JP7635978B2 - Prescription reading program, prescription reading device, and prescription creation computer - Google Patents

Description

The present invention relates to a prescription reading program that reads information such as text on a prescription, a prescription reading device, and a computer for creating prescriptions.

Patent Document 1 describes a pharmacy business support system. It states, "Prescription information is, for example, image data obtained by photographing a prescription issued by a medical institution. Note that the photographed image may be subjected to character recognition to convert the prescription contents into text data, and the image data and text data may be used as prescription information."

JP 2020-086833 A

Pharmacies receive prescriptions from medical institutions on paper or via fax. When a pharmacy wishes to convert a paper prescription received by pharmacy into text data, it scans the prescription and converts it into text data using OCR (Optical Character Recognition). Alternatively, a pharmacy may convert image data received by fax into text data using OCR (Optical Character Recognition). However, scanned image data and image data received by fax may contain unclear characters, and conventional OCR often cannot properly recognize the characters.

The object of the present invention is to provide a prescription reading program, a prescription reading device, and a computer for creating prescriptions that can properly read prescriptions even if the characters are unclear.

The prescription reading program according to the present invention includes a receiving process for receiving prescription data, a text recognition process for performing a text data recognition process on the prescription data received in the receiving process using a machine learning character recognition algorithm, a first NLP algorithm process for extracting words defined in a prescribed prescription standard from the text data recognized in the text recognition process and attaching tags corresponding to the standard to each word, a second NLP algorithm process for understanding a syntax defined in the standard and arranging each word in an order defined in the standard based on the tags of each word assigned by the first NLP algorithm, and a data generation process for generating general-purpose data readable by a prescription creation computer based on the data organized by the second NLP algorithm.
The information processing device is caused to execute the above.

The prescription reading program uses the above-mentioned first NLP algorithm and second NLP algorithm to extract words defined in the standard specifications for a specific prescription established by JAHIS or the like from the read text data, and organizes the data in the order defined by the standard specifications. The texts written on a prescription are not necessarily arranged in the order defined by the standard specifications such as JAHIS. For example, the order of writing on a prescription may differ depending on the hospital or facility, or the prescription may be corrected by hand or additional content may be added. However, the prescription reading program of this embodiment organizes the read text data into a format conforming to the standard specifications defined by JAHIS or the like, using the first NLP algorithm and the second NLP algorithm. Therefore, the prescription reading program can organize the data into usable data for devices such as receipt computers that comply with the standard specifications defined by JAHIS or the like.

The prescription reading program may also generate general-purpose data (e.g., CSV data) that can be read by a prescription creation computer based on the data organized by the second NLP algorithm, or the prescription reading program may input the data organized by the second NLP algorithm to a prescription computer via an API or the like.

The prescription reading program can also perform a correction process to correct errors in the text data recognized in the text recognition process using a machine learning algorithm, thereby further improving the accuracy of reading prescriptions. In this case, the extraction process is performed on the text data corrected in the correction process.

The correction process is performed using a drug name dictionary and a pattern dictionary that contains specific patterns contained in prescriptions. The pattern dictionary contains information such as the character "han" used in generic drugs, the katakana character "ro" and the kanji character "kuchi". This further improves the accuracy of reading prescriptions.

The present invention makes it possible to properly read prescriptions even if the characters are unclear.

FIG. 1 is a block diagram showing a configuration of a prescription reading system 1. FIG. 2 is a block diagram showing the configuration of the server 11. 4 is a flowchart showing the operation of the prescription reading system 1. 13 is a flowchart showing the operation of the server 11 when reading a prescription. 13 is a flowchart showing the operation of the server 11 when reading a prescription. FIG. 2 is a diagram showing an example of prescription data.

The following describes the prescription reading system 1 according to an embodiment of the present invention with reference to the drawings.

Figure 1 is a block diagram showing the configuration of a prescription reading system 1. The prescription reading system 1 includes a server 11, a reception terminal 12, a user terminal 13, and a receipt computer 14. The server 11, the reception terminal 12, the user terminal 13, and the receipt computer 14 are connected via a network.

The reception terminal 12 is realized by a general information processing device (personal computer). The reception terminal 12 is connected to a fax machine or scanner (not shown) and receives prescription data converted from a prescription into image data. The prescription is sent from the medical institution by paper or fax. When a pharmacist at the pharmacy receives the prescription on paper, the pharmacist inputs the prescription data into the reception terminal 12 using a scanner (not shown). When the prescription is received by fax, the reception terminal 12 receives the prescription data from the fax machine. Alternatively, the pharmacist may print the prescription data using a fax machine, read the printed prescription with a scanner, and input it into the reception terminal 12. The reception terminal 12 transmits the prescription data to the server 11.

The server 11 is an example of a prescription reading device of the present invention. The server 11 may be installed in a pharmacy, or may be a group of servers on the Internet. The server 11 receives prescription data from the reception terminal 12 and performs an operation of reading the contents of the prescription.

The user terminal 13 is a terminal used by a person (user) who performs pharmacy work, such as a pharmacist, and is, for example, a portable information processing device such as a smartphone or a tablet computer. The user uses the user terminal 13 to check the contents of the prescription read by the server 11 and to input the contents of the prescription into the receipt computer 14. The user can also use the user terminal 13 to check the patient's medical history and create a new medical history.

The receipt computer 14 is a receipt creation computer that creates receipts based on the contents of the prescription entered from the user terminal 13. The receipt computer 14 is realized by an application program installed on a general personal computer.

Figure 2 is a block diagram showing the configuration of the server 11. The server 11 includes a network I/F 51, a storage unit 52, a CPU 53, and a RAM 54.

The network I/F 51 is connected to a network such as the Internet. In this embodiment, the server 11 is connected to the reception terminal 12 and the user terminal 13 via the network I/F 51.

The CPU 53 reads out the programs stored in the memory unit 52 into the RAM 54 and performs various operations. The memory unit 52 stores the basic program (OS) of the server 11 and the prescription reading program in this embodiment. The memory unit 52 also includes various databases such as information indicating the layout type of the prescription data described below and personal information of patients.

The CPU 53 functions as a prescription reading device by reading the prescription reading program from the memory unit 52 into the RAM 54.

Figure 3 is a flowchart showing the operation of the prescription reading system 1. First, a prescription is sent from a medical institution by paper or fax (S11). A pharmacist at a pharmacy receives the prescription sent by paper or fax (S12).

Then, the pharmacist at the pharmacy uses the reception terminal 12 to scan the paper prescription and send it as prescription data to the server 11, or sends the prescription data received by FAX to the server 11 (S13). The pharmacist may use an application program on the user terminal 13 to send the prescription data to the server 11. The reception terminal 12 or the user terminal 13 may automatically send the prescription data to the server 11 when it receives the prescription data via FAX or when it receives it from the scanner. In particular, at-home pharmacies such as elderly care facilities receive many prescriptions from medical institutions at the same time. Therefore, the prescription reading system 1 of this embodiment can significantly reduce the load on pharmacy operations by automatically sending the prescription data to the server 11 when it receives the prescription data and performing the following reading operation.

Next, the server 11 executes the prescription reading program to perform the prescription reading operation (S14).

Figures 4 and 5 are flowcharts showing the operation of the server 11 when reading a prescription. First, the server 11 extracts image features from the prescription data (S101). This process is performed using an algorithm for extracting features, such as SIFT (Scale Invariant Feature Transform).

As described above, pharmacies receive prescription data by fax or scan it with a prescription scanner. Therefore, the image size and inclination of the prescription data are not constant. By using the SIFT feature extraction algorithm, server 11 can extract features of the prescription image that are independent of image size and inclination.

Next, the server 11 determines the corresponding layout type from the database based on the extracted features (S102). The database stores multiple layout types of prescriptions. The server 11 refers to the database using an estimation algorithm such as RANSAC (Random Sample Consensus) to determine the layout type that matches the layout type of the received prescription data.

The server 11 determines whether a matching layout type exists (S103). If the server 11 determines that a matching layout type exists, it performs a text area determination process (S104). The text area includes at least an area in which patient information such as the patient's name is recorded, an area in which the date is recorded, and an area in which drug information is recorded. The layout types registered in the database include position information that indicates where each text area exists in the prescription data. The server 11 determines the text area by reading the position information of the text area from the layout types registered in the database.

On the other hand, if the server 11 determines that there is no matching layout type, it performs a process of accepting a text area from the user (S105). A user such as a pharmacist manually specifies an area in which text is written from within the prescription data using the user I/F of the reception terminal 12 or the user terminal 13. The server 11 registers the specified text area in the database as a new layout type (S106) and performs a process of determining the text area (S104).

New layout types may arise for prescriptions due to legal changes, etc., and may not match the layout types stored in the database. However, if there is no matching layout type in the database, the server 11 accepts a text area specification operation from the user and performs a registration process to register the text area in the database, so that even if a new layout type arises, the layout type can be automatically determined from the next time.

Then, the server 11 performs text recognition processing on the text area in which the patient information is written (S107). The patient information includes information such as the patient's name and date of birth. The text recognition processing is performed using any character recognition algorithm (for example, Tesseract-OCR equipped with an algorithm such as a neural network).

The server 11 compares the patient information recognized by the text recognition process with the database to determine whether or not corresponding patient information exists (S108). If the server 11 determines that a corresponding patient exists, it continues the process of S201 to S207 shown in FIG. 5. If the server 11 determines that a corresponding patient does not exist, it registers the recognized patient information in the database (S109) and proceeds to the process of S201 to S207.

Next, as shown in FIG. 5, the server 11 performs segmentation processing on the text area in which the drug information is described (S201). FIG. 6 is a diagram showing an example of prescription data. The segmentation processing is a process of dividing the text area in which the drug information is described into segments of one line each, as shown in the hatched portion of FIG. 6.

Then, the server 11 performs text recognition processing on each segment (S202). The text recognition processing is performed using any character recognition algorithm (for example, Tesseract-OCR equipped with an algorithm such as a neural network). It is preferable for this processing to use a machine-learned character recognition algorithm that has been machine-learned for prescriptions. Of course, the server 11 may also perform text recognition processing using a general machine-learned algorithm.

Next, the server 11 performs a spelling correction process on the recognized text data (S203). The spelling correction process uses any algorithm, such as SymSpell. The SymSpell algorithm uses a drug name dictionary provided by the government to correct text recognition errors related to drug names. This allows the server 11 to appropriately correct errors that occurred in the text recognition process of S202.

Furthermore, the server 11 performs a correction process for the specific pattern in the prescription (S204). The correction process for the specific pattern uses any algorithm, such as n-gram. The n-gram algorithm uses a pattern dictionary that includes the specific pattern in the prescription to correct text recognition errors that could not be corrected by the process in S203. The specific pattern in the prescription includes, for example, the character "han" used in generic drugs. The n-gram algorithm also corrects the difference between the katakana character "ro" and the kanji character "kuchi".

The server 11 performs a first word extraction process (S205) and a second word extraction process (S206) using the corrected text data. Both the first word extraction process and the second word extraction process are performed using a machine-learned word extraction algorithm such as a neural network. The machine-learned word extraction algorithm is, for example, an NLP (Natural Language Processing) algorithm. The first NLP algorithm used in the first word extraction process of this embodiment extracts words specified in a standard specification of a predetermined prescription and attaches a tag corresponding to the standard specification to each word. The standard specification of a predetermined prescription is, for example, a standard specification for prescription data specified by JAHIS. The standard specification specifies the order of various information indicating the contents of a prescription. Data conforming to the standard specification specified by JAHIS can be input to the receipt computer 14 and used. The first NLP algorithm is learned to extract and tag words specified in the standard specification specified by JAHIS. Each tag includes, for example, the name of the medication, dosage, duration of medication, and timing of intake.

The second NLP algorithm used in the second word extraction process is trained to understand the relationship (syntax and order) between keywords specific to prescriptions. In other words, the second NLP algorithm learns syntax and order according to the contents of the standard defined by JAHIS. For example, an NLP algorithm trained to read English sentences, which is different from the present embodiment, tags the form of the English words read (nouns, verbs, etc.) and rearranges the words in the order defined by the English syntax. In contrast, the second NLP algorithm of the present embodiment understands the syntax of the standard defined by JAHIS and rearranges and organizes the words in the order defined by the JAHIS standard based on the tags of each word assigned by the first NLP algorithm. For example, in the prescription data shown in FIG. 6, the second NLP algorithm performs the following steps: 1. Generic name prescription with the name of the drug "XXXX tablets" 12 mg, 2. Dosage "3 tablets", 3. Medication period "14 days", 4. The timing of intake, "once a day before going to bed," is associated in this order. Similarly, the second NLP algorithm associates, in this order, 1. generic name prescription, drug name "○△△ tablets" 7.5 mg, 2. dosage "1 tablet," 3. dosing period "14 days," and 4. timing of intake, "once a day before going to bed."

By using the above-mentioned first NLP algorithm and second NLP algorithm, the server 11 can organize the read text data into the data order specified by the JAHIS standard. The texts written on a prescription are not necessarily arranged in the order specified by the JAHIS standard. For example, the order of writing on a prescription may differ depending on the hospital or facility, or the prescription may be corrected by hand or additional content may be added. However, the prescription reading program of this embodiment organizes the read text data into a format conforming to the JAHIS standard by using the first NLP algorithm and the second NLP algorithm. Therefore, the prescription reading program extracts the unique contents of the prescription with high accuracy and organizes them into a format that can be appropriately registered in the database and used. In other words, the prescription reading program organizes all prescriptions into data in an order conforming to the JAHIS standard, so that any format of prescription can be entered into the receipt computer and organized into usable data.

The server 11 registers the words extracted and organized in the above manner in a database (S207). Then, returning to the flowchart of FIG. 3, the server 11 converts the read result into a two-dimensional code (QR Code (registered trademark)) that meets a standard (e.g., JAHIS) that can be read by the medical receipt computer 14 (S15). The server 11 also registers the read prescription data, the position information of the words registered in S207 (information indicating the position of each word in the prescription data), and the QR Code (registered trademark) generated in S15 in a database (S16).

Meanwhile, the server 11 transmits the information registered in the database to the user terminal 13 and updates the information on the user terminal 13 (S17). The pharmacist can check the contents of each prescription using an application program on the user terminal 13.

Then, the QR code (registered trademark) generated by the server 11 is input to the receipt computer 14 via an application program on the user terminal 13 (S18). For example, the application program displays the QR code (registered trademark) generated by the server 11 on a display (not shown). The pharmacist reads the displayed QR code (registered trademark) with a scanner. As a result, the prescription data read by the prescription reading program is input to the receipt computer 14.

The prescription reading program may also generate general-purpose data such as CSV format in which text is written in the order organized by the second NLP algorithm. The prescription reading program inputs the generated CSV data into a data import folder in the medical receipt computer 14. The medical receipt computer 14 receives prescription data from the prescription reading program via the data import folder. The medical receipt computer 14 is configured to be able to input CSV data that conforms to the JAHIS standard. Therefore, the prescription reading program of this embodiment can input prescription data into the medical receipt computer 14 accurately and automatically.

Alternatively, the prescription reading program may input prescription data to the receipt computer 14 via an API (Application Programming Interface). In this case, the receipt computer 14 is configured to be able to input data conforming to the JAHIS standard specifications via the API. This allows the prescription reading program to also input prescription data to the receipt computer 14 via the API. Therefore, in this case too, the prescription reading program of this embodiment can input prescription data accurately and automatically into the receipt computer 14.

In this way, the prescription reading program can also input prescription data into the receipt computer 14 without using a QR code (registered trademark). In this case, the prescription reading program can also reduce the workload of pharmacists and other workers who read the data with a scanner, and can automate the input of prescription data into the receipt computer 14.

The prescription reading program of this embodiment determines the layout type of the prescription to determine the text area, and extracts words such as the name and amount of the medicine from the determined text area using a machine-learned character recognition algorithm and a machine-learned word extraction algorithm such as a neural network. Therefore, the prescription reading program of this embodiment can perform character recognition and word recognition with much higher accuracy than conventional OCR. Therefore, the prescription reading program of this embodiment can properly read the prescription even if the prescription data is unclear and has been captured by fax or scanner. In addition, since the prescription reading program organizes the read text data into a data format that conforms to the JAHIS standard, the prescription data can be accurately and automatically entered into the receipt computer even if the order of various information in the prescription is different, or if the information has been corrected by hand or additional information has been added. In particular, at home pharmacies, many prescriptions are received from medical institutions at the same time, and the prescription reading program of this embodiment can significantly reduce the burden of reading prescriptions and entering them into the receipt computer.

The server 11 stores the information of the reading results for each patient in a database. The pharmacist can refer to all past reading results for each patient using the application program of the user terminal 13. When the information of the user terminal 13 is updated in the process of S17 in FIG. 3, the pharmacist can check the contents of the prescription and confirm the explanation of the prescribed medicine and the patient's taking the medicine by telephone or the like. The pharmacist uses the user terminal 13 to record these contents as a drug taking history management record (hereinafter referred to as "medication history"). The drug history is composed of subjective information (what the patient said, subjective symptoms) (S: Subjective), objective information (prescription contents, changes, pharmacist's observations and findings, test data) (O: Objective), evaluation and guidance (the pharmacist's evaluation and judgment from S and O and the guidance contents based on it) (A: Assessment), and plan (points to ask next time and things to be careful about) (P: Plan). After the information on the user terminal 13 is updated in the process of S17, the pharmacist can easily create a new medication history by referring to past information in addition to the current prescription. In this way, the prescription reading system 1 of this embodiment also functions as a medication history creation work support program.

The user terminal 13 may also transmit the medication history created by the pharmacist to the server 11. By registering the new medication history in the database, the server 11 can present the medication history to the pharmacist via the user terminal 13 the next time a new prescription is read.

Furthermore, the server 11 may automatically create a new medical history based on the contents of the newly read prescription and the past medical history. The medical history is created using a specified algorithm. The server 11 sends the new medical history to the user terminal 13. The pharmacist checks the medical history created by the server 11, and modifies the medical history created by the server 11 depending on the results of explaining the medical history to the patient and checking the situation. Alternatively, if the pharmacist determines that there is no problem with the medical history created by the server 11 as a result of explaining the medical history to the patient and checking the situation, the pharmacist confirms the medical history created by the server 11. When the pharmacist confirms the medical history, the confirmed medical history is registered in the server 11 and the receipt computer 14 from the user terminal 13.

In this way, the server 11 of this embodiment can automatically create a medication history, thereby supporting the medication history creation process and further reducing the burden on pharmacy operations.

The description of this embodiment should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is indicated by the claims, not by the above-described embodiment. Furthermore, the scope of the present invention includes the scope equivalent to the claims.

For example, the prescription reading program may be executed not by the server 11 but by the reception terminal 12, the user terminal 13, or the receipt computer 14. When the receipt computer 14 executes the prescription reading program, the receipt computer 14 performs reception processing to directly accept the data organized by the second NLP algorithm in a format conforming to the JAHIS standard. In this case, the prescription reading program does not impose a workload on an operator such as a pharmacist to read with a scanner, and there is no need to send and receive data via CSV data or API, making it possible to completely automate the reading of prescriptions and the acceptance of prescription data within a single program. Note that the receipt computer 14 may execute the prescription reading program as plug-in software.

1... Prescription reading system 11... Server 12... Reception terminal 13... User terminal 14... Receipt computer 51... Network I/F
52...Storage unit 53...CPU
54...RAM

Claims (8)

A receiving process for receiving prescription data;
A text recognition process that performs a recognition process of text data on the prescription data received in the receiving process using a machine-learned character recognition algorithm;
A first NLP algorithm process for extracting words defined in a standard specification for a predetermined prescription from the text data recognized by the text recognition process and attaching a tag corresponding to the standard to each word;
A second NLP algorithm process that understands the syntax defined in the standard and organizes each word in an order defined in the standard based on the tag of each word assigned in the first NLP algorithm process;
A data generation process for generating general-purpose data readable by a computer for creating a medical receipt based on the data organized by the second NLP algorithm process ;
A prescription reading program that causes an information processing device to execute the above. A receiving process for receiving prescription data;
A text recognition process that performs a recognition process of text data on the prescription data received in the receiving process using a machine-learned character recognition algorithm;
A first NLP algorithm process for extracting words defined in a standard specification for a predetermined prescription from the text data recognized by the text recognition process and attaching a tag corresponding to the standard to each word;
A second NLP algorithm process that understands the syntax defined in the standard and organizes each word in an order defined in the standard based on the tag of each word assigned in the first NLP algorithm process;
An input process of inputting the data organized by the second NLP algorithm process into a computer for creating a medical receipt via an application programming interface;
A prescription reading program that causes an information processing device to execute the above. A receiving process for receiving prescription data;
A text recognition process that performs a recognition process of text data on the prescription data received in the receiving process using a machine-learned character recognition algorithm;
A first NLP algorithm process for extracting words defined in a standard specification for a predetermined prescription from the text data recognized by the text recognition process and attaching a tag corresponding to the standard to each word;
A second NLP algorithm process that understands the syntax defined in the standard and organizes each word in an order defined in the standard based on the tag of each word assigned in the first NLP algorithm process;
A reception process for receiving data organized by the second NLP algorithm process ;
A prescription reading program that causes a prescription creation computer to execute the above. Further executing a correction process for correcting errors in the text data recognized by the text recognition process using a machine learning algorithm;
The process of extracting words is performed on the text data corrected in the correction process.
The prescription reading program according to any one of claims 1 to 3. The correction process is performed using a drug name dictionary and a pattern dictionary that indicates a specific pattern included in the prescription.
The prescription reading program according to claim 4. A receiving unit for receiving prescription data;
a text recognition unit that performs a text data recognition process on the prescription data received by the receiving unit;
A first NLP algorithm processing unit that extracts words defined in a standard specification of a predetermined prescription from the text data recognized by the text recognition unit and attaches a tag corresponding to the standard to each word;
A second NLP algorithm processing unit that understands a syntax defined in the standard and organizes each word in an order defined in the standard based on the tag of each word assigned by the first NLP algorithm processing unit;
A data generation processing unit that generates general-purpose data that can be read by a prescription creation computer based on the data organized by the second NLP algorithm processing unit;
A prescription reading device comprising: A receiving unit for receiving prescription data;
a text recognition unit that performs a text data recognition process on the prescription data received by the receiving unit;
A first NLP algorithm processing unit that extracts words defined in a standard specification of a predetermined prescription from the text data recognized by the text recognition unit and attaches a tag corresponding to the standard to each word;
A second NLP algorithm processing unit that understands a syntax defined in the standard and organizes each word in an order defined in the standard based on the tag of each word assigned by the first NLP algorithm processing unit;
An input processing unit that inputs the data organized by the second NLP algorithm processing unit into a prescription creation computer;
A prescription reading device comprising: A receiving unit for receiving prescription data;
a text recognition unit that performs a text data recognition process on the prescription data received by the receiving unit;
A first NLP algorithm processing unit that extracts words defined in a standard specification of a predetermined prescription from the text data recognized by the text recognition unit and attaches a tag corresponding to the standard to each word;
A second NLP algorithm processing unit that understands a syntax defined in the standard and organizes each word in an order defined in the standard based on the tag of each word assigned by the first NLP algorithm processing unit;
A reception processing unit that receives the data organized by the second NLP algorithm processing unit;
A prescription creation computer equipped with the following:

Images