CN114387602A - Medical OCR data optimization model training method, optimization method and equipment - Google Patents

Abstract

The invention discloses a medical OCR data optimization model training method, an optimization method and equipment, wherein the training method comprises the following steps: acquiring large-scale label-free medical text data, and identifying medical terms and characters in the text data to form a training set; pre-training the training set to obtain a pre-training data set for training the medical OCR optimization model, and training the medical OCR optimization model by using the pre-training data set; the pre-training process comprises: carrying out data augmentation processing on low-frequency terms and low-frequency characters in the training set; randomly replacing a first target character in a training set with an error character; shielding a second target character in the training set; and segmenting the training set into a plurality of text paragraphs to obtain a pre-training data set for training the medical OCR optimization model. According to the invention, the pre-training language model in the medical field is utilized to perform structured extraction, error recognition and optimization on the medical OCR result, so that the accuracy of the medical OCR is improved.

Description

Medical OCR data optimization model training method, optimization method and equipment

technical field

The invention relates to the technical field of intelligent medical data processing, in particular to a medical OCR data optimization model training method, an optimization method, and related electronic equipment and computer-readable storage media.

Background technique

With the rapid development of machine learning, Optical Character Recognition (OCR) has made great progress in character recognition, and there have been various commercial applications such as Baidu OCR. In the medical field, paper data must be structured for clinical medical research, medical record structuring, and underwriting claims. How to convert paper medical data into computer-processable structured data has become the key to the development of intelligent medical care. The structuring of medical image data also requires optical recognition, and the result of the recognition determines the subsequent process. However, there are still many problems with the accuracy of optical text recognition in the medical field. Different from image text recognition in the general field, medical image text recognition contains a large number of medical professional terms, such as the names of diseases and field names in medical records, and the term base is large, with more than 1 million commonly used medical terms. Moreover, the medical field contains a large number of rare and very used characters, which appear very infrequently in general texts, and medical terms such as rare diseases and other non-rare words appear in the corpus very low frequency, such as "Kawasaki" Diseases", these low-frequency terms have low recognition accuracy (eg "lid" is often recognized as "face"). In addition, characters commonly used in the medical field refer to various diseases and states, etc., but the frequency of these characters is low, so it is difficult for the OCR character recognition model to accurately recognize these characters. In terms of the structure of medical records, different from ordinary text materials, medical data usually has a specific structure, such as the information of multiple fields contained in the medical record report, and the data types contained in different fields are different. However, the current text recognition system lacks structural information. use. The language style of texts in the medical field is very concise, and medical personnel often omit a large number of non-medical words when forming documents. All of the above aspects pose new challenges to OCR and existing post-processing models.

SUMMARY OF THE INVENTION

In order to solve the problem in the prior art that OCR cannot accurately identify abnormal or incorrect data, the present invention provides the following technical solutions.

The present invention provides a medical OCR optimization model training method in a first aspect, including:

acquiring large-scale unlabeled medical text data, and identifying medical terms and characters in the large-scale unlabeled medical text data to form a training set;

Performing pre-training processing on the training set to obtain a pre-training data set for training the medical OCR optimization model, and using the pre-training data set to train the medical OCR optimization model;

Wherein, the pre-training process includes:

performing data augmentation processing on the low-frequency terms and low-frequency characters in the training set;

Randomly replace the first target character in the training set with an incorrect character;

occluding the second target character in the training set; and

The training set is divided into multiple text paragraphs to obtain a pre-training data set for training the medical OCR optimization model.

Preferably, before the data augmentation processing is performed on the low-frequency terms and low-frequency characters in the training set, the method further includes:

The frequency of each medical term and character in the training set identified is counted, and the low-frequency term and low-frequency character in the training set are determined according to the corresponding low-frequency threshold.

Preferably, after forming the training set, it further includes:

Representation learning of medical terms is performed on the training set using the medical knowledge graph, and mapping is performed in the representation space.

Preferably, the randomly replacing the first target character in the training set with an error character further includes:

First target characters are screened from medical terms and characters in the training set, wherein the first target characters include characters contained in a glyph similarity dictionary and/or commonly used medical characters.

Preferably, the use of the pre-training data set to train the medical OCR optimization model further includes:

After the first target characters have been randomly replaced with wrong characters, the current training set is taken as the first data set, the wrong characters in the first data set are iteratively extracted according to the current context, and predictions that match the wrong characters are performed. The first target character corresponding to the character is used to train the character error correction capability of the medical OCR optimization model.

Preferably, the use of the pre-training data set to train the medical OCR optimization model further includes:

After the second target character has been occluded, using the current training set as the second data set, iteratively predicts the second target character corresponding to the occluded position in the second data set according to the current context to train the The ability of the medical OCR optimization model to identify occluded content.

Preferably, the use of the pre-training data set to train the medical OCR optimization model further includes:

Iteratively predicts end-of-paragraph sentences in the pre-trained dataset based on the current context to train the medical OCR optimization model's ability to automatically segment.

The present invention provides a medical OCR data optimization method in a second aspect, comprising:

Obtain the target medical image, perform OCR recognition on the target medical image, and obtain the text data to be optimized;

Inputting the text data to be optimized into a medical OCR optimization model, so that the medical OCR optimization model outputs medical terms and character recognition results corresponding to the text data to be optimized;

Wherein, the medical OCR optimization model is obtained in advance based on the medical OCR optimization model training method described in the first aspect.

Another aspect of the present invention provides an electronic device, comprising a processor and a memory, the memory stores a plurality of instructions, the processor is configured to read the instructions and execute the medical OCR data optimization model described in the first aspect training method, or executing the medical OCR data optimization method based on the second aspect.

Yet another aspect of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores a plurality of instructions, and the plurality of instructions can be read by a processor to execute the medical OCR data described in the first aspect The model training method is optimized, or the medical OCR data optimization method based on the second aspect is executed.

The beneficial effects of the present invention are: on the basis of data enhancement, the technical solution of the present invention uses a pre-trained language model in the medical field to perform structured extraction, error recognition and optimization of medical OCR results, thereby improving the accuracy of medical image text recognition, In particular, the recognition accuracy of key words such as medical terminology and medical record keywords is improved, and at the same time, text paragraphs can be subdivided for subsequent medical knowledge extraction and event extraction.

Description of drawings

FIG. 1 is a flowchart of the medical OCR optimization model training method according to the present invention.

FIG. 2 is a schematic diagram of the formation process of the pre-training data set used for model training according to the present invention.

FIG. 3 is a schematic diagram of the training process of the pre-trained language model for post-processing of text recognition according to the present invention.

FIG. 4 is a flow chart of the medical OCR data optimization method according to the present invention.

FIG. 5 is a detailed flowchart of the method for recognizing text in medical images according to the present invention.

Detailed ways

In order to better understand the above technical solutions, the above technical solutions will be described in detail below with reference to the accompanying drawings and specific embodiments.

The method provided by the present invention may be implemented in the following terminal environment, and the terminal may include one or more of the following components: a processor, a memory and a display screen. Wherein, at least one instruction is stored in the memory, and the instruction is loaded and executed by the processor to implement the methods described in the following embodiments.

A processor may include one or more processing cores. The processor uses various interfaces and lines to connect various parts of the entire terminal, and executes various functions of the terminal by running or executing the instructions, programs, code sets or instruction sets stored in the memory, and calling the data stored in the memory. and processing data.

The memory may include random access memory (Random Access Memory, RAM), or may include read-only memory (Read-Only Memory, ROM). Memory may be used to store instructions, programs, codes, sets of codes, or instructions.

The display is used to display the user interface of each application.

In addition, those skilled in the art can understand that the structure of the above-mentioned terminal does not constitute a limitation on the terminal, and the terminal may include more or less components, or combine some components, or arrange different components. For example, the terminal also includes components such as a radio frequency circuit, an input unit, a sensor, an audio circuit, and a power supply, which will not be repeated here.

In view of the above problems, in order to optimize the results of medical image OCR, the present invention proposes a medical OCR optimization model training method, and a medical image text recognition (OCR) optimization method based on a pre-trained language model. The method utilizes medical pre-training The language model post-processes the results of OCR. Specifically, a preset training model is used to identify the wrong characters in the text obtained by OCR, and predict the correct characters to obtain the correct text recognition result. On the basis of data enhancement, the present invention utilizes a pre-trained language model in the medical field to perform structured extraction, error identification and optimization of medical OCR results, aiming to improve the accuracy of medical OCR.

Example 1

As shown in FIG. 1, an embodiment of the present invention provides a medical OCR optimization model training method, including:

S101. Obtain large-scale unlabeled medical text data, and identify medical terms and characters in the large-scale unlabeled medical text data to form a training set;

Among them, the original training data of the pre-trained language model is large-scale medical text data, which comes from text data in clinical guidelines, medical textbooks, medical encyclopedias, medical forums and accumulated medical records. Based on the large-scale unlabeled medical text data, the existing medical named entity recognition model is used for recognition processing to obtain preliminary medical terminology and character recognition results, such as disease, diagnosis, surgery, medicine, medical record keywords (main complaint, ultrasound diagnosis) etc.), the specific term recognition model can adopt deep learning-based entity recognition, statistical learning-based entity recognition or dictionary-based methods.

S102. Perform pre-training processing on the training set to obtain a pre-training data set for training the medical OCR optimization model, and use the pre-training data set to train the medical OCR optimization model.

In order to improve the accuracy of medical image text recognition, the present invention is mainly aimed at the pre-trained language model for text recognition post-processing, and optimizes the text recognition results to obtain a model with better processing capability for medical image text recognition. In a preferred embodiment, the better processing capability can be reflected in that the model can correctly identify and correct errors in character recognition, supplement missing characters, and divide long texts into reasonable paragraphs. Therefore, in step S102, pre-training is performed on the training set obtained in step S101 to obtain a new training set that enhances the above identification processing capability, and the original medical OCR optimization model is trained according to the new training set.

In a further preferred embodiment, representation learning of medical terms can be performed on the training set by using a medical knowledge graph, and mapping is performed in the representation space. Specifically, dictionaries and models can be used to label large-scale unlabeled medical text data, and medical knowledge graphs and labeled texts can be used to perform term representation learning. The knowledge graph can use a graph neural network (GNN), and the text can use Transformer for representation learning. And map in the representation space, so that the data learned by different modalities have the same vector representation.

Wherein, the pre-training process may include one or more of the following aspects, as shown in FIG. 2:

S1021. Perform data augmentation processing on the low-frequency terms and low-frequency characters in the training set.

For the recognition of uncommon words, the present invention utilizes paraphrasing generation to realize data augmentation, that is, increasing the word frequency of low-frequency words, so as to better meet the training requirements. The specific method can choose the sentence-level paraphrase generation method, which can be combined with the seq2seq model, that is, directly copy and paste the sentences with low-frequency terms and low-frequency characters. In specific operations, statistical analysis can be performed on medical texts and terms to obtain the frequency of each medical term and character in the training set. The low-frequency terms and low-frequency characters in the training set are determined according to the corresponding low-frequency thresholds, such as terms that appear less than 20 times, or characters that appear less than 5 times. This part of the data is augmented to enhance this type of text data to obtain more balanced data.

S1022. Randomly replace the first target character in the training set with an error character.

Specifically, the first target characters may be screened from the recognized medical terms and characters in the training set, where the first target characters may include characters contained in a glyph similarity dictionary and/or commonly used medical characters. Since OCR recognition is based on image recognition, characters with similar glyphs are easily misidentified. Therefore, for the identification of abnormal characters, the present invention utilizes a glyph similarity dictionary to randomly replace the characters in the medical text with homographs, and replaces the correct characters of the homograph library with wrong glyphs in the training data. The glyph similarity dictionary contains common characters (such as "person" and "such as") as well as commonly used medical characters (such as "face" and "lid"). For example, the correct medical term "meiibomian gland" in the original training set can be used to determine the first target character "eyelid". The character "face" corresponding to the homograph is replaced by the wrong term "face gland". During language model training, the training set containing errors can be input into the pre-training model, and the model can be positively motivated to predict the character "face" that does not conform to the current context, and correspondingly predict the correct target character "eyelid", thereby improving The text error correction capability of the model.

In a further preferred embodiment, when the target characters to be replaced are randomly selected, the replacement frequency of commonly used medical characters can be increased compared with ordinary characters, so that the model can predict correct commonly used medical characters. Since the first target character can be more than one, that is, multiple error-prone characters can be randomly replaced, and the current training set can be used as the first data set, accordingly, during the training of the medical OCR optimization model, iterative Each erroneous character in the first data set is extracted according to the current context, and each correct first target character corresponding to each erroneous character is predicted, so as to train the character error correction ability of the medical OCR optimization model.

S1023, occlude the second target character in the training set.

Aiming at the problem of missing characters that often occurs after the medical image text recognition, the optimized language model of the present invention needs to have the ability to recognize the lack of words and characters. The second target character to be occluded can be screened from the recognized medical terms and characters in the training set according to a preset probability, and then the content of the occluded position is predicted during language model training. That is, after completing the occlusion operation of one or more characters, the current training set is used as the second data set. Accordingly, during the training of the medical OCR optimization model, each correct second target character corresponding to the occluded position in the second dataset is iteratively predicted from the current context to train the medical OCR optimization model to identify missing content capability.

Unlike common language model occlusion operations, the present invention preferably increases the probability of medical term vocabulary being occluded. In a preferred embodiment, the probability of random replacement of medical terms can be 3 times that of other terms, and at the same time, the characters in the medical terms are also partially occluded. The replacement frequency of a character can be inversely proportional to the number of times the character appears in the corpus, the lower the frequency the higher the probability of the character being replaced.

S1024. Divide the training set into multiple text paragraphs to obtain a pre-training data set for training the medical OCR optimization model.

In view of the structural characteristics of medical records, in the pre-training stage, the present invention uses keywords and language models to divide different text blocks, automatically divides medical data into multiple independent text blocks, and trains language models as separate tasks. Enables the model to have field partitioning capabilities. Correctly split paragraphs play an important role in subsequent medical knowledge extraction and event extraction. The method of segmentation can use pre-acquired keywords and the like. Different paragraphs in medical texts usually have large differences in content. The language model of the present invention performs paragraph segmentation by predicting whether the current sentence is the last sentence of the current paragraph. The prediction can be expanded based on the current text and the next sentence, and when there is no obvious semantic relationship between the two sentences, paragraph segmentation is performed. Among them, the prediction method of whether the current sentence is the last sentence of the current paragraph can also refer to formula (1) to predict the probability that the current sentence is correct according to the current context.

During the training of the medical OCR optimization model, one or more end-of-paragraph sentences in the pre-training dataset are iteratively predicted based on the current context to train the medical OCR optimization model's ability to automatically segment.

By performing the pre-training process of S1022, S1023 and S1024, a pre-training data set is obtained. Next, the medical OCR optimization model is iteratively trained and updated by using the obtained pre-training data set, and it can be obtained that the medical image text recognition has a Models for better processing power, including the ability to correctly identify and correct text recognition errors, missing characters, and optimal segmentation of text segments, respectively.

Figure 3 shows the complete training process of the pre-trained language model in the offline phase. It should be noted that, in the above process of the present invention, character error correction, character supplement and text segmentation are relatively independent model optimization processes. Therefore, in the pre-training process of practical application, selecting at least one of the above-mentioned pre-training methods can enhance the training data of the language model. Moreover, the sequence of steps S1022, S1023 and S1024 can be adjusted arbitrarily, and is not limited to the sequence described in the above embodiment. For example, the training set may be first divided into multiple text paragraphs, and then the preset target characters in the training set may be incorrectly replaced and/or occluded to obtain a pre-training data set, and so on.

In a further embodiment, for the identification of medical language styles, the pre-training process of the present invention may further include:

S1025 , extracting a large amount of diagnosis result texts to fine-tune the language model.

Model fine-tuning is performed by learning and training a large number of texts of diagnostic results, thereby enhancing the model's understanding of medical personnel's grammar habits and writing style, and improving the recognition accuracy of diagnostic results.

In addition, as a specific method for realizing sentence error detection, in the model training stage after the random replacement of wrong characters is completed in step S1022, the probability that the current sentence is a correct sentence can be estimated first, the wrong characters can be identified next, and the correct characters can be predicted according to the context. Calculate the probability that the sentence after error correction is the correct sentence. Specifically, the calculation method of sentence error detection can be expressed as formula (1):

P(s)=P(w ₁ , w ₂ ,w ₃ ,...,w _n ) Formula (1)

Where s is an OCR sentence, the sentence is composed of character sequences w ₁ , w ₂ , w ₃ ,..., _wn , and P(s) is the probability that the sentence is a correct sentence.

When the value of P(s) is less than a given threshold, it is determined to contain an error. That is, after identifying the wrong sentence and predicting the wrong character, the model predicts the wrong character as shown in formula (2):

P _error ( _wi )=minP( _wi |w ₁ ,..,wi _-1 ,wi ₊₁ ,..,w _n ) Formula (2)

where is the character w _i is the misidentified character and P( _wi |w ₁ ,..,wi _-1 ,w _i+1 ,..,w _n ) is the probability of w _i appearing in a given context , the character with the lowest probability in the sentence is the error character P _error ( _wi ). The way the model gives the correct character is shown in formula (3):

w'=maxP(w'|w ₁ ,w ₂ ,w ₃ ,...) Formula (3)

Where w' is the predicted correct character, P(w'|w ₁ ,w ₂ ,w ₃ ,...) Given the context, the probability that the character w' can constitute a reasonable sentence.

It can be seen that, compared with the prior art, the above method of the present invention can further improve the accuracy of text recognition in medical images, especially the recognition accuracy of key words such as medical terms, medical record keywords (main complaint, etc.), etc. At the same time, it can perform auxiliary segmentation on text paragraphs, which is beneficial to subsequent medical knowledge extraction and event extraction. The experimental data show that after using the medical OCR optimization model training method of the present invention, the text recognition error detection rate can reach 78%, and the correct character prediction accuracy rate can reach 85%, so the accuracy of medical image OCR recognition can be significantly optimized. The medical OCR optimization model training method of the present invention has important value for the structuring of medical records, clinical medical statistics, underwriting claims and other applications.

Embodiment 2

As shown in FIG. 4 , the present invention provides a medical OCR data optimization method in a second aspect, including:

S201, obtaining a target medical image, and performing OCR identification on the target medical image to obtain text data to be optimized;

The target medical image to be identified may include image files such as scanned or photographed medical records and diagnostic reports. After acquiring the target medical image, the image text data is extracted according to the existing OCR recognition algorithm as the initial text data.

S202. Input the text data to be optimized into a medical OCR optimization model, so that the medical OCR optimization model outputs medical terms and character recognition results corresponding to the text data to be optimized.

As mentioned above, before the OCR online identification, the final medical OCR optimization model has been obtained in advance based on the medical OCR optimization model training method of the first embodiment in the offline stage. The complete medical image OCR recognition method in the online stage is shown in Figure 5. Input the initial text data to be optimized into the medical OCR optimization model, so that the model outputs the corresponding optimized text data. Since the model is trained to have a higher ability to correct text recognition errors, identify missing characters, and optimize text segmentation, the optimized text data contains at least the segment segmentation of the initial text data, and there are errors in the initial text data In the case of medical terms or characters, identify erroneous medical terms or characters in the initial text data; then replace the erroneous terms with the corresponding correct element terms, and in the case of missing terms in the initial text data, predict the missing Medical term or character.

Embodiment 3

Another aspect of the present invention also includes a functional module architecture that is completely consistent with the aforementioned method flow, that is, an embodiment of the present invention also provides a medical OCR data optimization model training device, including:

The obtaining module 301 is used for obtaining large-scale unlabeled medical text data, and identifying medical terms and characters in the large-scale unlabeled medical text data to form a training set;

A pre-training module 302, configured to perform pre-training processing on the training set to obtain a pre-training data set for training the medical OCR optimization model, and use the pre-training data set to train the medical OCR optimization model ;

Wherein, the pre-training processing module includes:

Augmentation module 3021, configured to perform data augmentation processing on low-frequency terms and low-frequency characters in the training set;

Replacement module 3022, for randomly replacing the first target character in the training set with an incorrect character;

an occlusion module 3023, configured to occlude the second target character in the training set; and

The segmentation module 3024 is configured to segment the training set into multiple text paragraphs to obtain a pre-training data set for training the medical OCR optimization model.

The apparatus can be implemented by the medical OCR data optimization model training method provided in the first embodiment. For the specific implementation method, reference may be made to the description in the first embodiment, which will not be repeated here.

Embodiment 4

Correspondingly, the embodiment of the present invention also provides a medical OCR data optimization device, including:

The identification module 401 is used for acquiring the target medical image, and performing OCR identification on the target medical image to obtain text data to be optimized;

An optimization module 402, configured to input the text data to be optimized into a medical OCR optimization model, so that the medical OCR optimization model outputs medical terms and character recognition results corresponding to the text data to be optimized;

Wherein, the medical OCR optimization model is obtained in advance based on the medical OCR optimization model training method described in the first embodiment.

Embodiment 5

Another aspect of the present invention provides an electronic device, including a processor and a memory, the memory stores a plurality of instructions, and the processor is configured to read the instructions and execute the medical OCR data optimization model of the first embodiment training method, or execute the medical OCR data optimization method described in Embodiment 2. The processor and the memory may be connected by a bus or in other ways, and the connection by a bus is taken as an example. The processor may be a central processing unit (Central Processing Unit, CPU). The processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other Chips such as programming logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above types of chips.

As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs and modules, such as the medical OCR data optimization model training method and the optimization method in the embodiments of the present application corresponding to program instructions/modules. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory, that is, to implement the methods in the above method embodiments.

The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system and an application program required by at least one function; the storage data area may store data created by the processor, and the like. Additionally, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory may optionally include memory located remotely from the processor, such remote memory being connectable to the processor via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

Embodiment 6

Another aspect of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores a plurality of instructions, and the plurality of instructions can be read by a processor to execute the medical OCR data described in the first embodiment Optimize the model training method, or execute the medical OCR data optimization method described in Embodiment 2. The computer-readable storage medium may be a tangible storage medium such as random access memory (RAM), memory, read only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

Although the preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiment and all changes and modifications that fall within the scope of the present invention. It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit and scope of the invention. Thus, provided that these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (10)

1. a medical OCR optimization model training method, is characterized in that, comprises: acquiring large-scale unlabeled medical text data, and identifying medical terms and characters in the large-scale unlabeled medical text data to form a training set; Performing pre-training processing on the training set to obtain a pre-training data set for training the medical OCR optimization model, and using the pre-training data set to train the medical OCR optimization model; Wherein, the pre-training process includes: performing data augmentation processing on the low-frequency terms and low-frequency characters in the training set; Randomly replace the first target character in the training set with an incorrect character; occluding the second target character in the training set; and The training set is divided into multiple text paragraphs to obtain a pre-training data set for training the medical OCR optimization model. 2. The method according to claim 1, wherein before the data augmentation processing is performed on the low-frequency terms and low-frequency characters in the training set, the method further comprises: The frequency of each medical term and character in the training set identified is counted, and the low-frequency term and low-frequency character in the training set are determined according to the corresponding low-frequency threshold. 3. The method according to claim 1, characterized in that, after said forming a training set, further comprising: Representation learning of medical terms is performed on the training set using the medical knowledge graph, and mapping is performed in the representation space. 4. The method according to claim 1, wherein the randomly replacing the first target character in the training set with an error character further comprises: First target characters are screened from medical terms and characters in the training set, wherein the first target characters include characters contained in a glyph similarity dictionary and/or commonly used medical characters. 5. The method according to claim 1 or 4, wherein the training of the medical OCR optimization model using the pre-training data set further comprises: After the first target characters have been randomly replaced with wrong characters, the current training set is taken as the first data set, the wrong characters in the first data set are iteratively extracted according to the current context, and predictions that match the wrong characters are performed. The first target character corresponding to the character is used to train the character error correction capability of the medical OCR optimization model. 6. The method according to claim 1, wherein the training of the medical OCR optimization model using the pre-training data set further comprises: After the second target character has been occluded, using the current training set as the second data set, iteratively predicts the second target character corresponding to the occluded position in the second data set according to the current context to train the The ability of the medical OCR optimization model to identify occluded content. 7. The method according to claim 1, wherein the training of the medical OCR optimization model using the pre-training data set further comprises: Iteratively predicts end-of-paragraph sentences in the pre-trained dataset based on the current context to train the medical OCR optimization model's ability to automatically segment. 8. A method for optimizing medical OCR data, comprising: Obtain the target medical image, perform OCR recognition on the target medical image, and obtain the text data to be optimized; Inputting the text data to be optimized into a medical OCR optimization model, so that the medical OCR optimization model outputs medical terms and character recognition results corresponding to the text data to be optimized; Wherein, the medical OCR optimization model is obtained in advance based on the medical OCR optimization model training method described in any one of claims 1 to 7. 9. An electronic device, comprising a processor and a memory, wherein the memory stores a plurality of instructions, and the processor is configured to read the instructions and execute the medical treatment according to any one of claims 1 to 7. The OCR data optimization model training method, or the medical OCR data optimization method as claimed in claim 8 is implemented. 10. A computer-readable storage medium, wherein the computer-readable storage medium stores a plurality of instructions, and the plurality of instructions can be read and executed by a processor according to any one of claims 1 to 7 The medical OCR data optimization model training method, or the medical OCR data optimization method according to claim 8 is executed.

Images