JP2005235075A - System for supporting replacement of parts of medical inspection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To predict a replacement time of parts of medical inspection devices, with high accuracy. <P>SOLUTION: A system for supporting replacement of parts of medical inspection devices, which collects information of a medical inspection device 21 in a medical institution 20 through electric communication lines 22 and 23, is provided with; a measurement value request component 11 which receives information relating to a use condition from the medical inspection device; a storage component 12 for storing received information relating to the use condition; a prediction component 15 which predicts a future use condition based on a prescribed-period use condition on the basis of the stored use condition and predicts a more future use condition on the basis of the predicted future use condition and predicts a replacement time of parts of the medical inspection device on the basis of the stored use condition and predicted use conditions in future; and a information component 16 which transmits information relating to the predicted replacement time to a service engineer system. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a medical test equipment part replacement support system that supports parts replacement work of medical test equipment.

  Parts such as CT X-ray tubes and circuits deteriorate due to inspection and usage environment. Replacing parts requires special equipment, and it takes time to arrange and replace parts. Therefore, replacement after a failure occurs only reduces the reliability and operating rate of the inspection equipment. However, the burden on the medical institution such as interrupting the inspection or changing the inspection plan thereafter is increased. Therefore, there has been a demand for grasping and predicting the deterioration state of parts so that advance replacement can be planned.

The deterioration status of parts can be grasped by the number of times of shooting and the usage status such as voltage. For example, the number of shots is automatically measured inside the product, acquired by a service engineer visiting a place where an inspection device such as a hospital or inspection center is installed, or sent to the maintenance center via the network as remote maintenance. ing.

  For example, the deterioration is determined by comparing the number of times of photographing with a reference value. Conventionally, the deterioration is compared inside the inspection apparatus or at the maintenance center. In addition, it has been performed to express the fluctuation of the number of shootings by expressing it with some function.

  Japanese Patent Laid-Open No. 2001-290891 discloses that an inspection device and a customer support device are connected via the Internet, information on operation is reported from the inspection device to the customer support device by e-mail, and the customer support device creates a database. It is described. It also describes that the life of the X-ray tube is predicted from this database and notified to the inspection equipment.

However, Japanese Patent Application Laid-Open No. 2001-290891 shows that the part life time is predicted from the past number of times of photographing, the accumulated operation time, etc., but the important prediction method is not described and the problem of the prediction accuracy is not described. Is not solved. Further, although the inspection device is notified, the maintenance work is a work of a maintenance center engineer or a service engineer, and it is necessary to notify these engineers, but the means is not shown. Information on operation includes information on hospital facilities and operations, but there is a possibility of leakage or tampering with asynchronous transmission means such as e-mail.
JP 2001-290891 A

  An object of the present invention is to provide a medical test equipment part replacement support system that can predict the replacement time of parts of a medical test equipment with high accuracy.

  An aspect of the present invention is a medical test equipment part replacement support system that collects information on medical test equipment of a medical institution via an electrical communication line, and a receiving unit that receives information on the usage status from the medical test equipment; A storage unit that stores information about the received usage status; a future usage status is predicted based on the usage status for a predetermined period based on the stored usage status; and further based on the predicted future usage status Predicting the future usage status, predicting the parts replacement time of the medical examination equipment based on the stored usage status and the predicted usage status at the future time, and information on the predicted replacement time And a transmission unit for transmitting to the service engineer system.

  ADVANTAGE OF THE INVENTION According to this invention, the replacement time of the components of a medical test device can be estimated with high accuracy.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
As shown in FIG. 1, a medical test equipment part replacement support system 10 according to the present embodiment includes a plurality of medical test equipment 21 installed in a plurality of medical institutions 20 and an electrical communication line 22 as a dedicated line. Connected. The medical examination equipment 21 includes an X-ray diagnostic apparatus, an ultrasonic diagnostic apparatus, an X-ray computed tomography apparatus (CT scanner), a magnetic resonance imaging apparatus (MRI), a gamma camera, and the like. Further, the medical test equipment part replacement support system 10 is connected to a service engineer system 31 installed in a maintenance center 30 provided for each area in charge through an electrical communication line 32 as a dedicated line.

  The medical test equipment part replacement support system 10 collects, for example, the past operation status per day as a unit period of the medical test equipment 21 at a frequency of once a day for each medical test equipment 21 and operates over the past several days. Predict the tomorrow's operating status as a near future time from the situation, and repeat the prediction process one after another using the predicted usage status of the next day to predict the operating status at the future time of several days or tens of days ahead In addition, it is configured to predict the replacement time for each part by comparing the cumulative operating status up to the future time point with the service life determined for each part in advance.

  That is, the medical test equipment component replacement support system 10 includes a measurement value request component 11, a measurement value storage component 12, a past pattern learning component 13, a past pattern storage component 14, a prediction component 15, and a notification component 16.

  The measurement value requesting component 11 indicates the usage status such as the number of times of imaging and tube voltage self-measured by each medical examination device 21 at a predetermined frequency, here once a day, for a plurality of medical examination devices 21. In addition to requesting transmission of measured value information (hereinafter referred to as usage status information), the usage status information transmitted from the medical examination device 21 in response to the request is received under, for example, a file transfer protocol (FTP). The measured value storage component 12 stores the usage status information in association with the identification number of the medical examination device 21.

  In the following, for convenience of explanation, “measurement number” will be described as an example of measurement value information representing the usage status, but the measurement value targeted in this embodiment is not limited to the number of acquisitions, and is not limited to a circuit. It may be other matters such as the voltage of the X-ray tube, the temperature and humidity of the apparatus installation location, the temperature and humidity inside the apparatus.

  The prediction component 15 predicts the usage status of the next day from the usage status for several days, and repeats the prediction process one after another using the predicted usage status of the next day, thereby making the future several days or tens of days ahead. In order to predict the operation status at the time and to predict the replacement time for each part by comparing the cumulative operating status up to the future time with the service life determined for each part in advance, here the input layer, It is composed of a neural network consisting of three layers, a hidden layer and an output layer.

  The past pattern learning component 13 predicts the usage status of the next day from the usage status for several days, and inputs the prediction component 15 so that the predicted usage status of the next day matches or approximates the actual usage status. The connection strength between the layer node (neuron) and the hidden layer node (weight of weighted addition) and the weight between the hidden layer node and the output layer node are changed (learning). Here, these weights are collectively referred to as parameters together with the standard deviation and average value of the normal distribution function of the hidden layer, and parameters obtained by repeating learning over a period longer than a certain learning period are referred to as past patterns. The past pattern storage component 14 stores a past pattern derived by repetition of learning for each medical examination device 21. When the prediction component 15 predicts the part replacement time (life), the notification component 16 transmits information on the predicted replacement time to the service engineer system 31 installed in the maintenance center 30.

  As illustrated in FIG. 2, the input layer of the prediction component 15 includes six nodes (neurons), and predicts the number of photographing on the seventh day of the next day from the number of consecutive six days of photographing. The learning process requires a record of the number of shootings for seven consecutive days. The number of shootings for the first 6 days is input to each node in the input layer. It is assumed that the number of shootings input to the i-th node is I (i) {i = 1, 2,... 6}.

The hidden layer is composed of at least six nodes. The output from each node in the input layer becomes the input to the node in the hidden layer. At each node of the hidden layer, the input weighted sum is calculated. Let J (i, j) be the input from the i-th node of the input layer to the j-th node of the hidden layer. When the weight is W (i, j), the weighted sum S (j) is as follows. The sum of Σ is performed for all i.
S (j) = ΣW (i, j) × J (i, j)
Next, the node output is calculated using the membership function with S (j) as an input. A normal distribution function is used as the membership function. Assuming that the average value and standard deviation at the j-th node are X (j) and σ (j), respectively, the node output K (j) is as follows.
K (j) = N (S (j), X (j), σ (j))
The output layer is composed of one node. The output from the hidden layer becomes the input to the output layer. At the node in the output layer, the input weighted sum is calculated and used as the output. If the weight from the jth node of the hidden layer to the input to the output layer is U (j), the output L of the output layer is as follows. The sum of Σ is performed for all j.
L = Σ U (j) × K (j)
The difference between the predicted value L and the actual number of shots is calculated, and the weighted sum weight W (i, j) in the hidden layer and the weighted sum weight in the output layer so that the difference is smaller than the threshold value. The learning process is to adjust U (j), the average value X (j) of the normal distribution function of the hidden layer, and the standard deviation σ (j) of the normal distribution function of the hidden layer using the error back propagation method.

  The learning process will be described with reference to FIG. In the following, the weight W (i, j) of the weighted sum in the hidden layer, the weight U (j) of the weighted sum in the output layer, the average value X (j) of the normal distribution function of the hidden layer, The standard deviation σ (j) of the normal distribution function is generically called a parameter.

  In step S <b> 1, the learning component 13 causes the prediction component 15 to read the initial parameter values from the storage component 14. In step S <b> 2, the past pattern learning component 13 causes the measurement value storage component 12 to cause the prediction component 15 to read the number of shootings for the first six days of the period (for example, one month) used for learning. In step S3, the prediction component 15 calculates the number of shootings on the next seventh day corresponding to the number of shootings for the first six days. In step S4, the learning component 13 calculates, as an error, the difference between the predicted number of shootings and the actual number of shootings on the seventh day. In step S5, the learning component 13 compares the error with a threshold value. If the error is greater than or equal to the threshold, the learning component 13 compares the number of iterations (the number of passes in S7) with the scheduled number in step S6. If the number of iterations is less than or equal to the scheduled number, the learning component 13 updates the parameter by the error propagation method in step S7. While updating the parameters, the processes of S3 to S6 are repeated until the error converges to be smaller than the threshold or the number of iterations exceeds the scheduled number.

  When the error has converged to less than the threshold value, or when the number of iterations exceeds the scheduled number, in step S8, the learning component 13 determines that the last day of the 6 consecutive days used for input in the immediately preceding S2 is the learning. It is determined whether or not the last day of the period of use (for example, one month) has been reached. When the last day of the six consecutive days used for input in S2 immediately before has not reached the last day of the period used for learning, in step S9, the learning component 13 selects the period of six consecutive days used for input. Shift by one day and return to step S2. When the last day of the six consecutive days used for input in the immediately preceding S2 has reached the last day of the period used for learning, in step S10, the parameter at that time is used as past pattern information as the past pattern storage component 14 Save to and exit.

  Such a learning process by the past pattern learning component 13 may be started at regular intervals by a timer process. In addition, after the start, the learning process may be kept executed, and the process may be waited without ending after step S9, and step S2 may be started at regular intervals. In this case, if the number of times of photographing to be read is not stored in the measured value storage component 12, the process proceeds to the standby state without executing the steps after step S2, or the process ends. Before the end, it may be output that the number of photographing is not read. In addition, after the start, the learning process may be performed, and the process may be waited for without ending after step S <b> 9, and step S <b> 2 may be started each time the number of photographing is stored in the measured value storage component 12. Further, when step S1 is started, the previous learning pattern information may be read as an initial value. Moreover, you may start the period used for learning by step S2 from the day which moved 1 day or more from the 6th last used by the last learning. The input may be a period longer than 6 days or a shorter period. Further, a plurality of nodes in the output layer may be provided, and prediction may be performed in a period longer than the next day.

  Next, a prediction process by the prediction component 15 will be described with reference to FIG. First, in step S <b> 11, the prediction component 15 reads the latest past pattern information from the past pattern storage component 14. Subsequently, in step S <b> 12, the prediction component 15 reads from the measurement value storage component 12 the oldest number of consecutive six days of shooting that is unused in the prediction process. In step S13, the number of shootings for six consecutive days is input to each node of the input layer, weighted addition is repeated according to the past pattern information, and a predicted value of the number of shootings for the next day is output from the node of the output layer. In step S14, it is determined whether or not the date of the number of photographing predicted in S13 has reached the end of the prediction period, for example, two weeks ahead from today. When the date of the number of times of photographing predicted in S13 has not reached the end of the prediction period, the prediction component 15 moves the input date for six days by one day in step S15.

  In step S <b> 16, the prediction component 15 determines whether or not the number of photographings measured on the new input date after movement is stored in the measurement value storage component 12. When the number of times of photographing is measured on the new input date, the process returns to step S12, and the number of times of photographing for six consecutive days including the new input date is read from the measured value storage component 12. Similarly, the number of times of photographing on the next day is read in step S13. Predict. When the number of shootings is not measured on the new input day, the process proceeds to step S17, and the number of shootings predicted in the immediately preceding step S13 is read in step S12 as the number of shootings on the new input day together with the number of shootings for the remaining five days. Similarly, the number of shootings for the next day is predicted in step S13.

  In step S14, the processing of S12 to S17 is repeated until the predicted date of S13 reaches the end of the scheduled prediction period, and the number of times of shooting for each day in the scheduled end of the predicted period is predicted. Note that the number of shootings at a future time of seven days or more from the execution date (today) of the prediction process is predicted based on the number of shootings for all six days predicted.

  When the prediction date of S13 reaches the end of the scheduled prediction period, the prediction component 15 newly calculates the cumulative number of shootings after the installation time or the previous parts replacement time in step S18 as the prediction process execution date (today). Each day from the next day to the end of the prediction period is calculated from the measured number of shots (actually measured value) and the predicted number of shots (predicted value). In step S <b> 19, the prediction component 15 compares the calculated cumulative number of times of each day with the number of times of photographing (threshold value) representing the life defined for each part, and identifies the day that exceeds the threshold value for the first time. In step S <b> 20, the prediction component 15 transmits the specific date to the service engineer system 31 via the notification component 16 as the parts replacement limit time.

  FIG. 5 shows an actual prediction result using this embodiment. The learning period was set to 20 days, and the number of shootings per day from the day after the last day of the measurement of the number of shootings to the point when 50 days had elapsed was predicted. High prediction accuracy can be understood.

  In the above description, the number of shootings has been described as an example of the usage situation. However, when the value itself indicates a degradation situation, such as the tube voltage or the tube temperature, the predicted value is not the cumulative value. Compare with the threshold as it is.

  The prediction component 15 may start the prediction process at regular intervals by timer processing. In addition, after the start, the prediction process may be executed, step S11 may be started without ending after step S20, and step S11 to step S20 may be repeated. In this case, if the past pattern information read in step S11 is not output, step S11 is started again or is not executed without executing the subsequent steps. Before the end, it may be output that the past pattern information has not been read.

  In addition, after the start, the process may be executed, and the process may be waited for without ending after step S20, and step S11 may be started at regular intervals. In this case, if the past pattern information to be read is not stored in the past pattern storage component 12, the process proceeds to the standby state without executing the subsequent steps or ends. Before the end, it may be output that the past pattern information has not been read. In addition, after the start, the prediction process may remain executed, and the process may wait without finishing after step S20, and step S11 may be started each time the learning component 13 outputs past pattern information. The input may be a period longer than 6 days or a shorter period. Further, a plurality of nodes in the output layer may be provided, and prediction may be performed in a longer period than the next day.

  FIG. 6 shows a process in which the notification component 16 notifies the replacement time in step S20 of FIG. After the start, in step S21, the notification component 16 reads the replacement time output by the prediction component 15. In step S22, the notification component 16 creates a notification document including the replacement time. In step S23, the notification component 16 transmits the notification document to the service engineer by e-mail and ends.

  The notification process of the notification component 16 may be started at regular intervals by timer processing. In addition, after the start, the notification process may be executed, step S21 may be started without ending after step S23, and step S21 to step S23 may be repeated. In this case, if the replacement time read in step S21 is not output, step S21 is started again or is not executed without executing the subsequent steps. Before the end, it may be output that the replacement time has not been read. In addition, after the start, the process may be executed, and the process may be waited without ending after step S23, and step S21 may be started at regular intervals. In this case, if the replacement time to be read is not output, the process proceeds to the standby state without executing the subsequent steps or ends. Before the end, it may be output that the replacement time has not been read. In addition, after the start, the notification process may be executed, and the process may be waited without ending after step S23, and step S21 may be started each time the prediction component 15 outputs the replacement time. In addition, after the start, the notification process remains executed and waits without ending after step S23. When the prediction component 15 outputs the replacement time, the notification component is stored in the queue system and starts step S21. You may read the replacement time from. The replacement time may be stored in a storage system such as a file or database, the storage location may be read from the queue, and the replacement time may be read from the storage system. For the notification in step S23, notification means other than electronic mail such as fax and telephone may be used, or a notification document may be printed and notified by mail. The notification document may include not only the replacement time but also information such as the hospital name and the predicted date and time. Further, the notification document may include a URL to a WWW site describing detailed information and related information, and link information that can automatically refer to the site using mail software or the like.

  The past pattern learning component 13 described above may use a genetic algorithm in the learning process. FIG. 7 shows an example of a gene set when a genetic algorithm is used in the learning process of the learning component 13. FIG. 8 shows a learning process using a genetic algorithm in the learning component 13. A combination of the weight, the average value of the normal distribution function, and the standard deviation for each node in the hidden layer is set as a parameter (gene) of the hidden layer node in FIG. Further, the combination of weights for the nodes in the output layer is set as a parameter (gene) of the output layer node. Using a combination of these parameters as a gene set, a plurality of gene sets having different weights and average values are prepared. After the start, in step S31, prediction is performed using these sets, an error is calculated, and a set with a small error is selected. If the minimum error is equal to or smaller than the threshold value in step S32, the minimum error set is output as past pattern information, and the process ends. If the number of repetitions exceeds the scheduled number in step S33, the minimum error set is output as past pattern information, and the process ends. In the mating in step S34, weights and average values are exchanged between corresponding nodes of the selected set of genes (parent genes) to create a new gene (child gene). In the mutation in step S35, the weight or average value of the selected set of genes (parent genes) is changed to create a new gene (child gene). It repeats from step S31 using the created set of child genes.

  In the example of FIG. 8, only the child gene is used in step S31 when iterating, but the parent gene set may be included in the options when selecting the gene set. Further, the learning process may be executed using only one of step S34 and step S35. In step S33, the elapsed time may be used instead of the number of repetitions, and the process may be terminated when the scheduled time is exceeded. Further, in the example of FIG. 8, the prediction component 15 is described in the case of the neural network configuration of FIG. 2, but when the prediction component 15 performs prediction by expressing the variation of the measured value by an expression, the coefficient of the expression, etc. May be used for genes.

  The learning process of the past pattern learning component 13 is divided into two stages, learning is performed using the genetic algorithm shown in FIGS. 7 and 8 in the first stage, and the error back propagation method shown in FIG. 3 is used in the second stage using the output. You may make it learn using.

  Also, threshold values for determining the replacement time of parts are prepared for each type of inspection apparatus, type of part, and model number, and the threshold value corresponding to the target part is used when determining the replacement time. In the following, a circuit board used for an X-ray detector of an X-ray computed tomography apparatus (CT) will be described as an example. CT X-ray detectors rotate at high speed in the gantry using superconducting coils. When the test is repeated and the circuit ages, the voltage that drives the circuit becomes unstable and heat is generated. If the temperature exceeds a certain temperature, the superconducting state becomes unstable and cannot be rotated safely. Therefore, it is necessary to monitor the voltage and temperature, and to exchange the circuit if there is a fluctuation exceeding the threshold. Therefore, since the rotation speed differs depending on the model and model, and the circuits for controlling these differ, the prediction component of the system of the present invention holds a threshold value for each model and model, and determines the replacement time.

  In addition, the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

The figure which shows the structure of the medical examination equipment components replacement | exchange support system which concerns on embodiment of this invention. The figure which shows the structure of the prediction component of FIG. 2 is a flowchart illustrating a learning process by the learning component of FIG. 1. 2 is a flow diagram illustrating a prediction process by the prediction component of FIG. The figure which shows the prediction result by this embodiment with a measured value (result). 2 is a flow diagram illustrating a notification process by the notification component of FIG. The figure which shows the example of the gene set in 1 case which uses a genetic algorithm in the learning process of the learning component of FIG. 2 is a flowchart illustrating a learning process using a genetic algorithm in the learning component of FIG.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Medical test equipment component replacement support system, 11 ... Measurement value request component, 12 ... Measurement value storage component, 13 ... Past pattern learning component, 14 ... Past pattern storage component, 15 ... Prediction component, 16 ... Notification component, 20 ... Medical institution, 21 ... medical testing equipment, 22 ... electrical communication line, 30 ... maintenance center, 31 ... service engineer system, 32 ... electrical communication line.

Claims (4)

In the medical test equipment parts replacement support system that collects information on medical test equipment of medical institutions via electrical communication lines,
A receiving unit for receiving information on the usage status from the medical examination device;
A storage unit for storing information on the received usage status;
Based on the stored usage status, a future usage status is predicted based on the usage status for a predetermined period, a future usage status is predicted based on the predicted future usage status, and the stored usage status A prediction unit for predicting the replacement time of the medical examination device based on the predicted future use state;
A medical examination equipment parts replacement support system, comprising: a transmission unit that transmits information on the predicted replacement time to the service engineer system. The forecasting part is an operation for obtaining a future usage status by inputting the usage status for a predetermined period to the neural network and inputting the obtained future usage status to the neural network. The medical test equipment part replacement support system according to claim 1, wherein a future use situation is predicted repeatedly. The prediction part uses a genetic algorithm, inputs the usage status for a predetermined period to the gene set of the genetic algorithm to determine the future usage status, and inputs the calculated future usage status to the gene set, and further to the future 2. The medical test equipment part replacement support system according to claim 1, wherein a future use situation is predicted by repeating an operation of obtaining a use situation of the medical examination equipment. The usage state includes at least one of the number of times of imaging per day, the operating voltage of the X-ray tube, the temperature of the device installation location, the humidity of the device installation location, the temperature inside the device, and the humidity inside the device. The medical examination equipment component replacement support system according to claim 1.

Images