Automatic medical record classification method and system based on AI

Abstract

The invention belongs to the technical field of artificial intelligence, and particularly relates to an AI-based medical record automatic classification method and system, and the method comprises the steps: obtaining medical record data; performing cleaning and standardized preprocessing on the medical record data; performing natural language processing on the medical record data after standardization preprocessing to extract medical record text features; and inputting the medical record text features into a deep learning classification model to classify the medical record data, wherein the deep learning model is optimized by adopting a feedback optimization mechanism. According to the scheme of the invention, the classification accuracy is effectively improved, the rule does not need to be manually readjusted, the method has high intellectualization and self-adaptive capability, and manual intervention is reduced by adopting the deep learning model to automatically analyze and classify the medical record content.

Description

Technical Field

[0001] The present invention generally relates to the field of artificial intelligence technology. More specifically, the present invention relates to an AI-based automatic medical record classification method and system. Background Art

[0002] Automatic medical record classification is a crucial research area in healthcare informatization, contributing to the efficient management of hospital information systems, disease statistics, and analysis. However, existing medical record classification systems often rely on manual intervention or rule-based expert systems, which face numerous shortcomings.

[0003] Currently, existing technologies mainly rely on automatic classification methods based on keyword matching. By analyzing the text information in the medical records, keywords are extracted and classified based on the rule base. The advantage of these methods is that they are simple to implement, but their disadvantages are also very obvious:

[0004] 1. Low classification accuracy: Due to the rule-based or keyword matching, accurate classification often occurs. In particular, medical records may contain a large number of synonyms and ambiguous words, which traditional keyword matching methods often cannot handle.

[0005] 2. Lack of intelligence and adaptability: Existing classification methods mostly rely on manually defined rules, which are static and unable to adapt to new medical record data and changes in classification requirements. For example, when medical record content changes significantly, the system needs to readjust the rule base or re-annotate manually.

[0006] 3. Low computational efficiency: Since keyword matching cannot effectively process large-scale data, existing classification methods are often inefficient when faced with massive medical record data and cannot meet the needs of efficient processing.

[0007] Based on this, there is an urgent need for a method that can solve the current problems of low efficiency and poor accuracy in the classification and recognition of medical record information, so as to promote the intelligent development of the medical industry. Summary of the Invention

[0008] In order to solve the above-mentioned technical problems of low efficiency and poor accuracy in classification and recognition of medical record information, the present invention provides solutions in the following aspects.

[0009] In a first aspect, the present invention provides an AI-based automatic medical record classification method, comprising: acquiring medical record data; performing cleaning and standardization preprocessing on the medical record data; performing natural language processing on the standardized preprocessed medical record data to extract medical record text features; inputting the medical record text features into a deep learning classification model to classify the medical record data, wherein the deep learning model is optimized using a feedback optimization mechanism.

[0010] According to the solution of the present invention, through a text analysis model based on deep learning, important features are automatically extracted from the text information of medical records, solving the problems of low accuracy and inability to adapt to synonyms and ambiguous words in existing keyword matching methods. Utilizing natural language processing (NLP) and deep learning technology, the present invention can continuously optimize the classification model based on new medical record data without the need for manual readjustment of rules, and has a high degree of intelligence and adaptability. By using a deep learning model to automatically analyze and classify medical record content, manual intervention is reduced and computing efficiency is greatly improved, especially in large-scale data processing.

[0011] In one embodiment, data cleaning includes removing noise data and eliminating irrelevant information, and standardization preprocessing includes standardization processing and feature vectorization. The standardization processing includes converting unstructured text into structured data, and the feature vectorization includes converting the medical record text content into word vector representation.

[0012] In one embodiment, the deep learning classification model adopts a Transformer architecture based on an attention mechanism to capture text semantic associations.

[0013] In one embodiment, the deep learning classification model adopts an adaptive training mechanism, which includes incremental learning and active learning to achieve automatic updating of model parameters.

[0014] In one embodiment, the incremental learning includes: in response to the number of new medical record data input being greater than a set value, updating the model parameters using a linear gradient descent strategy.

[0015] In one embodiment, the calculation formula of the linear gradient descent strategy is:

[0016] η t =η0 / (1+γt)

[0017] Among them, η t is the current learning rate, η0 is the initial learning rate, γ is the decay coefficient, and t is the number of training rounds.

[0018] In one embodiment, active learning includes: selecting high-value unlabeled samples for training through uncertainty sampling.

[0019] In one embodiment, the feedback optimization mechanism includes: correcting the classification results according to user feedback data and generating corrected data; and adding the corrected data to the training set for incremental learning.

[0020] In one embodiment, the optimization formula of the loss function in the feedback optimization mechanism is:

[0021]

[0022] Where, is the new loss function, λ is the forgetting factor, and its value range is [0,1]. is the loss of the original training data, is the loss based on user correction data.

[0023] In a second aspect, the present invention also provides an AI-based automatic medical record classification system, comprising: a processor; a memory storing computer program instructions for automatic medical record classification, wherein when the computer program instructions are executed by the processor, the AI-based automatic medical record classification method according to one or more of the aforementioned embodiments is implemented.

[0024] The beneficial effects of the present invention are as follows: According to the solution of the present invention, through a deep learning-based text analysis model, important features are automatically extracted from the text information of medical records, effectively improving classification accuracy. Utilizing natural language processing (NLP) and deep learning technology, the present invention can continuously optimize the classification model based on new medical record data without the need for manual readjustment of rules, and has a high degree of intelligence and adaptability. By using a deep learning model to automatically analyze and classify medical record content, manual intervention is reduced and computing efficiency is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an illustrative and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0026] Figure 1 is a flow chart illustrating an AI-based automatic classification method for medical records according to an embodiment of the present invention;

[0027] Figure 2 2 is a block diagram showing the composition of an AI-based automatic medical record classification system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0029] The present invention provides a method for automatically classifying medical records using AI technology. This method can effectively overcome the shortcomings of existing technologies, improve the accuracy and efficiency of classification, and has adaptive and intelligent capabilities. The specific technical solution is as follows:

[0030] (1) Medical record data preprocessing: Clean and standardize the input medical record data, including removing noise data, eliminating irrelevant information, and extracting structured and unstructured information. Convert the medical record text content into a format that can be processed by the deep learning model.

[0031] (2) Deep learning-based text classification model: Using natural language processing (NLP) technology, feature information is extracted from medical records. Through these features, the model can automatically learn patterns in medical record data and perform high-precision classification.

[0032] (3) Training and Optimization: Optimize the parameters of the deep learning model through training with a large amount of labeled medical record data. This process uses an automated training strategy, and the model parameters are continuously optimized based on new medical record data to maintain a high level of accuracy.

[0033] (4) Automatic classification and feedback mechanism: The trained deep learning model is applied to the automatic classification of new medical records. Based on the classification results, the system can automatically assign medical record labels and adjust the classification strategy based on user feedback.

[0034] (5) System architecture: The entire system is deployed through a cloud platform to ensure real-time processing and analysis of large-scale data. Users can view classification results and provide feedback through the interface, and the feedback data is used to further optimize the model.

[0035] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0036] Figure 1 1 is a flowchart illustrating an AI-based automatic medical record classification method 100 according to an embodiment of the present invention. In the context of the present invention, the present invention is applicable to medical record data processing and classification in various medical institutions, such as hospitals and clinics, and can help medical personnel manage medical record data more efficiently and improve the quality of medical services.

[0037] like Figure 1 As shown, at step S101, medical record data is obtained. In some embodiments, the method of the present invention can be applied to different types of medical record data, including text medical records, image medical records, etc.

[0038] In step S102, the medical record data is cleaned and standardized. The input medical record data is cleaned and standardized, including removing noise data, eliminating irrelevant information, and extracting structured and unstructured information. The medical record text content is converted into a format that can be processed by the deep learning model.

[0039] In some embodiments, data cleaning includes removing noise data and eliminating irrelevant information. For example, special characters, stop words, and other noise data can be removed. For example, regular expressions can be used to remove punctuation and special characters from medical records.

[0040] Standardization preprocessing includes standardization and feature vectorization. Standardization involves converting unstructured text into structured data. This process can unify medical terminology and provide a reliable basis for medical record classification.

[0041] Feature vectorization involves converting medical record text into word vector representations, essentially converting unstructured data like medical record content into numerical vectors. For example, text can be converted into word vectors (N-dimensional vectors), such as using the BERT-base model to generate 768-dimensional word vectors. Embedding models can be used to convert high-dimensional, discrete data (such as text, images, and audio) into low-dimensional, continuous vector representations for machine learning.

[0042] In an application scenario, feature vectorization can be achieved using models such as Word2Vec and BERT. Word2Vec models are trained by predicting context words (Skip-Gram) or center words based on context (CBOW). Models such as BERT learn context-dependent embeddings using a masked language model.

[0043] In step S103, natural language processing (NLP) is performed on the standardized preprocessed medical record data to extract features from the medical record text. Natural language processing (NLP) technology is used to extract key features from the medical record text. These features can effectively represent the semantic information of the medical record and provide high-quality input for subsequent classification tasks. Using NLP technology, characteristic information from the medical record is extracted. Using these features, the model can automatically learn patterns in the medical record data, thereby performing high-precision classification.

[0044] At step S104, the medical record text features are input into a deep learning classification model to classify the medical record data. The deep learning model parameters are optimized through training on a large amount of labeled medical record data. This process uses an automated training strategy, and the model parameters are continuously optimized based on new medical record data to maintain a high level of accuracy.

[0045] In some embodiments, the deep learning classification model adopts a Transformer architecture based on an attention mechanism to capture text semantic associations.

[0046] The above-mentioned deep learning classification model adopts an adaptive training mechanism, which includes incremental learning and active learning to achieve automatic updating of model parameters.

[0047] In some embodiments, incremental learning includes: in response to the number of new medical record data input being greater than a set value, updating the model parameters using a linear gradient descent strategy. The calculation formula of the linear gradient descent strategy is:

[0048] η t =η0 / (1+γt)

[0049] Among them, η t is the current learning rate, η0 is the initial learning rate, γ is the decay coefficient, and t is the number of training rounds.

[0050] Active learning includes selecting high-value unlabeled samples for training through uncertainty sampling.

[0051] The trained deep learning model is applied to the automatic classification of new medical records. Based on the classification results, the system can automatically assign medical record labels and adjust the classification strategy based on user feedback.

[0052] Furthermore, the above deep learning model is optimized using a feedback optimization mechanism. The feedback optimization mechanism includes: correcting the classification results based on user feedback data and generating corrected data; and adding the corrected data to the training set for incremental learning.

[0053] The optimization formula of the loss function in the feedback optimization mechanism is:

[0054]

[0055] Where, is the new loss function, λ is the forgetting factor, and its value range is [0,1]. is the loss of the original training data, is the loss based on user correction data.

[0056] Deployment on a cloud platform ensures real-time processing and analysis of large-scale data. Users can view classification results and provide feedback through the interface, which is then used to further optimize the model. The cloud platform deployment architecture can utilize the Spark or Flink framework to process massive amounts of data, enabling parallel model training and classification inference. New medical record data is received in real time through the Kafka message queue, triggering incremental model updates.

[0057] Taking the case data classification in the case management system of a certain tertiary hospital as an example, the solution of the present invention will be described in detail.

[0058] Deploy cloud platform services and configure GPU computing resources. After going live, 2000 - 3000 new medical records need to be processed every day. Before feature extraction, it is necessary to preprocess the original medical record text to ensure the quality and consistency of the data. In this embodiment, it includes processes such as word segmentation, stop word removal, normalization, and word vector representation.

[0059] (1) Word Segmentation (Tokenization): Split continuous text into independent words or sub - word units. For example: Original text: "The patient's main complaint is persistent headache for three days". The word segmentation result obtained through the above word segmentation process is: ["patient", "main complaint", "persistent", "headache", "three days"]

[0060] (2) Stop Word Removal: Filter out common words with no practical meaning, such as "of", "is", "in", etc., but keep medical key stop words, such as "negative", "positive".

[0061] (3) Normalization:

[0062] Spelling correction: Correct spelling mistakes in medical records, such as "headache" (misspelled as "head pain") → "headache".

[0063] Medical term normalization: Unify synonyms into standard terms, such as "myocardial infarction" (synonym "MI") → "myocardial infarction".

[0064] Number and unit normalization: For example, "3 days" is unified as "three days".

[0065] (4) Word Embedding:

[0066] Convert text into numerical vectors for model processing. Common methods include static word vectors (such as Word2Vec, GloVe), which map each word to a vector of a fixed dimension (such as 300 - dimensional) through a pre - trained model. For example: "headache" → [0.25, - 0.71,..., 0.42].

[0067] Dynamic word vectors (such as BERT, BioClinicalBERT): Generate word vectors based on context, and the vectors of the same word are different in different contexts. It is more suitable for handling ambiguity in medical texts (such as the different meanings of "positive" in test results and symptom descriptions).

[0068] Next, context feature extraction is performed. In this embodiment, the deep learning classification model for feature extraction adopts a Transformer architecture based on an attention mechanism. The contextual semantic information of the text can be captured by a deep learning model. For example, the text is encoded from both the front and back directions through a bidirectional long short-term memory network to capture long-distance dependencies. At the same time, an attention mechanism is set to assign weights to different words, highlighting key information (such as disease names, symptom descriptions), and the text representation is obtained by weighted summation based on the attention weights. The deep Transformer architecture (such as BERT) in this embodiment globally models text relationships through a self-attention mechanism (Self-Attention). The context-related representation of each word is output, and the [CLS] label is extracted as the overall text feature.

[0069] Furthermore, in view of the professionalism of medical record texts, medical field features can be added. In an application scenario, entity recognition content can be added to identify and label medical entities (such as diseases, drugs, and examination items). For example: "ECG shows ST segment elevation" → [Examination: ECG, Result: ST segment elevation]. Relationships between entities (such as "drug-indication") can also be extracted. For example: "Taking aspirin to relieve headaches" → (aspirin, treatment, headache). Integrate entity relationships in medical knowledge graphs (such as UMLS) into feature representation. For example: Use the TransE model to associate "myocardial infarction" with nodes such as "chest pain" and "CK-MB elevation".

[0070] Take the medical record text "65-year-old male patient complained of chest pain for 2 hours, ECG showed ST-segment elevation, and elevated troponin" as an example: The above process is used to segment and retain key terms, and BioBERT generates context-sensitive vectors. BiLSTM is used to capture the temporal relationship of "chest pain → ST-segment elevation → elevated troponin." The attention mechanism assigns high weights to "ST-segment elevation" and "elevated troponin." NER identifies the key entities: [symptom: chest pain, examination: ECG, indicator: troponin]. These features are input into a classifier, which outputs the "acute myocardial infarction" category.

[0071] Through the above steps, NLP technology can extract highly discriminative and semantically rich features from medical record texts, significantly improving classification accuracy.

[0072] The aforementioned deep learning classification model can be initially trained using pre-imported historical medical record data. For example, training can be performed using 500,000 pre-imported historical medical record data. Furthermore, within the feedback optimization mechanism of the automatic medical record classification system, incremental learning dynamically adjusts model parameters to incorporate user feedback data, ensuring that the model can continuously adapt to new classification requirements without losing existing knowledge.

[0073] Specifically, medical staff can correct the classification results through the web interface. Using the corrected classification results, the system automatically performs incremental learning and updates the model every week.

[0074] For example, during the model classification process, the system receives feedback from a medical professional: the user corrects a case misclassified as 'pneumonia' to 'tuberculosis'. The corrected sample is then added to the feedback dataset. M old samples are randomly drawn from the memory bank. The loss is calculated by combining the old and new samples. This calculation can be performed using the loss function formula in the feedback optimization mechanism described above. The model is then updated using the Adam optimizer based on the calculated results.

[0075] In one application scenario, incremental training is initiated when 1,000 new annotated medical records are added. It should be noted that the aforementioned number of new medical records is merely illustrative and non-limiting. Those skilled in the art may also set other numbers based on their needs. This is not a limitation of the present invention. When the number of new cases reaches a certain value, incremental training is initiated to optimize and update the model parameters.

[0076] Furthermore, the above method is highly scalable and can be applied to different types of medical record data, including text and imaging, further improving classification accuracy through multimodal fusion. For example, it can expand multimodal medical record classification by fusing imaging data with text data.

[0077] Specifically, image data can be processed, for example, using ResNet50 to extract visual features from CT / MRI images. Text features can then be fused with image features using an attention mechanism. This fusion process can be achieved, for example, through weighted summation. The fused features are then input into a deep learning classification model for classification.

[0078] Through the above-mentioned solution of the present invention, natural language processing technology is used to pre-process and extract features from medical record texts, ensuring that the deep learning model can effectively extract useful information from complex medical terms and medical record descriptions. This technical solution can automatically optimize the classification model based on new medical record data. The system's adaptive ability can self-adjust as the amount of data increases or the type of medical record changes without manual intervention, maintaining a high classification accuracy rate. The present invention can perform real-time processing on large-scale medical record data, and through cloud platform deployment, the entire system can process large amounts of data, maintain an efficient classification speed, and provide accurate classification results for each medical record. A user feedback mechanism is introduced, and users can correct the automatic classification results during use. The system uses these feedbacks as input to further optimize the classification model, so that the model is gradually improved in practical applications.

[0079] This invention is applicable to various medical institutions, such as hospitals and clinics, and can help medical personnel manage medical record data more efficiently, thereby improving the quality of medical services. Compared with existing technologies, the solution of this invention can automatically identify and classify medical record content with high accuracy, intelligence, and adaptability. It can also process massive amounts of data, meeting the needs of hospital information development.

[0080] Figure 2 2 is a block diagram showing the composition of an AI-based automatic medical record classification system according to an embodiment of the present invention.

[0081] The present invention also provides an AI-based automatic medical record classification system. Figure 2 As shown, the system includes a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, an AI-based automatic medical record classification method as described above is implemented.

[0082] The system further includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and thus will not be described in detail here.

[0083] In the present invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, the computer-readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc., or any other medium that can be used to store the required information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible or connectable to a device. Any application or module described in the present invention can be implemented using computer-readable / executable instructions that can be stored or otherwise retained by such a computer-readable medium.

[0084] In the description of this specification, "multiple" and "several" mean at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0085] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. An AI-based automatic medical record classification method, characterized in that: include: Obtain medical record data; Clean and standardize medical record data; Perform natural language processing on the standardized pre-processed medical record data to extract medical record text features; The medical record text features are input into the deep learning classification model to classify the medical record data, where the deep learning model is optimized using a feedback optimization mechanism.

2. The AI-based automatic medical record classification method according to claim 1, characterized in that: Data cleaning includes removing noise data and eliminating irrelevant information, and standardization preprocessing includes standardization processing and feature vectorization. The standardization processing includes converting unstructured text into structured data, and the feature vectorization includes converting the medical record text content into word vector representation.

3. The AI-based automatic medical record classification method according to claim 1, characterized in that: The deep learning classification model adopts the Transformer architecture based on the attention mechanism to capture the semantic association of text.

4. The AI-based automatic medical record classification method according to claim 3, characterized in that: The deep learning classification model adopts an adaptive training mechanism, which includes incremental learning and active learning to achieve automatic updating of model parameters.

5. The AI-based automatic medical record classification method according to claim 4 is characterized in that: Incremental learning includes: In response to the number of new medical record data input being greater than the set value, the model parameters are updated using a linear gradient descent strategy.

6. The AI-based automatic medical record classification method according to claim 5, characterized in that: The calculation formula of the linear gradient descent strategy is: or t =η0 / (1+γt) Among them, η t is the current learning rate, η0 is the initial learning rate, γ is the decay coefficient, and t is the number of training rounds.

7. The AI-based automatic medical record classification method according to claim 4, characterized in that: Active learning includes: High-value unlabeled samples are selected for training through uncertainty sampling.

8. The AI-based automatic medical record classification method according to claim 1, characterized in that: The feedback optimization mechanism includes: Correct the classification results based on user feedback data and generate corrected data; The correction data is added to the training set for incremental learning.

9. The AI-based automatic medical record classification method according to claim 8, characterized in that: The optimization formula of the loss function in the feedback optimization mechanism is: Where, is the new loss function, λ is the forgetting factor, and its value range is [0,1]. is the loss of the original training data, is the loss based on user correction data.

10. An AI-based automatic medical record classification system, characterized by: include: processor; A memory storing computer program instructions for automatic classification of medical records, wherein when the computer program instructions are executed by the processor, the AI-based automatic classification method for medical records according to any one of claims 1 to 9 is implemented.

Images