attached to the TTL model lung.

[0082] In the machine learning to generate a trained calculation model, graphic monitor images of TTL model lungs showing four types of lung conditions (respiratory disease types), including a healthy state, measured by a ventilator, were used as training data, and machine learning was performed using the respiratory disease types corresponding to the graphic monitor images as output labels. The respiratory disease types were normal, asthma, COPD, and ARDS. In the following examples, graphic monitor images were used that included all of the waveforms for multiple cycles contained in the 12-second span graphic monitor of the ventilator.

[0083] The calculation model in the present device 10 may be a neural network, a random forest, a support vector machine (SVM), or the like.

[0084] The ventilator settings were assumed to be for an adult male of 170 cm and 66 kg, and the tidal volume was set to 400 mL (6 mL / kg), 530 mL (8 mL / kg), and 790 mL (12 mL / kg), the respiratory rate to 8, 16, 24, and 32 times, the inspiratory time to 0.6 seconds, 1.2 seconds, and 1.8 seconds, and PEEP to 2 cmH2O, 5 cmH2O, and 10 cmH2O. The graphic monitor image displayed on the graphic monitor 30 of the ventilator was obtained using the screen capture function.

[0085] The TTL model lung settings were adjusted to create respiratory disease conditions: normal, asthma, COPD, and ARDS, by varying compliance and airway resistance. For normal, airway resistance was set to 5 cmH2O / L / sec and lung compliance was fixed at 50 mL / cmH2O. For asthma, airway resistance was set to 20 and 50 cmH2O / L / sec and lung compliance was fixed at 50 mL / cmH2O. For COPD, airway resistance was set to 20 and 50 cmH2O / L / sec and lung compliance was fixed at 80 mL / cmH2O. For ARDS, airway resistance was set to 5 cmH2O / L / sec and lung compliance was set to 10, 20, and 30 mL / cmH2O.

[0086] After the above settings for the TTL model lung were made and operation was stable, a graphic monitor image displayed on the graphic monitor 30 of the ventilator was acquired using the screen capture function. Figure 3 shows an example of the acquired graphic monitor image.

[0087] 108 graphic monitor images of the normal TTL model lungs were acquired. 144 graphic monitor images of the asthma TTL model lungs were acquired. 198 graphic monitor images of the COPD TTL model lungs were acquired. 324 graphic monitor images of the ARDS TTL model lungs were acquired.

[0088] The acquired graphic monitor image was trimmed to include only the measured waveform. Figure 4 shows an example of the trimmed graphic monitor image.

[0089] Approximately 80% of the trimmed graphic monitor images were used as training data, and the remaining approximately 20% were used as test data for inference evaluation. Four types of output labels for the training data TD were used: Normal (0), Asthma (1), COPD (2), and ARDS (3), and a neural network calculation model was trained using machine learning.

[0090] For machine learning to generate a trained computational model, we used transfer learning with VGG16, which consists of 16 layers: 13 convolutional layers and 3 fully connected layers.

[0091] The neural network (deep learning library) used was Keras version 2.14.0. The analysis hardware used was an Intel Core i9 CPU, an NVIDIA RTX A6000 48GB GPU, and Microsoft Windows 10 Home OS, and JMP Pro 14 was used for statistical analysis.

[0092] The calculation model was trained using binary cross-entropy as the function loss, Adam as the optimization method, with 30 epochs and a batch size of 8. A confusion matrix was created, and the F1 score, precision, recall, and accuracy were obtained, with the Kappa coefficient used as the final evaluation index.

[0093] After generating the trained computational model, the image dataset for inference evaluation was used to visualize the regions evaluated by the trained computational model using a gradient class activation map (heat map) (Grad-CAM). The heat map is a two-dimensional image created by calculating the importance of each region based on the classification results of four types: Normal, Asthma, COPD, and ARDS. The red and yellow areas on the heat map indicate the regions that the trained computational model considered important in distinguishing between the four types.

[0094] Five-fold cross-validation was performed to mitigate the impact of data bias caused by randomly splitting the training and test data. Furthermore, stratified kFold was used to avoid bias in the classification of respiratory diseases when creating the five datasets in the five-fold cross-validation. A calculation model was trained and evaluated on each dataset, and the Kappa coefficient for each was calculated. The median Kappa coefficient for the five trained calculation models was used as the final result.

[0095] Figure 10 shows the learning curves and loss functions when the acquired graphic monitor images are classified into four categories: Normal, Asthma, COPD, and ARDS. The solid lines (Train Loss, Train Accuracy) represent the training data, and the dashed lines (Validation Loss, Validation Accuracy) represent the test data. Train Loss is the loss in the training data, and Validation Loss is the loss in the accuracy validation data. Furthermore, Train Accuracy is the accuracy rate in the training data, and Validation Accuracy is the accuracy rate (precision) in the accuracy validation data. The accuracy rate is a numerical value that indicates "what percentage of the output results of the trained calculation model are correct." The learning curve and loss function of the test data follow the learning curve and loss function of the training data, indicating that the trained calculation model has been trained appropriately.

[0096] Five trained calculation models (Models 1 to 5) were generated for Dataset 1 to Dataset 5. Model 4 was the median trained calculation model, with a Kappa coefficient of 0.924. The Accuracy, Precision, Recall, and F1Score of Model 4 were 0.946, 0.958, 0.928, and 0.937, respectively.

[0097] The Kappa coefficient for Model 4, which obtained the median value, was a high value of 0.924, indicating that a trained calculation model was generated that can distinguish between the four TTL model lung conditions of Normal, Asthma, COPD, and ARDS using graphic monitor images.

[0098] 11 to 14 show four gradient class activation maps (heat maps) based on graphic monitor images obtained using Grad-CAM for four types of TTL model lungs: normal, asthma, COPD, and ARDS.

[0099] In images classified as normal, there was a tendency for gazes to focus on all areas of the pressure waveform, flow waveform, and ventilation volume waveform. In images classified as asthma, there was a tendency for gazes to focus on the expiratory timing area of ​​the ventilation volume waveform. In images classified as COPD, there was also a tendency for gazes to focus on the expiratory timing area of ​​the ventilation volume waveform. In images classified as ARDS, there was a tendency for gazes to focus on the pressure waveform area. In the resulting heat map, gazes were focused on the same areas as those viewed by intensive care specialists.

[0100] We obtained graphic monitor images from the ventilator showing four types of TTL model lung conditions: normal, asthma, COPD, and ARDS, and were able to generate a trained calculation model that could distinguish the condition of each TTL model lung.

[0101] Example 2: Spontaneous Breathing Trial Using this device, we conducted a spontaneous breathing trial (SBT) on 27 patients on ventilators to determine whether they could maintain breathing without ventilator support, i.e., whether they could be extubated with the ventilator set to its weakest setting.

[0102] Using a neural network as a calculation model, the patient's SBT was performed by the device 10. As the ventilator connected to the device 10, a Puritan Bennett (registered trademark) 980 series manufactured by Medtronic was used.

[0103] After 30 minutes of ventilator operation with each patient on the lowest setting, 17 graphic monitor images were acquired for 17 patients who passed the SBT and 10 graphic monitor images for 10 patients who failed the SBT. The graphic monitor images were then cropped to include only the measured waveforms. Passing the SBT meant the patient was ready to be extubated, while failing the SBT meant the patient could not be extubated.

[0104] Of the 17 trimmed graphic monitor images that were judged to have passed the SBT, 11 were used as training data and 6 as test data, and of the 10 trimmed graphic monitor images that were judged to have failed the SBT, 6 were used as training data and 4 as test data to generate and verify a trained calculation model.

[0105] Figure 16 shows the learning curves and loss functions when classifying acquired graphic monitor images into four categories: Normal, Asthma, COPD, and ARDS. The solid lines (Train Loss, Train Accuracy) represent the training data, and the dashed lines (Validation Loss, Validation Accuracy) represent the test data. The learning curve and loss function for the test data follow those for the training data.

[0106] The accuracy rate for predicting SBT pass and SBT fail for the 10 test data was 90%. [Explanation of symbols]

[0107] 10 Respiratory disease risk prediction device 11 Image Reception Section 12 Storage section 121 Graphic monitor image data 125 Data on respiratory diseases 13 Display section 14 Control section 15 Processing section 20 Network 30 Graphics Monitor 100 Respiratory Disease Risk Prediction System 151 Receiving processing unit 152 Conversion processing section 153 Learning processing unit 154 Calculation processing unit 155 Output Processing Unit M1 calculation model TD teacher data

Claims

1. A respiratory disease risk prediction device that predicts a patient's respiratory disease risk based on measurement information of a ventilator used by the patient, an image receiving unit that receives an input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; a calculation unit that calculates information about the patient's respiratory disease risk when the received graphic monitor image is input; and an output unit that outputs information related to the calculated respiratory disease risk Equipped with the calculation unit has a trained calculation model that has been subjected to machine learning using training data that includes, as an input, a graphic monitor image of the model lung, the subject, or both of them, and, as an output, information regarding the presence or absence of respiratory disease onset of the model lung, the subject, or both of them, so that information regarding the respiratory disease risk is calculated when the graphic monitor image is input; Respiratory disease risk prediction device.

2. The respiratory disease risk prediction device according to claim 1 , wherein the information relating to the respiratory disease risk includes information displayed as an image, a number, a letter, a symbol, a sound, or a combination thereof.

3. 2. The respiratory disease risk prediction device of claim 1, wherein the information regarding the respiratory disease risk includes whether or not the patient has a respiratory disease, the type of respiratory disease, whether or not the ventilator settings need to be changed, whether or not the ventilator can be extubated, the level of respiratory disease risk, a respiratory disease risk score, a heat map showing areas among each partial area included in the graphic monitor image that affect the calculation of the information regarding the respiratory disease risk, or a combination thereof.

4. The respiratory disease risk prediction device of claim 1, wherein the graphic monitor image is a captured image of a graphic monitor of the ventilator, a captured image of a graphic monitor displayed on a display of the respiratory disease risk prediction device connected to the ventilator, or a captured image of a graphic monitor displayed on a display of a computer connected to the ventilator.

5. The respiratory disease risk prediction device according to claim 1 , wherein the graphic monitor image includes a pressure waveform, a flow waveform, a ventilation volume waveform, or a combination thereof.

6. The respiratory disease risk prediction device according to any one of claims 1 to 5, the ventilator; A respiratory disease risk prediction system comprising:

7. A respiratory disease risk prediction method for predicting a patient's respiratory disease risk based on measurement information of a ventilator used by the patient, comprising: Accepting input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; inputting the received graphic monitor image to calculate information regarding the patient's respiratory disease risk; and outputting information relating to the calculated respiratory disease risk; Including, The calculation uses a trained calculation model that has been subjected to machine learning using training data including, as input, a graphic monitor image of the model lung, the subject, or both of them, and, as output, information regarding the presence or absence of respiratory disease onset, such that information regarding the respiratory disease risk is calculated when the graphic monitor image is input. Respiratory disease risk prediction methods.

8. A respiratory disease risk prediction program that predicts a patient's respiratory disease risk based on measurement information of a ventilator used by the patient, An image receiving function that receives an input of a graphic monitor image displayed on a graphic monitor of the ventilator that displays the measurement information; and a calculation function that calculates information about the patient's respiratory disease risk when the received graphic monitor image is input; Including, The calculation function has a trained calculation model that has been subjected to machine learning using training data including, as input, a graphic monitor image of the model lung, the subject, or both of them, and, as output, information regarding the presence or absence of respiratory disease onset of the model lung, the subject, or both of them, so that information regarding the respiratory disease risk is calculated when the graphic monitor image is input. Respiratory disease risk prediction program.

Images