Biological suction device and biological suction program

Abstract

To provide a biological suction device that achieves generation of information related to suction by means of AI technology.SOLUTION: A biological suction device according to the present invention comprises: a suction tank that communicates with a suction tube inserted into a biological body being a suction target to apply a negative pressure and suction a suction object from the biological body; an air suction unit that applies a suction pressure to the suction tank; and a control unit. Furthermore, the control unit comprises: a history part that acquires and stores the history of the suction treatment (hereinafter referred to as "suction history") of the suction object; a learning model part that collects the suction history from the history part and / or the outside as learning data and performs machine learning of the characteristics of the suction history using the learning data; and an estimation part that generates information related to the suction by inputting the suction history stored in the history part into the machine-learned learning model part.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a suction device for a living body that sucks unnecessary objects (such as secretions such as sputum, nasal mucus, and blood, and foreign substances) from a living body, and a suction program for a living body. In particular, the present invention relates to AI (Artificial Intelligence)-related technologies adapted to suction technologies for living bodies.

Background Art

[0002] Conventionally, a technique of a sputum suction device that sucks sputum entangled in the airway of a suction target (such as a person under nursing or care) and removes it as a suction product has been known.

[0003] For example, Patent Document 1 discloses a technique of "a sputum suction device including a storage container that communicates with a suction catheter and stores the sucked sputum, and an air pump that communicates with the storage container and generates a negative pressure in the suction catheter via the storage container".

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In a conventional sputum suction device such as Patent Document 1, it is preferable that a user who uses the device performs an appropriate suction treatment based on knowledge and experience regarding the suction target and the suction treatment.

[0006] However, in the case of a user with little experience, there has been a problem that it is difficult to appropriately perform a suction treatment due to a lack of knowledge and information regarding suction (for example, prediction of the amount of suction product generated and prediction of the frequency at which suction should be performed).

[0007] For example, in the object to be aspirated, there may be a small amount of sputum and it may not be clogged very much. If the state is not fully grasped and the aspiration frequency becomes excessively high, an unnecessary burden due to sputum aspiration will be imposed on the object to be aspirated, which is not preferable.

[0008] Also, for example, in the object to be aspirated, there may be a large amount of sputum and it may become clogged repeatedly. If the state is not fully grasped and the aspiration frequency becomes excessively low, the burden of sputum clogging will be repeatedly imposed on the object to be aspirated, which is not preferable.

[0009] Therefore, an object of the present invention is to realize the generation of information related to aspiration by AI technology in a suction device for a living body in order to solve at least one of the above-described problems.

Means for Solving the Problem

[0010] The suction device for a living body of the present invention includes a suction tank that applies a negative pressure in communication with a suction tube inserted into the living body of the object to be aspirated to aspirate an aspirate from the living body, an air suction unit that applies a suction pressure to the suction tank, and a control unit.

[0011] Among these, the control unit is characterized by having the following configuration. The history unit stores the history of the aspiration treatment of the aspirate (hereinafter referred to as "aspiration history"). The learning model unit collects the aspiration history from the history unit and / or externally as learning data, and machine-learns the characteristics of the aspiration history using the learning data.

[0012] The estimation unit generates the information related to aspiration by inputting the aspiration history stored in the history unit into the machine-learned learning model unit.

Advantages of the Invention

[0013] According to the present invention, by machine-learning the characteristics of the aspiration history, it becomes possible to realize the generation of information related to aspiration by AI technology.

[0014] For details of problems, configurations, and effects other than those described above, they will be described in the embodiments described later.

Brief Description of the Drawings

[0015]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Modes for Carrying Out the Invention

[0016] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

Embodiment

[0017] "Explanation of the Configuration of Example 1" Figure 1 is a block diagram illustrating the schematic configuration of Example 1. In Figure 1, the entire system of Example 1 includes a biological suction device 100, an integrated container 200, and a suction tube A.

[0018] Among these, the biological suction device 100 includes a suction tank 110, a discharge unit 111, an air suction unit 120, a control unit 130, a sensor group 140, a suction switch 160, a voice / display unit 170, and a suction dial 180, etc.

[0019] Furthermore, as the air suction unit 120, it includes an air pump 121, a pump drive circuit 121a, an outside air adjustment valve 122, a valve drive circuit 122a, and a pipe 123, etc.

[0020] Also, as the breakdown of the sensor group 140, it includes a pressure detection unit 141, a first sensor 142, a second sensor 143, and a weather sensor 144, etc.

[0021] Note that the functions of the control unit 130 shown in Figure 1 will be described later. Hereinafter, each of the above-described configurations will be specifically explained.

[0022] "Explanation of the Configuration of the Air System" Figure 2 is a diagram illustrating the air system between the air pump 121 and the suction tank 110.

[0023] In Figure 2, the suction tank 110 is an airtight container that communicates with the suction tube A inserted into the living body to be suctioned and applies a negative pressure. Furthermore, the suction tank 110 also serves as a primary container that receives the suction material suctioned by the suction tube A and accumulates it primarily.

[0024] The suction tank 110 communicates via a filter 112 and a pipe 123. When the suction tank 110 is removed from the biological suction device 100 for cleaning, the connection between the suction tank 110 and the pipe 123 can be disconnected. The filter 112 has a function of blocking liquids and allowing gases to pass through.

[0025] The air pump 121 is the main part of the air suction section 120, and applies a negative pressure to the suction tank 110 by drawing the air in the suction tank 110 that communicates via the pipe 123.

[0026] On the path of the pipe 123, an outside air adjustment valve 122 for allowing outside air to flow in and out of the suction tank 110 and a pressure detection section 141 for detecting the suction pressure applied to the suction tank 110 are provided.

[0027] The discharge section 111 includes a valve section 111a and a fastener 111b. The valve section 111a is screwed to the tip of the funnel-shaped bottom surface of the suction tank 110 while being held down by the fastener 111b.

[0028] The discharge section 111 closes the valve section 111a due to the negative pressure in the suction tank 110. As a result, the negative pressure in the suction tank 110 does not escape from the discharge section 111, and the negative pressure acts on the suction tube A. Further, due to the closing of the discharge section 111 (valve section 111a), the suction material in the suction tube A does not pass through the discharge section 111 and is primarily accumulated in the suction tank 110.

[0029] The discharge section 111 opens the valve section 111a when the negative pressure in the suction tank 110 is released. As a result, the suction material primarily accumulated in the suction tank 110 is discharged from the discharge section 111 along the funnel-shaped bottom surface of the suction tank 110. The accumulation container 200 arranged at the discharge destination secondarily accumulates the discharged suction material.

[0030] 《Configuration of the mounting portion of the suction tank 110》 FIG. 3 is a diagram illustrating the mounting portion of the suction tank 110. As shown in FIG. 3, the suction tank 110 is detachably attached to the storage portion 100a of the biological suction device 100 and fixed by the lock portion 100b.

[0031] The bottom surface of the storage portion 100a is formed in a funnel shape to match the bottom surface of the suction tank 110, and a through hole 100c for penetrating the discharge portion 111 into the lower space is provided at the tip of the funnel shape.

[0032] In the storage portion 100a, a first sensor 142 for optically detecting the vertical position of the liquid level of the accumulated aspirated material (such as a float floating in the tank) is provided through the transparent portion of the suction tank 110.

[0033] This first sensor 142 detects the amount of the aspirated material aspirated from the suction tube A by the suction treatment and outputs information to the control unit 130 as the amount of the aspirated material.

[0034] 《Appearance of the Biological Suction Device 100》 FIG. 4 is a diagram illustrating the appearance of the biological suction device 100. In FIG. 4, on the appearance of the biological suction device 100, a suction switch 160, a voice / display unit 170, a suction dial 180, and the like are provided.

[0035] Furthermore, an integration container 200 is detachably arranged in the lower space of the suction tank 110. A second sensor 143 for detecting the weight of the integration container 200 is arranged at the location where the integration container 200 is arranged.

[0036] The control unit 130 acquires the weight of the integration container 200 detected by the second sensor 143 and calculates the incremental change in weight over time. The control unit 130 regards this incremental change in weight as the amount of the aspirated material by the suction treatment.

[0037] In addition, a weather sensor 144 is disposed on the housing surface of the living body suction device 100 so as to be in contact with the outside air through an opening. This weather sensor 144 detects information on the outside air near the device (for example, atmospheric pressure, humidity, temperature, amount of house dust, composition ratio of the outside air, etc.), and outputs the information as weather information to the control unit 130.

[0038] 《Functional Blocks of Control Unit 130》 FIG. 5 is a block diagram illustrating the main functional blocks of the control unit 130. In FIG. 5, the control unit 130 includes main functions such as a suction control unit 131, a history unit 133, a learning model unit 134, an estimation unit 135, and a weather information acquisition unit 136. This control unit 130 performs overall control of the living body suction device 100 and also performs data communication with the outside via the Internet 300.

[0039] Among these, the suction control unit 131 controls each part of the air suction unit 120 to control the suction pressure applied to the suction tank 110.

[0040] The history unit 133 stores the history related to the suction treatment of the suction object as a suction history. For example, it is preferable to include the following information in this suction history in association with the time series. · History of suction date and time (by the time measurement function of the control unit 130, etc.) · History of the amount of suction object (by the measurement function of the first sensor 142, the second sensor 143, etc.) · History of weather information (by the weather sensor 144, acquired information from the outside, etc.) · History of the target situation of the suction target (by acquired information from the outside, the estimation function in the device, etc.) · Others

[0041] The learning model unit 134 collects the suction history from the history unit 133 and / or the outside (such as the communication destination of the Internet 300) as learning data. The learning model unit 134 performs machine learning on the characteristics of the suction history using the accumulated learning data.

[0042] The estimation unit 135 generates information regarding aspiration by inputting the aspiration history and the like (such as the most recent information after machine learning) stored in the history unit 133 into the machine-learned learning model.

[0043] The "information regarding aspiration" generated here is utilized as information contributing to the aspiration procedure by being transmitted to the user of the living body aspiration device 100 and the like via the voice / display unit 170 or a terminal device (such as a smartphone application).

[0044] In addition, the "information regarding aspiration" is utilized as information contributing to understanding the situation of the aspiration target by being communicated and output to an authorized external destination such as a medical institution, a nursing care institution, or a relative. Further, these "information regarding aspiration" may be utilized as learning data for machine learning, etc., by being accumulated on the cloud as big data after excluding information identifying an individual (such as a profile).

[0045] On the other hand, information on the situation to be grasped regarding the aspiration target (target situation) is transmitted to the control unit 130 from a medical institution, a nursing care institution, a relative, the aspiration target himself / herself, a user, and the like. The target situation here preferably includes one or more of the medical condition, diagnosis content, health condition, exercise condition, appetite, sleep situation, activity level, etc. of the aspiration target. The history unit 133 includes the history of this target situation in the aspiration history by associating it with the aspiration history. Such a target situation has a certain degree of correlation with the aspiration history by affecting the physical and mental state of the aspiration target.

[0046] Such a target situation of the aspiration target is also used in the control unit 130 to vary the control of the aspiration (such as aspiration frequency, upper limit value, rising speed, etc.) of the air aspiration unit 120. Further, the "information regarding aspiration (such as the estimation result of the target situation)" estimated by machine learning of the target situation is similarly used in the control unit 130 to vary the control of the air aspiration unit 120.

[0047] The weather information acquisition unit 136 acquires, as weather information, information on the outside air near the apparatus (for example, atmospheric pressure, humidity, temperature, amount of house dust, composition ratio of the outside air, etc.) from the weather sensor 144. Further, the weather information acquisition unit 136 acquires, as weather information, information on the outside air (for example, atmospheric pressure, humidity, temperature, weather, weather forecast, rainfall, weather map, arrangement map of clouds, sunshine amount, temperature difference, various cautionary reports, amount of scattered allergy-related substances, air pollution degree, composition ratio of the atmosphere, etc.) from an external weather-related information site via the Internet 300. The history unit 133 stores the weather information acquired by the weather information acquisition unit 136 in time series in association with the suction history. Such weather information has a certain degree of correlation with the suction history by affecting the physical and mental states of the suction target.

[0048] 《Configuration of the learning model unit 134》 Here, the learning model unit 134 will be further described. Since the learning model unit 134 supports a plurality of estimation processes as described later, it is preferable to include a plurality of individual learning models.

[0049] FIG. 6 is a block diagram illustrating the basic configuration of such a learning model. In FIG. 6, the learning model in the learning model unit 134 includes an input layer A10, a neural network A20, and an output layer A30.

[0050] Among these, the input layer A10 has the number of nodes necessary for inputting sections such as the suction history stored by the history unit 133.

[0051] The neural network A20 is configured by connecting neuron groups by the number of layers suitable for the processing of the input layer A10. Each neuron in the neural network A20 is implemented by arithmetic processing of the connection by the weight coefficient between the nodes and a non-linear activation function. Further, the neural network A20 may have a function of a recurrent neuron (such as RNN or LSTM) that handles neuron groups with a time shift by including a memory group corresponding to human short-term memory to long-term memory.

[0052] The output layer A30 outputs an estimation result (information regarding suction) in accordance with the previously performed machine learning by processing the node outputs of the neural network A20.

[0053] Regarding the specific configuration of each learning model, a plurality of basic configurations (see FIG. 6) may be prepared and combined in a developed manner. For example, a plurality of basic configurations may be provided in series or in parallel. Further, network elements such as branch connections, detours, feedback paths, and confluence paths may be provided in the middle or at connection points of the basic configuration.

[0054] 《Regarding the Biological Suction Program》 Note that part or all of the above-described control unit 130 may be configured as a computer system including a CPU (Central Processing Unit), a memory, etc. as hardware. By executing the "biological suction program" stored in a computer-readable medium by this hardware, part or all of the functions of the control unit 130 (such as the history unit 133, the learning model unit 134, the estimation unit 135, the weather information acquisition unit 136, etc.) are realized.

[0055] Part or all of such hardware may be replaced with a dedicated device, a machine learning device, a DSP (Digital Signal Processor), an FPGA (Field-Programmable Gate Array), a GPU (Graphics Processing Unit), a PLD (programmable logic device), etc.

[0056] Further, the control unit 130 does not have to be built into the biological suction device 100 and may be separated externally. For example, a part or all of the computer system or program of the control unit 130 may be centralized or distributed to a server on the cloud to form a cloud system. In this cloud system, it is possible to aggregate information such as suction histories from a plurality of biological suction devices 100 as big data. In this case, since a large amount of learning data can be collected from a plurality of biological suction devices 100, machine learning can be realized with even higher accuracy. Furthermore, communication services and control services related to the AI technology of the control unit 130 can be provided to a plurality of biological suction devices 100 at any time.

[0057] 《Operation Explanation of Example 1》 Subsequently, as an operation explanation of Example 1, specific machine learning and estimation processing for each individual learning model in the learning model unit 134 will be described in order.

[0058] 《Machine Learning and Estimation Processing Regarding Suction Date and Time》 FIG. 7 is a diagram for explaining machine learning using the history of suction date and time. First, the learning model unit 134 collects the "history of suction date and time" from the history unit 133 or externally, creates a time-series pattern of the suction date and time, and uses it as learning data B10.

[0059] The learning model unit 134 generates learning data B11 that is τ time differences earlier by delaying the learning data B10 by a time difference τ, and sequentially inputs it to the input layer A10. (See FIG. 7)

[0060] The time difference τ (τ≧0) here is preferably set to a value corresponding to the leading time when the learning model unit 134 makes a prediction. When estimating in real time, the time difference τ may be set to zero.

[0061] In the unlearned stage, meaningless estimation results B12 are output in time series to the output layer A30 via the neural network A20. The learning model unit 134 performs machine learning (such as the error backpropagation method) on the neural network A20 so that the time series data of this estimation result B12 approaches the learning data B10.

[0062] As such machine learning progresses, within the neural network A20, specific nodes become sensitive to activation at the periodic timing of the suction date and time leading by a time difference τ. As a result, machine learning of the periodic characteristics of the suction date and time is achieved.

[0063] The estimation unit 135 inputs the history of the suction date and time (the most recent time series data) stored in the history unit 133 into the learned learning model, and predicts the periodic characteristics of the suction treatment leading by a time difference τ. The estimation unit 135 estimates, for example, the following "information regarding suction" according to the predicted periodic characteristics of the suction treatment. · The implementation schedule of the suction treatment after the next time · The predicted time of the next suction treatment · Others

[0064] The control unit 130 notifies the user of this information as information useful for the suction treatment via the voice / display unit 170 or a terminal device (such as a smartphone application).

[0065] 《Machine Learning and Estimation Processing Regarding the Suction Quantity》 FIG. 8 is a diagram for explaining machine learning using the history of the suction quantity. As this "history of the suction quantity", the suction quantities measured for each suction treatment (measurement results by the first sensor 142 and the second sensor 143) arranged on the time axis may be used.

[0066] Note that since the amount of aspirated material for each aspiration procedure is also affected by the time interval from the previous procedure to the current procedure, it does not accurately reflect the generation amount of aspirated material per unit time. Therefore, the history unit 133 may convert it into the generation amount of aspirated material per unit time by equally distributing the amount of aspirated material that increases or decreases due to the influence of the time interval and performing smoothing or interpolation processing. In this case, it becomes possible to use the transition of the generation amount of aspirated material per unit time as the "history of the amount of aspirated material".

[0067] The learning model unit 134 collects such a "history of the amount of aspirated material" from the history unit 133 or externally, creates a time-series pattern of the amount of aspirated material, and uses it as the learning data C10.

[0068] The learning model unit 134 generates the learning data C11 that is τ time steps earlier by delaying the learning data C10 by the time difference τ, and sequentially inputs it to the input layer A10. (See Fig. 8)

[0069] Here, the time difference τ (τ≧0) is preferably set to a value corresponding to the leading time when the learning model unit 134 makes a prediction. When estimating in real time, the time difference τ may be set to zero.

[0070] At the unlearned stage, an insignificant estimation result C12 is output in time series to the output layer A30 via the neural network A20. The learning model unit 134 performs machine learning (such as the error backpropagation method) on the neural network A20 so that the time-series data of this estimation result C12 approaches the learning data C10.

[0071] As such machine learning progresses, within the neural network A20, specific nodes become sensitively activated at the periodic timing of the amount of aspirated material that is τ time steps ahead. As a result, machine learning of the periodic characteristics of the amount of aspirated material is achieved.

[0072] The estimation unit 135 predicts the periodic characteristics of the suction volume with a time difference τ in advance by inputting the "history of suction volume (latest time-series data)" stored in the history unit 133 into the learned learning model. The estimation unit 135 estimates, for example, the following "information regarding suction" according to the predicted periodic characteristics of the suction volume. · The period of the next suction volume (such as the peak time) · Prediction of the next suction volume · Implementation schedule of the suction treatment according to the change in the suction volume · Recommended date and time of the next suction treatment according to the change in the suction volume · Others

[0073] The control unit 130 notifies the user of this information as information useful for "prediction of suction volume" and "implementation of suction treatment" via the voice / display unit 170 or a terminal device (such as a smartphone application).

[0074] 《Machine Learning and Estimation Processing Regarding Weather Information Part 1》 FIG. 9 is a diagram for explaining machine learning with the addition of weather information history. First, the learning model unit 134 collects the "history of weather information" and the "suction history" from the history unit 133 and the outside, creates a time-series pattern (suction history D10a, weather information history D10b) in which both histories are associated with each other in time series, and sets it as learning data D10. The suction history here includes, for example, the history of the suction date and time and / or the suction volume.

[0075] The learning model unit 134 generates previous learning data D11 (suction history D11a, weather information history D11b) with a time difference τ by delaying the learning data D10 by the time difference τ, and sequentially inputs it to the input layer A10 (input layer A10a for suction history, input layer A10b for weather information). (See FIG. 9)

[0076] The time difference τ (τ≧0) here is preferably set to a value corresponding to the leading time when the learning model unit 134 makes a prediction. When estimating in real time, the time difference τ may be set to zero.

[0077] At the unlearned stage, meaningless estimation results D12 are output in time series to the output layer A30 via the neural network A20. The learning model unit 134 performs machine learning (such as the error backpropagation method) on the neural network A20 so that the time series data of this estimation result D12 approaches the suction history D10a in the learning data D10.

[0078] As such machine learning progresses, within the neural network A20, specific nodes become sensitively activated according to the relationship between the weather information and the suction history at a timing that precedes by the time difference τ. As a result, machine learning of the relationship (such as causal relationship or correlation) between the weather information and the suction history is achieved.

[0079] The estimation unit 135 inputs the "weather information and suction history (latest time series data)" stored in the history unit 133 and the machine-learned learning model into the input layers A10a and A10b in time series. As a result, an estimation result D12 of the suction history taking into account the influence of the weather information is output from the output layer A30 with a time difference τ in advance. The estimation unit 135 estimates, for example, the following "information regarding suction" according to the cycle of the suction history predicted in this way. · Next suction cycle taking into account the influence of weather information · Next suction volume taking into account the influence of weather information · Implementation schedule of the suction treatment taking into account the influence of weather information · Recommended date and time of the next suction treatment taking into account the influence of weather information · Others

[0080] The control unit 130 notifies the user of this information as information useful for "prediction of the suction volume" and "implementation of the suction treatment" taking into account the influence of the weather information via the voice / display unit 170 or a terminal device (such as a smartphone application).

[0081] "Machine Learning and Estimation Processing Regarding Weather Information Part 2" FIG. 10 is a diagram for explaining machine learning for estimating "information regarding suction" based on weather information.

[0082] First, the learning model unit 134 collects the "history of weather information" and the "suction history" from the history unit 133 or externally, and creates a time-series pattern (suction history D10a, history of weather information D10b) in which both histories are associated in time series, and sets it as learning data D10. The suction history here includes, for example, the history of the suction date and time and / or the suction amount.

[0083] The learning model unit 134 delays the history D10b of weather information by a time difference τ, thereby generating the history D10b of weather information τ time steps before, and sequentially inputs it to the input layer A10b of the weather information. (See Fig. 10)

[0084] The time difference τ (τ≧0) here is preferably set to a value corresponding to the leading time when the learning model unit 134 makes a prediction. When estimating in real time, the time difference τ may be set to zero.

[0085] At the unlearned stage, an insignificant estimation result D12 is output in time series to the output layer A30 via the neural network A20. The learning model unit 134 performs machine learning (such as the error backpropagation method) on the neural network A20 so that the time-series data of this estimation result D12 approaches the suction history D10a in the learning data D10.

[0086] As such machine learning progresses, within the neural network A20, at the timing τ time steps ahead, specific nodes become sensitively activated according to the relationship between the weather information and the suction history. As a result, machine learning of the relationship (such as a causal relationship or a correlation) between the weather information and the suction history is achieved.

[0087] The estimation unit 135 inputs the history of weather information stored in the history unit 133 in time series to the input layer A10b of the learned learning model, and as a result, an estimation result D12 of the suction history corresponding to the history of weather information is output from the output layer A30 τ time steps ahead. The estimation unit 135 estimates, for example, the next "information regarding suction" according to the period of the suction history predicted in this way. ·Next suction cycle taking into account the influence of meteorological information ·Next suction volume taking into account the influence of meteorological information ·Implementation schedule of the suction treatment taking into account the influence of meteorological information ·Recommended date and time for the next suction treatment taking into account the influence of meteorological information ·Others

[0088] The control unit 130 notifies the user of this information as information useful for "prediction of suction volume" and "implementation of suction treatment" via the voice / display unit 170 or a terminal device (such as a smartphone application).

[0089] 《Machine Learning and Estimation Processing Regarding the Target Situation Part 1》 FIG. 11 is a diagram for explaining machine learning taking into account the situation of the suction target (target situation). First, the learning model unit 134 collects the history unit 133 and the "history of the target situation" and the "suction history" from the outside as a combination, creates a time-series pattern (suction history E10a, history of the target situation E10c) associating both histories in time series, and sets it as learning data E10.

[0090] The learning model unit 134 delays the learning data E10 by a time difference τ to generate learning data E11 (suction history E11a, history of the target situation E11c) τ time before, and sequentially inputs it to the input layer A10 (input layer A10a of the suction history, input layer A10c of the target situation). (See FIG. 11) The time difference τ (τ≧0) here is preferably set to a value corresponding to the leading time when the learning model unit 134 makes a prediction. When estimating in real time, the time difference τ may be set to zero.

[0091] In the unlearned stage, meaningless estimation results E12 (estimation result E12a of the suction history, estimation result E12c of the target situation) are output in time series to the output layer A30 (output layer A30a of the suction history, output layer A30c of the target situation) via the neural network A20. The learning model unit 134 performs machine learning (such as the error backpropagation method) on the neural network A20 so that the estimation result E12a of the suction history approaches the suction history E10a. Further, the learning model unit 134 performs machine learning (such as the error backpropagation method) on the neural network A20 so that the estimation result E12c of the target situation approaches the history E10c of the target situation.

[0092] As such machine learning progresses, within the neural network A20, specific nodes become sensitively activated according to the relationship between the target situation and the suction history at a timing that precedes by a time difference τ. As a result, machine learning of the relationship (such as mutual causal relationship and correlation) between the target situation and the suction history is achieved.

[0093] The estimation unit 135 inputs the suction history stored in the history unit 133 to the input layer A10a of the machine-learned learning model, whereby an estimation result E12c of the target situation that can be estimated from the suction history is output from the output layer A30c with a time difference τ in advance.

[0094] Also, the estimation unit 135 inputs the history of the target situation stored in the history unit 133 to the input layer A10c of the machine-learned learning model, whereby an estimation result E12a of the suction history that can be estimated from the target situation is output from the output layer A30a with a time difference τ in advance.

[0095] Furthermore, the estimation unit 135 inputs the suction history stored in the history unit 133 to the input layer A10a of the machine-learned learning model and inputs the history of the target situation stored in the history unit 133 to the input layer A10c. As a result, from the output layer A30a, an estimation result E12a of the suction history taking both histories into account is output with a time difference τ in advance. Further, from the output layer A30c, an estimation result E12c of the target situation taking both histories into account is output with a time difference τ in advance.

[0096] The estimation unit 135 estimates, for example, the following "information regarding aspiration" according to the predicted target situation and the estimation result of the aspiration history as described above. · The next target situation predicted from the transition of the aspiration history (support information for medical institutions, nursing care facilities, etc.) · The next aspiration cycle predicted from the transition of the target situation · The next aspiration volume predicted from the transition of the target situation · The implementation schedule of the aspiration procedure taking into account the influence of the target situation · The recommended date and time of the next aspiration procedure taking into account the influence of the target situation · Others

[0097] The control unit 130 notifies the user of this information as information useful for "support information regarding the target situation", "prediction of the aspiration volume", "implementation of the aspiration procedure", etc. via the voice / display unit 170 or a terminal device (such as a smartphone application).

[0098] Machine Learning and Estimation Processing Regarding the Target Situation Part 2 FIG. 12 is a diagram for explaining machine learning for estimating the situation of the aspiration target (target situation) based on the aspiration history.

[0099] First, the learning model unit 134 collects the "history of the target situation" and the "aspiration history" from the history unit 133 and externally as a combination, and creates a time-series pattern (aspiration history E10a, history of the target situation E10c) associating both histories in time series, and sets it as learning data E10.

[0100] The learning model unit 134 generates the aspiration history E11a τ time steps before by delaying the learning data E10 by the time difference τ, and sequentially inputs it to the input layer A10a of the aspiration history. (See FIG. 12) The time difference τ (τ≧0) here is preferably set to a value corresponding to the leading time when the learning model unit 134 makes a prediction. When estimating in real time, the time difference τ may be set to zero.

[0101] At the unlearned stage, meaningless estimation results E12 (estimation results E12c of the target situation) are output in time series to the output layer A30c via the neural network A20. The learning model unit 134 performs machine learning (such as the error backpropagation method) on the neural network A20 so that the estimation result E12c of the target situation approaches the history E10c of the target situation.

[0102] As such machine learning progresses, within the neural network A20, specific nodes become sensitively activated according to the relationship between the target situation and the attraction history at a timing that precedes by a time difference τ. As a result, machine learning of the relationship (such as mutual causal relationship and correlation) between the target situation and the attraction history is achieved.

[0103] The estimation unit 135 inputs the attraction history stored in the history unit 133 to the input layer A10a of the machine-learned learning model, whereby the estimation result E12c of the target situation that can be estimated from the attraction history is output from the output layer A30c with a time difference τ ahead.

[0104] The estimation unit 135 estimates, for example, the following 'information regarding suction' according to the estimation result of the target situation predicted in this way. · The next target situation expected from the transition of the attraction history (support information for medical institutions, nursing care institutions, etc.) · The implementation schedule of the suction treatment according to the target situation · The recommended date and time of the next suction treatment according to the target situation · Others

[0105] The control unit 130 notifies the user of this information as information contributing to'support information regarding the target situation' or the like via the voice / display unit 170 or a terminal device (such as a smartphone app).

[0106] 《Effect of Example 1》 Example 1 has the following effects due to the above-described configuration and operation.

[0107] (1) In Example 1, the learning model unit 134 machine-learns the characteristics of the history of aspiration treatment (aspiration history). By inputting the aspiration history stored in the history unit 133 into this machine-learned learning model unit 134, it becomes possible to generate information regarding aspiration. Therefore, Example 1 is excellent in that it can generate information regarding aspiration using AI technology in the living body aspiration device 100.

[0108] (2) In Example 1, the learning model unit 134 machine-learns the periodic characteristics of the aspiration date and time. By inputting the history of the aspiration date and time stored in the history unit 133 into this machine-learned learning model unit 134, the cycle of the aspiration treatment is predicted. Therefore, Example 1 is excellent in that it can predict the date and time of the next and subsequent aspiration treatments using AI technology in the living body aspiration device 100.

[0109] (3) In Example 1, the learning model unit 134 machine-learns the periodic characteristics of the aspiration volume. By inputting the history of the aspiration volume stored in the history unit 133 into this machine-learned learning model unit 134, the cycle of the aspiration volume is predicted. Therefore, Example 1 is excellent in that it can predict the aspiration volume of the next and subsequent times using AI technology in the living body aspiration device 100.

[0110] (4) In Example 1, the learning model unit 134 machine-learns the relationship between the weather information and the aspiration history. By inputting the weather information and the like into this machine-learned learning model unit 134, the timing of the aspiration treatment taking into account the influence of the weather information is predicted. Therefore, Example 1 is excellent in that it can further improve the prediction accuracy of the aspiration treatment based on the influence of the weather information.

[0111] (5) In Example 1, the learning model unit 134 machine-learns the relationship between the aspiration history and the target situation. By inputting at least the aspiration history into this machine-learned learning model unit 134, an estimation result of the target situation corresponding to the aspiration history is generated as support information. Therefore, Example 1 is excellent in that it can generate the situation of the aspiration target as support information using AI technology in the living body aspiration device 100.

[0112] (6) In particular, in Example 1, when the characteristics of the aspiration history of each individual to be aspirated are machine-learned, it becomes possible to generate information useful for the aspiration treatment in accordance with the following individual characteristics of each person. · Those who need aspiration operations during specific time periods such as waking up or in the evening · Those who need regular aspiration operations during sleep · Those who have a high frequency of aspiration operations during specific periods such as hay fever · Those whose frequency of aspiration operations changes seasonally or monthly · Those whose frequency of aspiration operations changes with the passage of time and date · Others

[0113] 《Other Supplementary Matters》 Note that in the above-described embodiments, the case where the suction tube A is inserted from the oral cavity or nasal cavity has been described by way of assumption. However, the present invention is not limited thereto, and it is sufficient that the suction tube A is inserted into the living body to aspirate the inside of the living body. Further, for example, the suction tube A may be inserted into the living body in combination with a tracheostomy cannula, a ventilator, an endoscope, a laparoscope, or the like.

[0114] Furthermore, in the above-described embodiments, the description has been made by assuming the electric air pump 121. However, the air pump 121 of the present invention only needs to be a means for aspirating at least air, and is not limited to the electric type. For example, various air pump mechanisms such as a manual type, a chemical reaction type, and a mouth aspiration type may be adopted.

[0115] Also, in the above-described embodiments, the description has been made by assuming that all of the control unit 130 is arranged inside the living body aspiration device 100. However, the present invention is not limited thereto. For example, part or all of the control unit 130 may be arranged on the cloud or the terminal device, and machine learning, estimation processing, or the like may be performed on the cloud or the terminal device side.

[0116] Furthermore, in the above-described embodiments, for the purpose of clarifying the problems and solutions of individual data, the description has been made assuming that machine learning and estimation processing of the learning model are performed separately for each data. However, the present invention is not limited to this. For example, two or more of these data (such as suction date and time, suction volume, weather information, target situation, etc.) may be combined, and machine learning, estimation processing, etc. may be performed on the decoded learning model.

[0117] Note that the present invention is not limited to the content of the above-described embodiments, and various modifications are possible.

[0118] For example, the above-described embodiments have been described in detail in their entirety for the purpose of easily explaining the present invention, and the present invention is not necessarily limited to having all the configurations and all the steps described.

[0119] Also, the present invention is not limited to the types of individual components (such as parts like sensors). For example, each individual component may be changed to another type of component having an equivalent function.

[0120] Also, the individual elements of the embodiments may be partially combined. Furthermore, it is also possible to add or replace other configurations and other steps to the embodiments. Also, some configurations and some steps may be deleted from the embodiments.

Description of Reference Numerals

[0121] 100...Biological suction device, 100a...Storage part, 100b...Lock part, 100c...Through hole, 110...Suction tank, 111...Discharge part, 111a...Valve part, 111b...Fastener, 112...Filter, 120...Air suction part, 121...Air pump, 121a...Pump drive circuit, 122...Outside air adjustment valve, 122a...Valve drive circuit, 123...Pipe, 130...Control part, 131...Suction control part, 133...History part, 134...Learning model part, 135...Estimation part, 136...Weather information acquisition part, 140...Sensor group, 141...Pressure detection part, 142...First sensor, 143...Second sensor, 144...Weather sensor, 160...Suction switch, 170...Voice / display part, 180...Suction dial, 200...Collection container, 300...Internet, A...Suction tube, A10...Input layer, A10a...Input layer of suction history, A10b...Input layer of weather information, A10c...Input layer of target situation, A20...Neural network, A30...Output layer, A30a...Output layer of suction history, A30c...Output layer of target situation, B10...Learning data, B11...Learning data, B12...Estimation result, C10...Learning data, C11...Learning data, C12...Estimation result, D10...Learning data, D10a...Suction history, D10b...History of weather information, D11...Learning data, D11a...Suction history, D11b...History of weather information, D12...Estimation result, E10...Learning data, E10a...Suction history, E10c...History of target situation, E11...Learning data, E11a...Suction history, E11c...History of target situation, E12...Estimation result, E12a...Estimation result of suction history, E12c...Estimation result of target situation

Claims

1. A suction tank that applies a negative pressure to communicate with a suction tube inserted into a living body of an object to be attracted and sucks a suction material from the living body, an air suction unit that applies a suction pressure to the suction tank, and a control unit, wherein the control unit comprises a history unit that acquires and stores a history of the suction treatment of the suction material (hereinafter referred to as "suction history"), a learning model unit that collects the suction history from the history unit and / or externally as learning data and machine-learns the characteristics of the suction history using the learning data, and an estimation unit that generates information related to the suction by inputting the suction history stored in the history unit into the machine-learned learning model unit. A suction device for a living body, characterized by the above.

2. The suction device for a living body according to Claim 1, wherein the history unit acquires and stores at least a history of the suction date and time as the suction history, the learning model unit collects at least "the history of the suction date and time" from the history unit and / or externally as the learning data, machine-learns the periodic characteristics of the suction date and time, and the estimation unit inputs "the history of the suction date and time" stored in the history unit into the machine-learned learning model unit to predict information according to the timing of the suction treatment. A suction device for a living body, characterized by the above.

3. The suction device for a living body according to Claim 1, comprising a sensor that detects the amount of the suction material (hereinafter referred to as "suction material amount") by the suction treatment, wherein the history unit acquires and stores at least a history of the suction material amount as the suction history, the learning model unit collects at least "the history of the suction material amount" from the history unit and / or externally as the learning data, machine-learns the periodic characteristics of the suction material amount, and the estimation unit inputs "the history of the suction material amount" stored in the history unit into the machine-learned learning model unit to predict information according to the suction material amount. A suction device for a living body, characterized by the above.

4. The suction device for a living body according to Claim 1, wherein the control unit comprises a weather information acquisition unit that acquires weather information, the history unit stores the acquired weather information in association with the suction history, and the learning model unit collects the weather information and the suction history from the history unit and / or externally as the learning data and machine-learns the relationship between the weather information and the suction history. The estimation unit inputs at least the weather information acquired by the weather information acquisition unit into the learned learning model unit, and predicts information according to the suction history. A suction device for a living body, characterized in that.

5. The suction device for a living body according to claim 1, The learning model unit collects at least a combination of the suction history and the situation of the suction target (hereinafter referred to as "target situation") from the history unit and / or externally as the learning data, and learns the relationship between the suction history and the target situation. The estimation unit inputs at least the suction history stored in the history unit into the learned learning model unit, and generates an estimation result of the target situation as support information. A suction device for a living body, characterized in that.

6. Causing a computer system to function as the control unit according to any one of claims 1 to 5 A suction program for a living body, characterized in that.

Images