WO2023113452A1

WO2023113452A1 - Method, device, and program for filtering noise data of medical text on basis of artificial intelligence

Info

Publication number: WO2023113452A1
Application number: PCT/KR2022/020295
Authority: WO
Inventors: 김민승; 이호익; 최상민; 한요섭
Original assignee: 솔닥 주식회사; 연세대학교 산학협력단
Priority date: 2021-12-14
Filing date: 2022-12-14
Publication date: 2023-06-22
Also published as: KR102445098B1

Abstract

The present invention relates to a method, a device, and a program for filtering noise data of a medical text on the basis of artificial intelligence, and allows a medical text obtained by converting voice data corresponding to a telemedicine conversation into text data to be embedded and grouped by word so that noise word data is identified, and allow new text data to be filtered on the basis of the identified noise word data, and thus can provide a medical text with high accuracy and reliability.

Description

Method, apparatus and program for filtering noise data of medical text based on artificial intelligence

The present invention relates to a method for filtering noise data of medical text, and more particularly, to a method, apparatus and program for filtering noise data of medical text based on artificial intelligence capable of filtering noise data of medical text extracted from conversation contents in a telemedicine process. It is about.

Recently, as a new type of infectious disease is prevalent, the need for non-face-to-face medical services is increasing.

Due to the need for non-face-to-face medical services, telemedicine providing medical services such as diagnosis, treatment, and consultation online is increasing.

Telemedicine is an online medical examination using at least three of the five examination methods (interview, inspection, palpation, percussion, and auscultation) conducted by a doctor, as well as a urine test, blood test, and electrocardiogram test, similar to hospitals. It means conducting tests to diagnose, prescribe and treat.

When telemedicine is implemented, specialized medical care can be provided even to patients who are far away, eliminating regional concentration of medical services, and enabling medical resources to be operated as efficiently as possible, ultimately reducing medical expenses. can be expected

Telemedicine can be conducted in the form of a web or app on a mobile device, and based on voice-to-text conversion technology, conversations between doctors and patients during medical consultation are extracted in text format.

However, since the accuracy of the extracted text data depends on the voice-to-text conversion technology, if there is a problem in the voice-to-text conversion technology itself, inaccurate text may be extracted and an error may occur in the medical treatment data between the doctor and the patient.

Errors in these medical data may adversely affect not only artificial intelligence learning but also patient care, resulting in incorrect prescriptions.

Therefore, in the future, there is a need to develop a technology for filtering noise data of medical text that can provide medical text with high accuracy and reliability by filtering noise data of medical text extracted from conversation contents during a remote medical treatment process.

One object of the present invention to solve the above-described problems is to identify noise word data by embedding and clustering medical text by word by converting voice data corresponding to a telemedicine conversation into text data, and identifying the identified noise word. An object of the present invention is to provide a method, apparatus, and program for filtering noise data of medical text capable of providing medical text with high accuracy and reliability by filtering new text data based on data.

The problems to be solved by the present invention are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

A method for filtering noise data of medical text according to an embodiment of the present invention for solving the above problems includes: (a) generating medical text by converting voice data corresponding to a telemedicine conversation into text data; ) embedding the medical text word by word for each sentence, (c) identifying noise word data by clustering the embedded words, and storing the identified noise word data in a noise dictionary; generating a noise filter based on noise word data stored in a noise dictionary; (e) checking whether a new medical text corresponding to the telemedicine conversation is generated; and (f) when the new medical text is generated, and filtering the new medical text through a noise filter to reconstruct the medical text from which noise word data is removed.

In an embodiment, the step (a) is characterized in that the medical text is generated by converting voice data corresponding to the telemedicine conversation into text data through speech-to-text (STT).

In an embodiment, the medical text generated by STT (Speech-to-Text) includes prescription information including at least one of a disease name and medication guidance, and at least one of age, gender, and region of residence. Characterized in that it includes patient information.

In an embodiment, the step (b) may include inputting sentence data of the medical text into a pre-learned neural network model and embedding the input sentence data word by word.

In an embodiment, the step (c) generates a plurality of clusters by clustering words through a k-means clustering algorithm based on the location information of the embedded words, and generating the clusters. If there is word data that does not belong to the word data, it is characterized in that the corresponding word data is regarded as noise word data.

In an embodiment, in the step (c), if a cluster consisting of less than k words, which is the minimum number of words, exists among the generated clusters, words included in the corresponding cluster are regarded as noise word data. .

In an embodiment, in the step (c), when a plurality of clusters are generated by clustering the words, the plurality of clusters are reclassified based on treatment features, and noise word data is obtained from the reclassified treatment feature based clusters. and storing the identified treatment feature-based noise word data in a noise dictionary.

In an embodiment, in the step (c), clusters are reclassified based on prescription information among the treatment characteristics, and noise word data based on prescription information is identified from the reclassified clusters and stored in the noise dictionary. to be characterized

In an embodiment, the step (c) includes reclassifying clusters based on patient information among the treatment characteristics, identifying noise word data based on patient information from the reclassified clusters, and storing the noise word data in the noise dictionary. to be characterized

In an embodiment, the step (d) generates a general noise filter from general information-based noise word data stored in the noise dictionary, and generates a treatment feature-based noise filter from treatment feature-based noise word data stored in the noise dictionary. It is characterized by doing.

In an embodiment, the step (d) is characterized in that, when generating the treatment feature-based noise filter, a treatment feature-based noise filter including a prescription information-based noise filter and a patient information noise filter is generated.

In an embodiment, in step (f), when the new medical text is generated, noise words included in each sentence of the new medical text are removed based on a noise filter corresponding to each sentence of the new medical text; It is characterized in that the medical text from which noise words are removed is reconstructed.

In an embodiment, in step (f), if the sentence of the new medical text is a general information related sentence, noise words included in the general information related sentence of the medical text are removed based on a general noise filter, and the new medical text is a general information related sentence. If the sentence of the text is a sentence related to the treatment feature, noise words included in the treatment feature sentence of the medical text are removed based on the treatment feature-based noise filter.

In an embodiment, in step (f), if the sentence of the new medical text is a prescription information sentence, noise words included in the prescription information sentence of the medical text are removed based on a prescription information-based noise filter, and the new medical text is a prescription information sentence. If the text sentence is a patient information sentence, noise words included in the patient information sentence of the medical text are removed based on a patient information based noise filter.

In addition, a computing device according to an embodiment of the present invention is a computing device for providing a method for filtering noise data of medical text, and includes a processor including one or more cores and a memory, wherein the processor corresponds to a telemedicine conversation. Medical text is generated by converting speech data to text data, the medical text is embedded by word for each sentence, and noise word data is identified by clustering the embedded words, and the identified noise word data is converted into a noise dictionary. and generates a noise filter based on the noise word data stored in the noise dictionary, checks whether a new medical text corresponding to the telemedicine conversation is generated, and if the new medical text is generated, the noise filter It is characterized in that the new medical text is reconstructed into a medical text from which noise word data is removed by filtering the new medical text.

A computer program providing a method for filtering noise data of medical text according to another embodiment of the present invention for solving the above problems is combined with a computer that is hardware and stored in a medium to perform any one of the above methods. do.

In addition to this, another method for implementing the present invention, another system, and a computer readable recording medium recording a computer program for executing the method may be further provided.

As described above, according to the present invention, medical texts obtained by converting voice data corresponding to telemedicine conversations into text data are embedded and clustered for each word to identify noise word data, and new text data is generated based on the identified noise word data. By filtering, medical text with high accuracy and reliability can be provided.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a block diagram of a computing device performing an operation for providing a method for filtering noise data of medical text according to an embodiment of the present invention.

2 to 7 are conceptual diagrams for explaining a method of filtering noise data of medical text according to an embodiment of the present invention.

8 is a flowchart illustrating a method of filtering noise data of medical text according to an embodiment of the present invention.

Advantages and features of the present invention, and methods of achieving them, will become clear with reference to the detailed description of the following embodiments taken in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, only these embodiments are intended to complete the disclosure of the present invention, and are common in the art to which the present invention belongs. It is provided to fully inform the person skilled in the art of the scope of the invention, and the invention is only defined by the scope of the claims.

Terminology used herein is for describing the embodiments and is not intended to limit the present invention. In this specification, singular forms also include plural forms unless specifically stated otherwise in a phrase. As used herein, "comprises" and/or "comprising" does not exclude the presence or addition of one or more other elements other than the recited elements. Like reference numerals throughout the specification refer to like elements, and “and/or” includes each and every combination of one or more of the recited elements. Although "first", "second", etc. are used to describe various components, these components are not limited by these terms, of course. These terms are only used to distinguish one component from another. Accordingly, it goes without saying that the first element mentioned below may also be the second element within the technical spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used with meanings commonly understood by those skilled in the art to which the present invention belongs. In addition, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless explicitly specifically defined.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Prior to the description, the meaning of the terms used in this specification will be briefly described. However, it should be noted that the description of terms is intended to help the understanding of the present specification, and is not used in the sense of limiting the technical spirit of the present invention unless explicitly described as limiting the present invention.

In this specification, neural networks, artificial neural networks, and network functions may often be used interchangeably.

Also, throughout this specification, a neural network, a neural network, and a network function may be used with the same meaning. A neural network may consist of a set of interconnected computational units, which may be generally referred to as “nodes”. These “nodes” may also be referred to as “neurons”. A neural network includes at least two or more nodes. Nodes (or neurons) constituting neural networks may be interconnected by one or more “links”.

The configuration of the computing device 100 shown in FIG. 1 is only a simplified example. In one embodiment of the present invention, the computing device 100 may include other components for performing a computing environment of the computing device 100, and only some of the disclosed components may constitute the computing device 100.

The computing device 100 may include a processor 110 , a memory 130 , and a network unit 150 .

In the present invention, the processor 110 converts voice data corresponding to the telemedicine conversation into text data to generate medical text, embeds the medical text by word in each sentence, and clusters the embedded words to generate noise words. Identify data, store the identified noise word data in a noise dictionary, create a noise filter based on the noise word data stored in the noise dictionary, check whether new medical text corresponding to the telemedicine conversation is generated, and check whether a new medical text corresponding to the telemedicine conversation is created Once the text is generated, the new medical text can be reconstructed into medical text from which noise word data has been removed by filtering the new medical text through a noise filter.

Here, the processor 110 may generate medical text by converting voice data corresponding to the telemedicine conversation into text data through speech-to-text (STT).

For example, medical text generated by STT (Speech-to-Text) includes prescription information including at least one of a disease name and medication map, and patient information including at least one of age, gender, and region of residence. It may include, but this is only one embodiment, but is not limited thereto.

Next, the processor 110 may input the sentence data of the medical text into the pretrained neural network model and embed the input sentence data word by word.

For example, the neural network model may include a skip-gram algorithm of a Word2Vec model, which is only an example, but is not limited thereto.

Here, if one sentence data in the medical text is composed of n words, the processor 110 may perform a one-hot-vector for each word based on a position in the sentence data.

In addition, the processor 110 outputs one one-hot vector corresponding to the central word as a plurality of one-hot vectors corresponding to neighboring words through a projection layer, and converts each output data into soft It is converted using a softmax algorithm, and an error between the converted output data and real data can be calculated using a cross-entropy algorithm.

Here, the processor 110 may minimize an error between output data and actual data by utilizing a gradient descent algorithm.

Next, the processor 110 generates a plurality of clusters by clustering the words through a k-means clustering algorithm based on the location information of the embedded words, and generates word data that does not belong to the generated clusters. If exists, the corresponding word data may be regarded as noise word data.

Here, the processor 110 may regard words included in the cluster as noise word data if a cluster consisting of less than k words, which is the minimum number of words, exists among the generated clusters.

In addition, when a plurality of clusters are generated by clustering words, the processor 110 reclassifies the plurality of clusters based on treatment features, identifies noise word data from the reclassified treatment feature-based clusters, and identifies the identified treatment features. Feature-based noise word data may be stored in a noise dictionary.

Here, the processor 110 may reclassify clusters based on prescription information among treatment characteristics, identify noise word data based on prescription information from the reclassified clusters, and store the noise word data in a noise dictionary.

For example, the prescription information may include at least one of a disease name and a medication guide, which is only an example, but is not limited thereto.

In some cases, the processor 110 may reclassify clusters based on patient information among treatment characteristics, identify noise word data based on patient information from the reclassified clusters, and store them in a noise dictionary.

For example, patient information may include at least one of age, gender, and region of residence, which is only an example, but is not limited thereto.

Next, the processor 110 may generate a general noise filter from general information-based noise word data stored in the noise dictionary, and may generate a diagnosis feature-based noise filter from treatment feature-based noise word data stored in the noise dictionary.

Here, the processor 110 may generate a treatment feature-based noise filter including a prescription information-based noise filter and a patient information noise filter when generating a treatment feature-based noise filter.

Then, when the new medical text is generated, the processor 110 removes noise words included in each sentence of the medical text based on a noise filter corresponding to each sentence of the new medical text, and outputs the medical text from which the noise words have been removed. can be reconstructed.

Here, if the sentence of the new medical text is a sentence related to general information, the processor 110 removes noise words included in the sentence related to general information of the medical text based on the general noise filter, and the sentence of the new medical text is related to treatment characteristics. If it is a sentence, noise words included in the treatment feature sentence of the medical text may be removed based on the treatment feature-based noise filter.

In some cases, if the sentence of the new medical text is a prescription information sentence, the processor 110 removes noise words included in the prescription information sentence of the medical text based on a prescription information-based noise filter, and the sentence of the new medical text is a patient information sentence. If it is an information sentence, noise words included in the patient information sentence of the medical text may be removed based on the patient information-based noise filter.

According to an embodiment of the present invention, the processor 110 may be composed of one or more cores, a central processing unit (CPU) of a computing device, a general purpose graphics processing unit (GPGPU) ), a processor for data analysis and deep learning, such as a tensor processing unit (TPU). The processor 110 may read a computer program stored in the memory 130 and perform data processing for machine learning according to an embodiment of the present invention. According to an embodiment of the present invention, the processor 110 may perform an operation for learning a neural network. The processor 110 performs neural network learning, such as processing input data for learning in deep learning (DL), extracting features from input data, calculating errors, and updating neural network weights using backpropagation. calculations can be performed for At least one of the CPU, GPGPU, and TPU of the processor 110 may process learning of the network function. For example, the CPU and GPGPU can process learning of network functions and data classification using network functions. In addition, in one embodiment of the present invention, the learning of a network function and data classification using a network function may be processed by using processors of a plurality of computing devices together. In addition, a computer program executed in a computing device according to an embodiment of the present invention may be a CPU, GPGPU or TPU executable program.

According to an embodiment of the present invention, the memory 130 may store a computer program for performing a method of filtering noise data of medical text, and the stored computer program may be read and driven by the processor 120 . The memory 130 may store any type of information generated or determined by the processor 110 and any type of information received by the network unit 150 .

According to an embodiment of the present invention, the memory 130 is a flash memory type, a hard disk type, a multimedia card micro type, or a card type memory (eg SD or XD memory, etc.), RAM (Random Access Memory, RAM), SRAM (Static Random Access Memory), ROM (Read-Only Memory, ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Memory) Read-Only Memory), a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium. The computing device 100 may operate in relation to a web storage that performs a storage function of the memory 130 on the Internet. The description of the above memory is only an example, and is not limited thereto.

The network unit 150 according to an embodiment of the present invention may transmit/receive result information of a method for filtering noise data of medical text, etc. to another computing device or server. In addition, the network unit 150 enables communication between a plurality of computing devices so that operations for filtering noise data of medical text or learning a model may be performed in a distributed manner in each of the plurality of computing devices. The network unit 150 may enable communication between a plurality of computing devices to perform distributed processing of filtering noise data of medical text or calculation for learning a model using a network function.

The network unit 150 according to an embodiment of the present invention may operate based on any type of wired or wireless communication technology currently used and implemented, such as short-distance (short-distance), long-distance, wired, and wireless, and other networks. can also be used in

The computing device 100 of the present invention may further include an output unit and an input unit.

The output unit according to an embodiment of the present invention may display a user interface (UI) for performing a method of filtering noise data of medical text. The output unit may output any type of information generated or determined by the processor 110 and any type of information received by the network unit 150 .

In one embodiment of the present invention, the output unit is a liquid crystal display (liquid crystal display, LCD), thin film transistor liquid crystal display (thin film transistor-liquid crystal display, TFT LCD), organic light-emitting diode (organic light-emitting diode, OLED) , a flexible display, and a 3D display. Some of these display modules may be of a transparent type or a light transmissive type so that the outside can be seen through them. This may be referred to as a transparent display module, and a representative example of the transparent display module is TOLED (Transparent OLED) and the like.

The input unit according to an embodiment of the present invention may receive a user input. The input unit may include keys and/or buttons on a user interface for receiving user input, or physical keys and/or buttons. A computer program for controlling a display according to embodiments of the present invention may be executed according to a user input through an input unit.

The input unit according to embodiments of the present invention may detect a user's button operation or touch input and receive a signal, or may receive a user's voice or motion through a camera or microphone and convert it into an input signal. For this purpose, speech recognition technology or motion recognition technology may be used.

The input unit according to embodiments of the present invention may be implemented as an external input device connected to the computing device 100 . For example, the input device may be at least one of a touch pad, a touch pen, a keyboard, or a mouse for receiving a user input, but this is only an example and is not limited thereto.

The input unit according to an embodiment of the present invention may recognize a user touch input. The input unit according to an embodiment of the present invention may have the same configuration as the output unit. The input unit may include a touch screen implemented to receive a user's selection input. The touch screen may use any one of a contact capacitive method, an infrared light sensing method, a surface ultrasonic (SAW) method, a piezoelectric method, and a resistive film method. Detailed description of the touch screen described above is only an example according to an embodiment of the present invention, and various touch screen panels may be employed in the computing device 100 . The input unit configured as a touch screen may include a touch sensor. The touch sensor may be configured to convert a change in pressure applied to a specific portion of the input unit or capacitance generated at a specific portion of the input unit into an electrical input signal. The touch sensor may be configured to detect not only the touched position and area, but also the pressure upon touch. When there is a touch input to the touch sensor, the corresponding signal(s) is sent to the touch controller. The touch controller may process the signal(s) and then transmit corresponding data to processor 110 . Accordingly, the processor 110 can recognize which area of the input unit has been touched.

In one embodiment of the present invention, the server may include other configurations for performing the server environment of the server. The server may include any type of device. The server may be a digital device, such as a laptop computer, a notebook computer, a desktop computer, a web pad, or a mobile phone, equipped with a processor and having an arithmetic capability with a memory.

A server (not shown) performing an operation for providing a user interface displaying a filtering result of noise data of medical text according to an embodiment of the present invention to a user terminal may include a network unit, a processor, and a memory.

The server may generate a user interface according to embodiments of the present invention. The server may be a computing system that provides information to clients (eg, user terminals) over a network. The server may transmit the generated user interface to the user terminal. In this case, the user terminal may be any type of computing device 100 capable of accessing the server. The processor of the server may transmit the user interface to the user terminal through the network unit. A server according to embodiments of the present invention may be, for example, a cloud server. The server may be a web server that processes services. The types of servers described above are examples only and are not limited thereto.

In this way, the present invention embeds and clusters medical texts obtained by converting voice data corresponding to telemedicine conversations into text data by word, identifies noise word data, and filters new text data based on the identified noise word data. By doing so, it is possible to provide medical texts with high accuracy and reliability.

2 to 7, the present invention relates to a method for correcting medical text.

The medical text of the present invention may include data extracted in a remote medical treatment process.

Here, telemedicine is conducted in the form of a web or app on a mobile device, and conversation contents during medical treatment can be extracted in a text format through speech-to-text (STT).

At this time, since the accuracy of the extracted text data depends on the STT technology, when text data is extracted from voice data using the STT technology, inaccurate text data may be extracted if there is a problem in the STT technology itself.

Therefore, unless the STT itself is developed and learned, the extracted text data is dependent on the existing technology. Therefore, the present invention is a method for improving the accuracy of the text data by additionally correcting the extracted text while using the existing STT technology.

Here, the accuracy means the degree to which the extracted text data includes the contextual meaning of the voice data, not the degree of error that exists between the voice data and the extracted text data.

For example, assuming that the text data extracted through the voice data 'Kanada' is 'Kanada', and the voice data 'Kanada' is v1 and 'Kanada' is t1, the degree of error between the two data is It can be expressed as dist(v1, t1), which is an edit distance.

In this method, while there is a clear error value depending on the difference in text, in the context difference, the contextual similarity of the two data (similarity is assumed to be cosine similarity) sim(v1, t1) is calculated and the result is high. , can be assumed to be the correct extraction.

Therefore, according to the present invention, contextual similarity between extracted text data can be derived, and meaningful words and nonsensical words can be distinguished by utilizing this information.

And, the present invention is a method of filtering meaningless words using such a classification.

As shown in FIG. 2 , the present invention may proceed through three steps of text embedding, text clustering and classification, and constructing outliers that are noise data filters.

In the present invention, as a text embedding step, medical text may be generated by converting voice data corresponding to a telemedicine conversation into text data, and the medical text may be embedded word by word in each sentence.

Next, as a text clustering and classification step, noise word data may be identified by clustering embedded words.

Next, in the present invention, as an outlier constructing step, the identified noise word data may be stored in a noise dictionary, and a noise filter may be generated based on the noise word data stored in the noise dictionary.

Also, according to the present invention, when a new medical text is input, the new medical text is filtered through a noise filter to reconstruct the medical text from which noise word data is removed.

Meanwhile, in the text embedding step, medical text is generated by converting speech data corresponding to a telemedicine conversation into text data through STT (Speech-to-Text), and the sentence data of the medical text is pretrained with a neural network model. By inputting into , it is possible to embed the input sentence data for each word.

For example, the neural network model may include a skip-gram algorithm of a Word2Vec model.

3 shows a process of converting text data into a one-hot-vector to apply a skip-gram.

Here, as shown in FIG. 3, in the present invention, if one sentence data in the medical text is composed of n words, a one-hot-vector is performed for each word based on the position in the sentence data. can

For example, in a sentence with n words, if w1 is the first word in the sentence, then the one-hot-vector for w1 is n-dimensional, where only the first position is 1 and the rest are 0. may be a vector of

In addition, FIG. 4 shows a process of performing word-by-word embedding of an input sentence using a skip-gram.

That is, embedding can be performed by learning the relationship between the first word and the remaining words in the structure shown in FIG. 4 .

Here, in the present invention, as shown in FIG. 4, one one-hot-vector corresponding to the central word is output as a plurality of one-hot-vectors corresponding to neighboring words through a projection layer, and each output Data may be converted using a softmax algorithm, and an error between converted output data and real data may be calculated using a cross-entropy algorithm.

Then, in the text clustering and classification step, multiple clusters are generated by clustering words through a k-means clustering algorithm based on the location information of the embedded words, and words that do not belong to the generated clusters are generated. If data exists, corresponding word data may be regarded as noise word data.

That is, according to the present invention, text data in units of all words can be expressed as vectors, and word clustering is possible based on each word vector.

Figure 5(1) is an example showing word vectors expressed on a two-dimensional coordinate plane, and Figure 5(2) shows k-means clustering based on word location information as shown in Figure 5(1). An example of the result of applying the algorithm is shown.

In (2) of FIG. 5, it is largely composed of two clusters, and three word data can be regarded as noise word data that does not belong to any cluster.

Here, in the present invention, if a cluster consisting of less than k words, which is the minimum number of words, exists among the generated clusters, words included in the corresponding cluster may be regarded as noise word data.

In addition, the present invention, when a plurality of clusters are generated by clustering words, reclassifies the plurality of clusters based on treatment characteristics, identifies noise word data from the reclassified treatment feature-based clusters, and identifies noise word data based on the identified treatment characteristics. Noise word data can be stored in a noise dictionary.

Here, according to the present invention, clusters may be reclassified based on prescription information among treatment characteristics, and noise word data based on prescription information may be identified and stored in a noise dictionary from the reclassified clusters.

In some cases, the present invention may reclassify clusters based on patient information among treatment characteristics, identify noise word data based on patient information from the reclassified clusters, and store them in a noise dictionary.

Next, in the outlier construction step, words stored in the noise dictionary may be assumed as outliers.

Further, in the present invention, words of the medical text newly input through the STT may be filtered by first using the noise dictionary.

5 shows an example of filtering noise data of a new medical text input using an outlier, which is a noise filter configured through a noise dictionary.

As shown in FIG. 5 , sentence k (sentence_k) is an example of text data newly derived through STT. After removing noise word data w_15 and w_17 included in sentence k (sentence_k), sentence k (sentence_k) may be reconstructed. can

Here, the reconstructed sentence k may be assumed to be a sentence in which outlier words, which are noise word data, are removed.

As in the present invention, an advantage of selecting outliers through Word2vec and clustering is that words accidentally or incorrectly extracted during text data conversion can be detected.

As a result, words that are accidentally converted or incorrectly converted will have a lower frequency than words that are not, and the probability of being located in various sentences is also reduced.

Therefore, when these words are selected as outliers, the possibility of detecting incorrect words in the future increases.

Also, according to the present invention, an outlier may be configured based on classification information.

That is, in the present invention, when noise word data is identified based on information classified as treatment characteristics, a noise dictionary can be derived for each classification information.

In addition, it is possible to detect outliers suitable for telemedicine based on each noise dictionary.

6 shows an example of an outlier configuration based on classification information.

6 shows classification clusters for prescription A, and sentence k (sentence_k) is a word unit set of input sentences for prescription A.

That is, this means a sentence classified as prescription A among sentences input through the STT.

Then, outlier detection may be performed on the corresponding sentence.

Accordingly, the present invention may generate a general noise filter from general information-based noise word data stored in the noise dictionary, and a treatment feature-based noise filter from treatment feature-based noise word data stored in the noise dictionary.

Here, when generating a noise filter based on a treatment feature, the present invention may generate a noise filter based on a treatment feature including a noise filter based on prescription information and a noise filter based on patient information.

In addition, in the present invention, when a new medical text is generated, noise words included in each sentence of the medical text are removed based on a noise filter corresponding to each sentence of the new medical text, and the medical text from which the noise words are removed is reconstructed. can

Here, in the present invention, if the sentence of the new medical text is a sentence related to general information, noise words included in the sentence related to general information of the medical text are removed based on the general noise filter, and if the sentence of the new medical text is a sentence related to treatment characteristics, Noise words included in the treatment feature sentences of the medical text may be removed based on the treatment feature-based noise filter.

For example, in the present invention, if the sentence of the new medical text is a prescription information sentence, noise words included in the prescription information sentence of the medical text are removed based on the prescription information-based noise filter, and if the sentence of the new medical text is a patient information sentence, Noise words included in patient information sentences of the medical text may be removed based on the patient information-based noise filter.

As shown in FIG. 8 , according to the present invention, medical text may be generated by converting voice data corresponding to a telemedicine conversation into text data (S10).

Subsequently, the present invention may embed medical text for each word in each sentence (S20).

Next, the present invention may cluster the embedded words to identify noise word data, store the identified noise word data in a noise dictionary, and generate a noise filter based on the noise word data stored in the noise dictionary (S30). ).

Here, the present invention generates a plurality of clusters by clustering words through a k-means clustering algorithm based on positional information of embedded words, and word data that does not belong to the generated clusters exists. Then, corresponding word data may be regarded as noise word data.

In addition, the present invention may reclassify a plurality of clusters based on treatment characteristics, identify noise word data from the reclassified treatment feature-based clusters, and store the identified treatment feature-based noise word data in a noise dictionary. .

In addition, the present invention may generate a general noise filter from general information-based noise word data stored in the noise dictionary, and a treatment feature-based noise filter from treatment feature-based noise word data stored in the noise dictionary.

And, according to the present invention, a new medical text corresponding to the telemedicine conversation may be generated (S40).

Subsequently, when a new medical text is generated, the present invention filters the new medical text through a noise filter to reconstruct the medical text from which noise word data is removed (S50).

Here, in the present invention, when a new medical text is generated, noise words included in each sentence of the medical text are removed based on a noise filter corresponding to each sentence of the new medical text, and the medical text from which the noise words are removed is reconstructed. can

For example, in the present invention, if a sentence of the new medical text is a sentence related to general information, noise words included in the sentence related to general information of the medical text are removed based on a general noise filter, and the sentence of the new medical text is a sentence related to treatment characteristics. Noise words included in treatment feature sentences of the medical text can be removed based on the background treatment feature-based noise filter.

The method according to an embodiment of the present invention described above may be implemented as a program (or application) to be executed in combination with a server, which is hardware, and stored in a medium.

The aforementioned program is C, C++, JAVA, machine language, etc. It may include a code coded in a computer language of. These codes may include functional codes related to functions defining necessary functions for executing the methods, and include control codes related to execution procedures necessary for the processor of the computer to execute the functions according to a predetermined procedure. can do. In addition, these codes may further include memory reference related codes for which location (address address) of the computer's internal or external memory should be referenced for additional information or media required for the computer's processor to execute the functions. there is. In addition, when the processor of the computer needs to communicate with any other remote computer or server in order to execute the functions, the code uses the computer's communication module to determine how to communicate with any other remote computer or server. It may further include communication-related codes for whether to communicate, what kind of information or media to transmit/receive during communication, and the like.

The storage medium is not a medium that stores data for a short moment, such as a register, cache, or memory, but a medium that stores data semi-permanently and is readable by a device. Specifically, examples of the storage medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc., but are not limited thereto. That is, the program may be stored in various recording media on various servers accessible by the computer or various recording media on the user's computer. In addition, the medium may be distributed to computer systems connected through a network, and computer readable codes may be stored in a distributed manner.

Steps of a method or algorithm described in connection with an embodiment of the present invention may be implemented directly in hardware, implemented in a software module executed by hardware, or implemented by a combination thereof. A software module may include random access memory (RAM), read only memory (ROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, hard disk, removable disk, CD-ROM, or It may reside in any form of computer readable recording medium well known in the art to which the present invention pertains.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art to which the present invention pertains can be implemented in other specific forms without changing the technical spirit or essential features of the present invention. you will be able to understand Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive.

Claims

In the method performed by the device,

(a) generating medical text by converting voice data corresponding to the telemedicine conversation into text data;

(b) embedding the medical text for each word in each sentence;

(c) clustering the embedded words to identify noise word data, and storing the identified noise word data in a noise dictionary;

(d) generating a noise filter based on noise word data stored in the noise dictionary;

(e) checking whether a new medical text corresponding to the telemedicine conversation is generated; and

(f) filtering the new medical text through the noise filter when the new medical text is generated and reconstructing the medical text from which noise word data is removed.
According to claim 1,

In step (c),

Based on the location information of the embedded words, words are clustered through a k-means clustering algorithm to generate multiple clusters, and if there is word data that does not belong to the generated clusters, the corresponding word A method for filtering noise data in medical text, characterized in that the data is regarded as noise word data.
According to claim 2,

In step (c),

The method of filtering noise data of medical text, characterized in that, if a cluster consisting of less than k words, which is the minimum number of words, among the generated clusters exists, words included in the cluster are regarded as noise word data.
According to claim 2,

In step (c),

When a plurality of clusters are generated by clustering the words, the plurality of clusters are reclassified based on treatment characteristics, noise word data is identified from the reclassified treatment feature-based clusters, and the identified treatment feature-based noise word A method for filtering noise data in medical text, characterized in that the data is stored in a noise dictionary.
According to claim 1,

In step (d),

A method for filtering noise data of medical text, characterized in that generating a general noise filter from general information-based noise word data stored in the noise dictionary, and generating a treatment feature-based noise filter from treatment feature-based noise word data stored in the noise dictionary. .
According to claim 5,

In step (d),

The noise data filtering method of medical text, characterized in that when generating the treatment feature-based noise filter, a treatment feature-based noise filter including a prescription information-based noise filter and a patient information noise filter is generated.
According to claim 1,

In step (f),

When the new medical text is generated, noise words included in each sentence of the medical text are removed based on a noise filter corresponding to each sentence of the new medical text, and the medical text from which the noise words are removed is reconstructed. A method for filtering noise data in medical texts with .
According to claim 7,

In step (f),

If the sentence of the new medical text is a sentence related to general information, noise words included in the sentence related to general information of the medical text are removed based on a general noise filter, and if the sentence of the new medical text is a sentence related to treatment characteristics, based on the treatment feature A method for filtering noise data of medical text, characterized in that, based on a noise filter, noise words included in medical treatment characteristic sentences of the medical text are removed.
A method of filtering noise data of medical text of a device for filtering noise data of medical text stored in a medium to perform the method of filtering noise data of medical text according to any one of claims 1 to 8, combined with a computer that is hardware. computer program provided.
A computing device for providing a method for filtering noise data of medical text,

a processor comprising one or more cores; and

Memory;

including,

the processor,

Converting voice data corresponding to a telemedicine conversation into text data to generate medical text;

Embedding the medical text for each word in each sentence;

Clustering the embedded words to identify noise word data and storing the identified noise word data in a noise dictionary;

generating a noise filter based on noise word data stored in the noise dictionary;

Check whether a new medical text corresponding to the telemedicine conversation is generated; and

When the new medical text is generated, the new medical text is filtered through the noise filter to reconstruct the medical text from which noise word data is removed.