JP5867847B2 - Patient behavior identification method and patient behavior detection system - Google Patents

Description

  The present invention relates to a patient behavior identification method and a patient behavior detection system for remotely monitoring a patient's behavior and recognizing a behavior pattern to prevent an accident.

  In recent years, the number of patients who undergo rehabilitation at home during the recovery period after hospitalization has increased due to aging, a lack of treatment beds, and the desires of the person. In order to shift to home medical treatment, guidance by doctors, nurses, and caregivers until they can perform daily activities such as eating, excretion, preparation, movement, and bathing independently during hospitalization. After being judged that there is no risk of accidents, the shift to home care is permitted.

  However, there are an increasing number of cases where patients who continue to be treated at home are hospitalized again due to accidents such as falls or falls. These patients may not be able to reach the eyes of nurses or caregivers for patients who have shifted to home care, or may behave beyond hospital behavior restrictions due to patient negligence or overconfidence with the patient. Is the cause.

  On the other hand, a system as disclosed in Patent Document 1 has been proposed as a system for constantly monitoring the condition of an inpatient in a hospital or the like. This patient monitoring system includes a biological information measuring device for measuring the biological information of a patient, and the contents of the abnormality when there are abnormalities in the biological information carried by each of a plurality of nurses and measured by the biological information measuring device. A patient monitoring system including a plurality of portable terminal devices that output abnormality information representing the position of the plurality of portable terminal devices, and an alarm is output from the biological information measuring device. In this case, the patient includes abnormality information transmission control means for transmitting abnormality information to the portable terminal device closest to the biological information measuring device based on the positions of the plurality of portable terminal devices detected by the position detecting means. It is a monitoring system.

  In addition, as disclosed in Patent Document 2, a seating sensor is provided on the wheelchair to detect that the user has left the wheelchair and to sound an alarm buzzer, etc. to prevent a fall accident caused by leaving the wheelchair. Devices have also been proposed.

JP 2004-65471 A JP 2005-81053 A

  The patient monitoring system disclosed in Patent Document 1 is a monitoring system in a hospital and cannot be applied to prevent accidents at home patients. Also, as disclosed in Patent Document 2, when a sensor is provided on a device used by a patient, it detects when the patient leaves the device, and an alarm is issued if an accident cannot occur. As a result, unforeseen circumstances will occur.

  The present invention has been made in view of the above-mentioned background art. When a medical person sees the behavior of the patient who has detected and accumulated the behavior pattern of the patient afterwards, and the behavior includes dangerous behavior. The purpose is to provide a patient behavior identification method and a patient behavior detection system that allow the system to learn that the behavior corresponds to the dangerous behavior and to instruct the patient not to take the dangerous behavior in the future. .

  Furthermore, the present invention uses a system that learns the risk behavior of each patient during the patient's convalescent rehabilitation, and detects the patient's risk behavior after going home, and issues a warning when the patient takes the risk behavior. It is an object of the present invention to provide a patient behavior identification method and a patient behavior detection system capable of stopping the dangerous behavior.

In this invention, various sensors for detecting force and position are provided in a plurality of devices and members used by a patient, and sensor data from the sensors obtained by the various operations and actions of the patient are accumulated in advance for each individual patient. Based on the accumulated sensor data , a motion model is created and stored for each individual patient, and the sensor obtained from the motion of the individual patient during normal life based on the motion model. This is a patient behavior identification method for identifying data and determining its operating state.

The plurality of devices are at least a wheelchair and a bed used by the individual patient, and the sensor detects a load applied to the wheelchair or the bed and position information of each member, and the wheelchair and the bed Are provided with a plurality of sensors for detecting information, respectively, first acquiring data from each sensor, performing operation labeling of the acquired operation data, extracting features of the patient's operation, and dangerous operation The action classification information for discriminating whether or not each of the individual patients is stored on the basis of the label obtained by the action labeling and the extracted feature amount data of the feature. And various motion models for each individual patient are set, and based on the motion model, sensor data from each sensor for the motion of normal life for each individual patient. The patient motion is identified, and the motion of the identified sensor data is compared with the behavior rule recorded in the behavior rule storage unit to determine whether it is dangerous, and the motion in the wheelchair and the bed for each patient is determined. perform dangerous whether behavior determining process, when the wheelchair with a reference value or more values determined from either a dangerous state of the bed is obtained, notified its operating status to the patient and surrounding, And / or the operation state is accumulated and can be referred to later. The identification of the movement of the patient uses, for example, SVM (Support Vector Machine).

The present invention also provides a plurality of sensors attached to a plurality of devices and members used by a patient to detect forces and positions, and sensor data from the sensors obtained in advance by various operations and actions of the patient for each individual patient. And a motion model is created by classifying the movement of each individual patient based on the sensor data, the movement of each individual patient is identified based on the movement model, and the movement state It is a patient action detection system provided with the processing apparatus which performs the process which transmits.

The plurality of devices are at least a wheelchair and a bed used by the individual patient, and the sensors are a load sensor provided on the wheelchair, a load sensor and a force sensor provided on the bed, and the processing The apparatus acquires data from each sensor, performs operation labeling of the acquired operation data, extracts the characteristics of the patient's movement, and classifies movement classification information for determining whether the movement is dangerous or not. Based on the accumulated and obtained labels and the feature amount data of the extracted features, the various patterns of motion for each individual patient are classified, and various motion models for each individual patient are set. Based on the behavior model, the behavior of normal life for each individual patient is identified with the patient behavior of the sensor data from each sensor, and the behavior of the identified sensor data is the behavior rule storage unit Perform an action determination processing of determining whether dangerous or not by collating the recorded behavioral rules, to identify processing a plurality of the sensor data from one of the bed and the wheelchair, one of the said wheel chair bed When a value equal to or higher than a reference value determined to be a dangerous state is obtained , a process of notifying the patient and the surroundings of the operation state is performed.

  The plurality of devices include wireless communication transceivers, and exchange the sensor data with the processing device by wireless communication.

  The patient behavior identification method and patient behavior detection system according to the present invention grasps and accumulates the patient's behavior patterns in advance using various sensors, creates an operation model of sensor data that appears during the dangerous behavior of the patient, and uses this as a basis. For example, during home medical treatment, if the patient behavior is monitored and data that may cause dangerous behavior is obtained, an alarm can be sounded to warn the patient and the surrounding people. As a result, it is possible to prevent unexpected situations and accidents due to dangerous behavior during home medical treatment. In particular, accurate determination is possible by using SVM for classification of behavior patterns.

It is a conceptual diagram which shows the patient action detection system of one Embodiment of this invention. It is a schematic perspective view which shows the wheelchair used for the patient action detection system of this embodiment. It is a partial fracture front view of the horizontal direction of the footrest part of the wheelchair of this embodiment. It is the longitudinal cross-sectional view (a) of the left-right direction of the seat surface part of the wheelchair of the patient action detection system of this embodiment, and the partially broken side view (b) of the seat surface side part. It is the longitudinal cross-sectional view (a) of the left-right direction of the armrest part of the wheelchair of this embodiment, and the partially broken side view (b) of an armrest part. It is the schematic plan view (a) and schematic front view (b) of the bed of this embodiment. It is a schematic front view of the handrail of the bed of this embodiment. It is a perspective view which shows the wall handrail of this embodiment. It is a circuit diagram of the strain gauge sensor used for the patient action detection system of this embodiment. It is a functional block diagram of the patient action detection system of this embodiment. It is the flowchart (a) which shows the preliminary process of the patient action detection system of this embodiment, and the flowchart (b) which shows the process at the time of operation | movement.

  Hereinafter, a patient behavior detection system 10 according to an embodiment of the present invention will be described with reference to FIGS. As shown in FIG. 1, the patient behavior detection system 10 is provided with sensors for detecting a use state of each device by a patient on a handrail 18 attached to a wheelchair 12, a bed 14, and a wall of a toilet 16. Yes.

  First, in the wheelchair 12, as shown in FIGS. 2 to 5, a pressure distribution sensor 24 using a piezoelectric element or the like is disposed in the backrest portion 22a of the backrest 22 so that the patient's leaning state against the backrest portion 22a can be detected. Is provided. An optical sensor 32 such as a photo interrupter is attached to the front vertical frame 30 a of the vehicle body frame 30 of the wheelchair 12 in the vicinity of the brake lever 28 adjacent to the wheel 26 of the wheelchair 12. Thereby, it is possible to detect the wheel locked state and the unlocked state by the swinging operation of the brake lever 28. The footrest part 34 of the wheelchair 12 is also provided with an optical sensor 32 such as a photo interrupter attached to a footrest frame 30 b extending forward from the vehicle body frame 30. Thereby, the use state in which the footrest plate 34a is leveled and the non-use state in which the footrest plate 34a is flipped up can be detected. The optical sensor 32 detects a binary state with or without a lever or the like.

  Furthermore, the wheelchair 12 is provided with a load sensor 20 including a piezoelectric element and other pressure sensors for detecting how to apply a load in a use state of a patient who is a user. As shown in FIG. 3, the footrest plate 34a of the footrest portion 34 is provided with a load sensor 20 inside the footrest plate 34a, and can detect the load when the patient puts the foot on the footrest plate 34a. It is said.

  As shown in FIG. 4, the seat surface portion 38 of the wheelchair 12 is provided with a seat surface support structure 40 fixed to the vehicle body frame 30, and supports the seat surface fabric 38 a on the surface side of the seat surface 38. A pair of load sensors 20 are provided between both ends of the holding member 40a and the seat support structure 40 to detect the patient's sitting and to detect the load value.

  As shown in FIG. 5, the pair of load sensors 20 is provided between the mounting plate 42 b on the back surface of the armrest plate 42 a and the mounting plate 42 c on the vehicle body frame 30 side. Is attached. A holding member 44 for preventing displacement is inserted between the lower surface of the armrest plate 42 a and the upper surface of the vehicle body frame 30.

  As shown in FIGS. 6 and 7, the bed 14 used in the patient behavior detection system 10 of this embodiment has load sensors 20 attached to the four corners of the back surface of the two upper surface plates 52 placed on the upper surface of the bed 14. The state where the patient is on the bed 14 can be detected. On the handrail 54 of the bed 14, the strain gauge sensor 50 is attached to the upper surface of the horizontal member 54a, and the strain gauge sensor 50 is also attached to the two directions around the vertical member 54b. Thereby, the force applied to the bed 14 and the handrail 54 can be detected.

  Furthermore, in this embodiment, as shown in FIG. 8, the strain gauge sensor 50 is also wound around the attachment member 18a of the handrail 18 attached to the wall of the toilet 16, and the force applied to the handrail 18 and the force applied thereto. The direction can be detected. The strain gauge sensor 50 wound around the attachment member 18a has three strain gauge sensors 50 attached to one attachment member 18a, and can detect force in the vertical and horizontal directions.

  As shown in FIG. 1, the patient behavior detection system 10 of this embodiment is a personal computer (processing device for various programs and a recording device) that uses force values and member positions detected by each sensor by wireless communication. (Hereinafter referred to as “PC”). The PC 60 can be placed on the wheelchair 12 with a size about the size of a notebook computer, for example. For communication, the existing short-range wireless communication standard Zigbee or Bluetooth can be used. Further, data may be exchanged by wire via a USB terminal. For example, a communication adapter is connected to the PC 60 provided in the wheelchair 12, and a transceiver 62 is installed in the bed 14 and the toilet 16.

  The output of each load sensor 20 is a value proportional to the load via an amplifier (not shown), and digital data indicating the value of the load is output from the transceiver 62 via an A / D converter (not shown) to the PC 60. Then, predetermined processing described later is performed by the processing program, and the load data is recorded in the storage device 64.

  Similarly, as the output of the strain gauge sensor 50, a load value is output via the sensor circuit 56 shown in FIG. The sensor circuit 56 includes a bridge circuit 56a in which the strain gauge sensor 50 is arranged in one of the bridges, and a pair of outputs of the bridge circuit 56a is input to the amplifier 56b, and the output of the amplifier 56b is converted into A / D conversion and a load. The data is output to a data logger 58 that records data. The load data recorded in the data logger 58 is recorded in the storage device 64 in the PC 60 via the transceiver 62.

  As shown in FIGS. 1, 10, and 11, the processing in the PC 60 is performed by transmitting sensor data acquired by the sensors provided in the wheelchair 12, the bed 14, and the toilet 16 to the PC 60, according to a predetermined program. The sensor data is aggregated, and the operation identification processing of the operation is performed on the aggregated sensor data.

  Prior to the motion identification process, the data from each sensor is aggregated to create a motion model that serves as a reference for classification of the behavior pattern of the patient. As shown in FIG. 11A, the motion model is created by extracting the features of the patient's motion, accumulating motion classification information for determining whether the motion is dangerous, and storing the motion model information. It is created and recorded in the behavior model storage unit 64a of the storage device 64. Here, SVM (Support Vector Machine) is used for the sensor data classification process. SVM is one of identification methods using supervised learning, and is a learning model characterized by exhibiting excellent performance for nonlinear classification problems. Therefore, classification processing is performed using SVM for the operation classification of this embodiment.

  First, as shown in FIG. 11A, when sensor data is acquired (s1), operation labeling (s2) and feature amount extraction processing (s3) of the acquired operation data are performed. Thereafter, based on the obtained label and feature amount data, an operation model based on SVM classification is created using SVM (s4). Thereby, various patterns of the patient's motion are classified and various motion models are set. This motion model is later used, for example, to identify patient motion.

  After the SVM model is created, as shown in FIG. 10, an action identification process is performed for the movement of the patient, and then an action determination process for determining whether the patient's action is dangerous is performed. The detected value of each operation based on the obtained sensor data is recorded in the operation log storage unit 64b in the storage device 64 together with the operation time. As shown in FIG. 11B, in the motion identification process for the patient motion, sensor data is acquired (s11), and a feature amount extraction process for the sensor data is performed (s12). The extracted feature data is subjected to SVM classification processing (s13). In this SVM classification process, the previously created SVM model is called from the behavior model storage unit 64a and used. Thereby, the operation | movement identification by the patient's sensor data is made.

  The sensor data that has been subjected to the action identification process is subjected to an action determination process to determine whether it is a dangerous action. For the determination, an action rule recorded in the action rule storage unit 64c in the storage device 64 is applied. That is, when it is determined that the detected sensor data is dangerous in light of the action rules in the action rule storage unit 64c, an alarm signal is transmitted from the alarm generation unit 66 and the location of the operation is performed. An alarm (not shown) sounds and the patient and the people around him can recognize that the operation is dangerous.

  As described above, the patient behavior identification method and the patient behavior detection system 10 according to this embodiment grasps a patient's behavior pattern and uses the sensor data from each sensor obtained in advance based on the behavior of the patient to determine the patient's behavior. Recognize the value of sensor data that appears at the time of dangerous behavior, and based on this, monitor the above patient behavior during home medical treatment, for example, and if data that can cause dangerous behavior is obtained, an alarm is issued. It can be used to alert the patient and those around him. As a result, accidents due to dangerous behavior such as home medical treatment can be prevented. In particular, by using SVM for model creation and classification of sensor data obtained at the time of action, accurate classification is possible, and appropriate determination is possible.

  The patient behavior identification method and the patient behavior detection system according to the present invention are not limited to the above-described embodiment, and the installation location of the sensor can be selected as appropriate. The sensor type may be selected as appropriate.

DESCRIPTION OF SYMBOLS 10 Patient action detection system 12 Wheelchair 14 Bed 16 Toilet 18 Handrail 20 Load sensor 22 Backrest 24 Pressure distribution sensor 30 Body frame 32 Optical sensor 34 Footrest part 38 Seat surface part 50 Strain gauge sensor 52 Armrest part 60 PC
62 Transceiver 64 Storage device

Claims (4)

Various sensors and sensors that detect the force and position are provided on a plurality of devices and members used by patients, and sensor data from the sensors obtained by the various operations and actions of the patients are stored in advance for each individual patient. Based on the sensor data , a motion model is created and stored for each individual patient, and the sensor data obtained from the motion of the individual patient during normal life is identified based on the motion model. , A patient behavior identification method for determining the operation state ,
The plurality of devices are at least a wheelchair and a bed used by the individual patient, and the sensor detects a load applied to the wheelchair or the bed and position information of each member. A plurality of sensors each for detecting information;
First, data from each of the sensors is acquired, operation labeling of the acquired operation data is performed, characteristics of the operation of the patient are extracted, and operation classification information for determining whether or not the operation is dangerous is stored. And
After that, based on the label obtained by the motion labeling and the extracted feature amount data of the feature, various motion patterns for each individual patient are classified, and various motion models for each individual patient are obtained. Set,
Based on the behavior model, the behavior of the normal data for each individual patient is identified with the patient behavior of the sensor data from each sensor, and the behavior of the identified sensor data is recorded in the behavior rule storage unit. It is determined whether or not it is dangerous by collating with an action rule, and action determination processing is performed as to whether or not the operation on the wheelchair and the bed for each patient is dangerous,
A patient behavior identification method characterized by notifying the patient and the surroundings of an operation state when a value equal to or higher than a reference value determined as a dangerous state is obtained from either the wheelchair or the bed .   The patient behavior identification method according to claim 1, wherein the patient's movement is identified using an SVM (Support Vector Machine). A plurality of sensors attached to a plurality of devices and members used by a patient to detect forces and positions, and a storage device that stores sensor data from the sensors obtained in advance by various operations and actions of the patient for each individual patient When the processing of the sensor data said classifying the operation of the individual patient to patient based on creating an operation model, the identifying operation of the individual for each patient on the basis of this operation model, and sends the operating state A patient behavior detection system comprising a processing device to perform,
The plurality of devices are at least a wheelchair and a bed used by the individual patient, and the sensors are a load sensor provided in the wheelchair, a load sensor and a force sensor provided in the bed,
The processing device acquires data from each sensor, performs operation labeling of the acquired operation data, extracts the characteristics of the patient's operation, and classifies the operation to determine whether it is a dangerous operation Based on the obtained label and extracted feature value data of the feature, information is accumulated, and various patterns of motion for each individual patient are classified, and various motion models for each individual patient are set. Based on this behavior model, the patient behavior of the sensor data from each sensor is identified for the behavior of normal life for each individual patient, and the behavior of the identified sensor data is recorded in the behavior rule storage unit. Action discriminating process for judging whether or not it is dangerous by comparing with the action rule, discriminating a plurality of the sensor data from either the wheelchair or the bed, and either the wheelchair or the bed When the reference value or more values is determined to al dangerous condition is obtained, the patient behavior detection system characterized by performing a process of notifying the operation state to the patient and the surroundings. The patient behavior detection system according to claim 3, wherein the plurality of devices includes a transceiver for wireless communication, and exchanges the sensor data with the processing device by wireless communication.

Images